v3.9.7 (2023-01-10)
-------------------

* Allow cluster database servers to start even when there are existing databases
  that would violate the settings `--cluster.min-replication-factor` or
  `--cluster.max-replication-factor`.
  This allows upgrading from older versions in which the replication factor
  validation for databases was not yet present.

* Make the cache_oblivious option of jemalloc configurable from the environment.
  This helps to save 4096 bytes of RAM for every allocation which is at least
  16384 bytes large. This is particularly beneficial for the RocksDB buffer
  cache.

* Use intermediate commits in old shard synchronization protocol. This avoids
  overly large RocksDB transactions when syncing large shards, which is a remedy
  for OOM kills during restarts.

* Allow parallel access to the shards of smart edge collections in AQL via
  parallel GatherNodes.

* Added a configuration option (for the agency):
    --agency.supervision-failed-leader-adds-follower
  with a default of `true` (behavior as before). If set to `false`, a
  `FailedLeader` job does not automatically configure a new shard follower,
  thereby preventing unnecessary network traffic, CPU and IO load for the case
  that the server comes back quickly. If the server is permanently failed, an
  `AddFollower` job will be created anyway eventually.

* Disallow creating new databases with a `replicationFactor` value set to a
  value lower than `--cluster.min-replication-factor` or higher than
  `--cluster.max-replication-factor`. Previously the `replicationFactor`
  settings for new databases were not bounds-checked, only for new collections.

* Fixed Github issue #16451: In certain situations, a LIMIT inside a subquery
  could erroneously reduce the number of results of the containing (sub)query.

* Fixed BTS-1153 GeoJson columns added to cache for ArangoSearch cached columns.


v3.9.6 (2022-12-06)
-------------------

* Don't log Boost ASIO warnings such as `asio IO error: 'stream truncated'`
  when a peer closes an SSL/TLS connection without performing a proper
  connection shutdown.

* Added experimental per-operation option `refillIndexCaches` to write 
  operations, namely:

  - AQL INSERT/UPDATE/REPLACE/REMOVE modification operations
  - single-document insert, update, replace and remove operations
  - multi-document insert, update, replace and remove operations

  If the option is set to `true` every currently running transaction will 
  keep track of which in-memory edge index cache entries were invalidated 
  by the transaction, and will try to (re-)fill them later.
  Currently only edge indexes are supported. ArangoDB 3.11 will add support
  for velocypack-based indexes (persistent, hash, sklipist index).

  Example usages:
  - `db.<collection>.insert({ _from: ..., _to: ..., ... }, 
  { refillIndexCaches: true });`
  - `db.<collection>.update(key, { _from: ..., _to: ..., ... }, 
  { refillIndexCaches: true });`
  - `db.<collection>.replace(key, { _from: ..., _to: ..., ... }, 
  { refillIndexCaches: true });`
  - `db.<collection>.remove(key, { refillIndexCaches: true });`
  - `INSERT { ... } INTO <collection> OPTIONS { refillIndexCaches: true }`
  - `UPDATE { ... } WITH { ... } INTO <collection> 
  OPTIONS { refillIndexCaches: true }`
  - `REPLACE { ... } WITH { ... } INTO <collection> 
  OPTIONS { refillIndexCaches: true }`
  - `REMOVE { ... } IN <collection> OPTIONS { refillIndexCaches: true }`

  The refilling of the in-memory caches for indexes is performed by a 
  background thread, so that the foreground write operation shouldn't be 
  slowed down a lot. The background thread may however cause additional 
  I/O for looking up the data in RocksDB and for repopulating the caches.

  The background refilling is done in a best-effort way and is not guaranteed 
  to always succeed, e.g. if there is no memory available for the cache 
  subsystem, or when an in-memory cache table is currently in a migration phase
  (grow/shrink operation).

  There is a new experimental startup option 
  `--rocksdb.auto-refill-index-caches-on-modify` for DB-Servers and single 
  servers, which currently defaults to `false`. If it is set to `true`, the 
  cache refilling will be turned on automatically for all insert/update/
  replace/remove operations, so that it doesn't need to be specified on the 
  per-operation/per-query level.

  The experimental option `--rocksdb.auto-refill-index-caches-queue-capacity`
  can be used to limit the number of index cache entries that the background 
  thread will queue. This is a safeguard to keep the memory usage at bay in 
  case the background thread is slower than concurrent threads that perform 
  ingestions.

  There are also new experimental startup options to control whether or not 
  the in-memory caches should automatically be seeded upon server restart.
  The option `--rocksdb.auto-fill-index-caches-on-startup` for DB-Servers and 
  single servers enables this functionality. It currently defaults to `false`.
  If it is set to `true`, the in-memory caches of all eligible indexes will be
  automatically pre-seeded after the server startup. Note that this may cause 
  additional CPU and I/O load.
  The option `--rocksdb.max-concurrent-index-fill-tasks` is available to limit
  the impact of the automatic index filling at startup. It controls how many 
  full index filling operations can execute concurrently. The lower this number
  is, the lower the impact of cache filling, but the longer it will take. 
  The default value for this option depends on the number of available cores, 
  and is at least `1`. A value of `0` cannot be used. 
  This option is only relevant if `--rocksdb.auto-fill-index-caches-on-startup` 
  is set to `true`.

  The PR also adds the following metrics:
  - `rocksdb_cache_auto_refill_loaded_total`: Total number of queued items for 
    in-memory index caches refilling. It will always report a value of zero on 
    coordinators.
  - `rocksdb_cache_auto_refill_dropped_total`: Total number of dropped items 
    for in-memory index caches refilling (because number of queued items would 
    exceed the value of `--rocksdb.auto-refill-index-caches-queue-capacity`). 
    It will always report a value of zero on coordinators.
  - `rocksdb_cache_full_index_refills_total`: Total number of in-memory index
    caches refill operations for entire indexes. The counter gets increased for
    every index automatically loaded (because startup option 
    `--rocksdb.auto-fill-index-caches-on-startup` is set to `true`) or when
    full indexes are loaded into memory manually. 
    In cluster deployments the counter will be increased once per eligible 
    index per shard. It will always report a value of zero on coordinators.

* Added a feature to the ResignLeadership job. By default, it will now
  undo the leader changes automatically after the server is restarted,
  unless the option `undoMoves` is set to `false`. This will help to
  make rolling upgrades and restarts less troublesome, since the shard
  leaderships will not get unbalanced.

* Add missing metrics for user traffic: Histograms:
    `arangodb_client_user_connection_statistics_bytes_received`
    `arangodb_client_user_connection_statistics_bytes_sent`
  These numbers were so far only published via the statistics API.
  This is needed for ArangoGraph traffic accounting.

* Added agency options
    --agency.supervision-delay-add-follower
  and
    --agency.supervision-delay-failed-follower
  to delay supervision actions for a configurable amount of seconds. This is
  desirable in case a DBServer fails and comes back quickly, because it gives
  the cluster a chance to get in sync and fully resilient without deploying
  additional shard replicas and thus without causing any data imbalance.

* Fix HTTP/VST traffic accounting in internal statistics / metrics.

* Delay a MoveShard operation for leader change, until the old leader has
  actually assumed its leadership and until the new leader is actually in sync.
  This fixes a bug which could block a shard under certain circumstances. This
  fixes BTS-1110.

* Updated arangosync to v2.13.0.

* Fixed issue #17291: Server crash on error in the PRUNE expression.
  Traversal PRUNE expressions containing JavaScript user-defined functions
  (UDFs) are now properly rejected in single server and cluster mode.
  PRUNE expressions that use UDFs require a V8 context for execution, which is
  not available on DB-servers in a cluster, and also isn't necessarily available
  for regular queries on single servers (a V8 context is only available if a
  query was executed inside Foxx or from inside a JS transaction, but not
  otherwise).

* Fix setting query memory limit to 0 for certain queries if a global memory
  limit is set, but overriding the memory limit is allowed.

* Fixed BTS-1125 Fixes handling of primarySortCache and primaryKeyCache
  properties in ArangoSearch views.

* Do not query vertex data in K_PATHS queries if vertex data is not needed.


v3.9.5 (2022-11-10)
-------------------

* Enable "collect-in-cluster" optimizer rule for SmartGraph edge collections.

* Show number of HTTP requests in cluster query profiles.

* Repair "load indexes into memory" function in the web UI.

* Solve a case of excessive memory consumption in certain AQL queries with IN
  filters with very long lists. Free sub-iterators as soon as they are
  exhausted.

* BTS-1070: Fixed query explain not dealing with an aggregate function without
  arguments and the WINDOW node not being defined as an Ast node type name.

* Fixed error handling of coordinator commit of streaming transactions.
  This bug could mean that a streaming transaction is reported to be
  successfully committed when in reality it isn't.

* Updating properties of a satellite collection breaks replicationFactor.

* Log the documents counts on leader and follower shards at the end of each
  successful shard synchronization.

* Updated OpenSSL to 1.1.1s.

* Added startup option `--query.log-failed` to optionally log all failed AQL
  queries to the server log. The option is turned off by default.

* Added startup option `--query.log-memory-usage-threshold` to optionally log
  all AQL queries that have a peak memory usage larger than the configured
  value. The default value is 4GB.

* Added startup option `--query.max-artifact-log-length` to control the maximum
  length of logged query strings and bind parameter values.
  This allows truncating overly long query strings and bind parameter values to
  a reasonable length. Previously the cutoff length was hard-coded.

* Improve cardinality estimate for AQL EnumerateCollectionNode in case a `SORT
  RAND() LIMIT 1` is used. Here, the estimated number of items is at most 1.

* Improved shard distribution during collection creation.

* SEARCH-408 Added "cached" columns feature for the ArangoSearch.


v3.9.4 (2022-10-18)
-------------------

* Fixed BTS-852 (user's saved queries used to disappear after updating user
  profile).

* Changed the encoding of revision ids returned by the following REST APIs:
  - GET /_api/collection/<collection-name>/revision: the revision id was
    previously returned as numeric value, and now it will be returned as
    a string value with either numeric encoding or HLC-encoding inside.
  - GET /_api/collection/<collection-name>/checksum: the revision id in the
    "revision" attribute was previously encoded as a numeric value in single
    server, and as a string in cluster. This is now unified so that the
    "revision" attribute always contains a string value with either numeric
    encoding or HLC-encoding inside.

* Fixed handling of empty URL parameters in HTTP request handling.

* Fixed diffing of completely non-overlapping revision trees, which could lead
  to out-of-bounds reads at the right end of the first (smaller) tree.

* Fixed aborting the server process if an exception was thrown in C++ code that
  was invoked from the llhttp C code dispatcher. That dispatcher code couldn't
  handle C++ exceptions properly.

* Fixed BTS-1073: Fix encoding and decoding of revision ids in replication
  incremental sync protocol. Previously, the encoding of revision ids could be
  ambiguous under some circumstances, which could prevent shards from getting
  into sync.

* ES-1312: fix handling of reaching the WAL archive capacity limit.

* Log better diagnosis information in case multiple servers in a cluster are
  configured to use the same endpoint.

* BTS-908: Fixed WebUI GraphViewer not being able to create a new edge relation
  between two nodes in cases where only one edge definition has been defined
  inside the graph definition.

* Fixed BTS-850: Fixed the removal of already deleted orphan collections out of
  a graph definition. The removal of an already deleted orphan collection out of
  a graph definition failed and has been rejected in case the collection got
  dropped already.

* Fix an issue with replication of arangosearch view change entries in single
  server replication and active failover. Previously, when changing the
  properties of existing views, the changes were not properly picked up by
  followers in these setups. Cluster setups were not affected.

* Improve upload and download speed of hotbackup by changing the way we use
  rclone. Empty hash files are now uploaded or downloaded by pattern, and all
  other files are done in batches without remote directory listing, which allows
  rclone to parallelize and avoid a lot of unnecessary network traffic. The
  format of hotbackups does not change at all.

* Updated arangosync to v2.12.0.

* Fixed BTS-905 regarding read-only mode for expired license.

* Fixed issue BTS-1018: Improve logging of binary velocypack request data.

* Defer intermediate commits in the middle of a multi-document (array)
  operation. This is to ensure that the RocksDB key locks for all participating
  document keys are still held while the operations are replicating via the
  synchronous replication.

* Fixed issue BTS-1023:
  Added Linux-specific startup option `--use-splice-syscall` to control whether
  the Linux-specific splice() syscall should be used for copying file contents.
  While the syscall is generally available since Linux 2.6.x, it is also
  required that the underlying filesystem supports the splice operation. This is
  not true for some encrypted filesystems, on which splice() calls thus fail.
  By setting the startup option `--use-splice-syscall` to `false`, a less
  efficient, but more portable user-space file copying method will be used
  instead, which should work on all filesystems.
  The startup option is not available on other operating systems than Linux.

* Fixed BTS-1017: Fixed a graph search issue, where subqueries lead to incorrect
  results when they have been pushed down fully onto a DBServer when they are in
  a Hybrid Disjoint SmartGraph context and SatelliteCollections were part of it.

* Updated ArangoDB Starter to 0.15.5.

* Fixed SEARCH-374: Fixed ArangoSearch seek iteration over documents in sparse
  bitset.

* Added startup option `--arangosearch.skip-recovery` to skip the recovery of
  arangosearch view links or inverted indexes.
  The startup option can be specified multiple times and is expected to either
  contain the string `all` (will skip the recovery for all view links and
  inverted indexes) or a collection name + link id/name pair (e.g.
  `testCollection/123456`, where `123456` is a link/index id or an index name).
  This new startup option is an emergency means to speed up lengthy recovery
  procedures when there is a large WAL backlog to replay. The normal recovery
  will still take place even with the option set, but recovery data for
  links/indexes can be skipped. This can improve the recovery speed and reduce
  memory usage during the recovery process.
  All links or inverted indexes that are marked as to-be-skipped via the option,
  but for which there is recovery data, will be marked as "out of sync" at the
  end of the recovery.
  The recovery procedure will also print a list of links/indexes which it has
  marked as out-of-sync.
  Additionally, if committing data for a link/index fails for whatever reason,
  the link/index is also marked as being out-of-sync.

  If an out-of-sync link or index can be used in queries depends on another new
  startup option `--arangosearch.fail-queries-on-out-of-sync`. It defaults to
  `false`, meaning that out-of-sync links/indexes can still be queried. If the
  option is set to `true`, queries on such links/indexes will fail with error
  "collection/view is out of sync" (error code 1481).

  Links/indexes that are marked out-of-sync will keep the out-of-sync flag until
  they are dropped. To get rid of an out-of-sync link/index, it is recommended
  to manually drop and recreate it. As recreating a link/index may cause high
  load, this is not done automatically but requires explicit user opt-in.

  The number of out-of-sync links/indexes is also observable via a new metric
  `arangodb_search_num_out_of_sync_links`.

* Implement prefetch for revision trees, in case a batch is created with
  a distinguished collection as for `SynchronizeShard`. This ensures that
  the revision tree for the batch will be available when needed, even though
  the revision tree for the collection might already have advanced beyond
  the sequence number of the snapshot in the batch. This ensures that
  shards can get in sync more reliably and more quickly.

* Fixed BTS-926: UI showing the "create index" form to non-admin users.

* Fixed a rare occuring issue where paths inside a DisjointSmart traversal
  containing only satellite relevant nodes were not returned properly
  (ES-1265).


v3.9.3 (2022-09-02)
-------------------
      
* Added startup option `--rocksdb.periodic-compaction-ttl`.
  This option controls the TTL (in seconds) for periodic compaction of .sst
  files in RocksDB, based on the .sst file age. The default value from RocksDB
  is ~30 days. To avoid periodic auto-compaction, the option can be set to 0.

* Fixed SEARCH-366: Fixed extracting sub-attributes using projections for the
  `_id` attribute (e.g. `RETURN doc.sub._id`).

* Do not drop follower shard after too many failed shard synchronization
  attempts.

* Make AQL query explainer display "index scan + document lookup" for queries
  that use an index for filtering, but need to fetch more document data from the
  storage engine. Previously, those were labeled only as "index scan",
  obfuscating the fact that there are document lookups needed.
  Covering-index queries are still labeled "index scan, index only" as before.

* Disable optimization rule to avoid crash (BTS-951).

* Fix SEARCH-350: Crash during consolidation.

* Fix comparison of JSON schemas on DB servers after there was a schema change
  via a coordinator: the schema comparison previously did not take into account
  that some ArangoDB versions store an internal `{"type":"json"}` attribute in
  the schema, and some don't. Thus two identical schemas could compare
  differently.
  The correct schema version was always applied and used, and validation of
  documents against the schema was also not affected. However, because two
  schemas could compare unequal, this could have caused unnecessary repeated
  work for background maintenance threads.

* Add progress reporting to RocksDB WAL recovery, in case there are many WAL
  files to recover.

* Web UI: Now correctly handles the server error response when an error occurred
  during the modification of a document or an edge (BTS-934).

* Make graph search case-insensitive (fixes BTS-882).

* Make throttle for RocksDB take into account the complete compaction debt on
  all levels, not just level0. Previously it only looked at level0, which is not
  enough if there aren't any other levels in the column family.

* Improve error handling for passing wrong transaction ids / cursor ids / pregel
  job ids to request forwarding. Also prevent the error "transaction id not
  found" in cases when request forwarding was tried to a coordinator that was
  recently restarted.

* Fixed a potential deadlock in RocksDB compaction.
  For details see https://github.com/facebook/rocksdb/pull/10355.

* Updated warning messages raised for non accepted query OPTIONS, 
  distinguishing between when the OPTIONS attribute is correct, but the value 
  is in incorrect format, and when the OPTIONS attribute itself is incorrect.

* BTS-907: Fixed some rare SortNode related optimizer issues, when at least two
  or more SortNodes appeared in the AQL execution plan.

* Fixed BTS-902 (clicking on the search icon in the analyzers filter input used
  to take the user to the collections view).

* Fixed BTS-852 (user's saved queries used to disappear after updating user
  profile).

* BTS-913: check for proper timezone setup of the system on startup.
  This will then log errors that else would only occur in AQL-Functions at
  runtime.

* Fixed BTS-851: "Could not fetch the applier state of: undefined".

* Since ArangoDB 3.8 there was a loophole for creating duplicate keys in the
  same collection. The requirements were:
  - cluster deployment
  - needs at least two collections (source and target), and the target
    collection must have more than one shard and must use a custom shard key.
  - inserting documents into the target collection must have happened via an AQL
    query like `FOR doc IN source INSERT doc INTO target`.
  In this particular combination, the document keys (`_key` attribute) from the
  source collection were used as-is for insertion into the target collection.
  However, as the target collection is not sharded by `_key` and uses a custom
  shard key, it is actually not allowed to specify user-defined values for
  `_key`. That check was missing since 3.8 in this particular combination and
  has now been added back. AQL queries attempting to insert documents into a
  collection like this will now fail with the error "must not specifiy _key for
  this collection", as they used to do before 3.8.

* Fixed an invalid attribute access in AQL query optimization.
  Without the fix, a query such as

      LET data = {
        "a": [
          ...
        ],
      }
      FOR d IN data["a"]
        RETURN d

  could fail with error "invalid operand to FOR loop, expecting Array".

* Updated OpenSSL to 1.1.1q and OpenLDAP to 2.6.3.

* Updated arangosync to v2.11.0.

* Changed default macOS CI compiler to LLVM clang-13.

* Fixed issue 16337: arangoimport with `--headers-file` and `--merge-attributes`
  merges column names instead of row values on the first line of a CSV file.

  Additionally, floating-point numbers are now merged using their standard
  string representation instead of with a fixed precision of 6 decimal places.

* Optionally enable range deletions in the WAL for truncate operations in the
  cluster, too. This can speed up truncate operations for large
  collections/shards. Range deletions are turned off by default to ensure
  compatibility with all versions of arangosync. They can be turned on by
  using the startup option `--rocksdb.use-range-delete-in-wal` and setting it to
  `true`. The default value of that option will change from `false` to `true` in
  ArangoDB 3.10.

* Fixed potential resource leak when reusing `rocksdb::PinnableSlice` instances
  without resetting them first. This issue can lead to the block cache filling
  up with data that will not be evicted from the block cache anymore.


v3.9.2 (2022-06-07)
-------------------

* Enterprise only: Restricted behavior of Hybrid Disjoint Smart Graphs. Within a 
  single traversal or path query we now restrict that you can only switch
  between Smart and Satellite sharding once, all queries where more than one
  switch is (in theory) possible will be rejected. e.g:
  ```
    FOR v IN 2 OUTBOUND @start smartToSatEdges, satToSmartEdges
  ```
  will be rejected (we can go smart -> sat -> smart, so two switches)
  ```
  FOR v1 IN 1 OUTBOUND @start smartToSatEdges
    FOR v2 IN 1 OUTBOUND v1 satToSmartEdges
  ```
  will still be allowed, as each statement only switches once.
  We have decided to take these restrictions as especially for ShortestPath
  queries the results are not well-defined. If you have a use case where this
  restriction hits you, please contact us.

* Fixed issue BTS-875.

* Updated arangosync to v2.10.0.

* Make all requests which are needed for shard resync at least medium priority
  to improve getting-in-sync under load.

* Fix behavior when accessing a view instead of a collection by name in a REST
  document operation. Now return a proper error.

* Fix documentation of collection's `cacheEnabled` property default.

* Added startup option `--cluster.shard-synchronization-attempt-timeout` to
  limit the amount of time to spend in shard synchronization attempts. The
  default timeout value is 20 minutes.
  Running into the timeout will not lead to a synchronization failure, but will
  continue the synchronization shortly after. Setting a timeout can help to
  split the synchronization of large shards into smaller chunks and release
  snapshots and archived WAL files on the leader earlier.
  This change also introduces a new metric `arangodb_sync_timeouts_total` that
  counts the number of timed-out shard synchronization attempts.

* Make followers respond to synchronous replication requests with less data.
  Specifically, followers will not build detailed results with _id, _key and
  _rev for the inserted/modified/removed documents, which would be ignored by
  the leader anyway.

* Very verbose warning from failing to parse GEO JSON in search. Has lead to
  billions of log lines on deployed services.

* Fixed Github issue #16279: assertion failure/crash in AQL query optimizer when
  permuting adjacent FOR loops that depended on each other.

* Fixed a potential hang on shutdown, when there were still document operations
  queued.

* No good reason to fatal error in agency state, when local database entries
  lack local timestamp (legacy). In that situation, we will record epoch begin
  as local time.

* Fixed BTS-860. Changed ArangoSearch index recovery procedure to remove
  necessity to always fully recreate index if IndexCreation marker encountered.

* Removed separate FlushThread (for views syncing) and merged it with the
  RocksDBBackgroundThread.

* Fix some issues with WAL recovery for views. Previously it was possible that
  changes to a view/link were already recovered and persisted, but that the
  lower bound WAL tick was not moved forward. This could lead to already fully
  recovered views/links being recovered again on the next restart.

* Put hotbackup requests on the HIGH priority queue to make hotbackups work
  under high load (BTS-865).

* Allow starting with option value `--cache.size 0`, to turn off in-memory
  caches for indexes entirely (for performance testing or limiting memory
  usage).

* Updated OpenSSL to 1.1.1o and OpenLDAP to 2.6.2.

* Added option `--enable-revision-trees` to arangorestore, which will add the
  attributes `syncByRevision` and `usesRevisionsAsDocumentIds` to the
  collection structure if they are missing. As a consequence, these collections
  created by arangorestore will be able to use revision trees and a faster
  getting-in-sync procedure after a restart. The option defaults to `true`,
  meaning the attributes will be added if they are missing. If the option is
  set to `false`, the attributes will not be added to the collection structure.
  If the attributes are already present in the dump data, they will not be
  modified by arangorestore irrespective of the setting of this option.

* Fix agency inception when one of the gossip peers responds with HTTP 503 or
  another unexpected error.

* Make sure that newly created TTL indexes do not use index estimates, which
  wouldn't be used for TTL indexes anyway.

* Improve log output for WAL recovery, by providing more information and making
  the wording more clear.

* Fix: Highly unlikely race in cluster maintenance. For every shard only one
  operation (change attribute, change leadership) should be performed at the
  same time. However if two changes are detected in the same hearbeat  it could
  lead to both operations to be executed in parallel. In most cases this is also
  fine, but could lead to races on the same attribute, however the race will be
  sorted out in the next heartbeat interval.

* Fix: for the Windows build, the new Snappy version, which was introduced in
  3.9, generated code that contained BMI2 instructions which where introduced
  with the Intel Haswell architecture. However, our target architecture for 3.9
  is actually Sandy Bridge, which predates Haswell. Running the build on these
  older CPUs thus resulted in illegal instruction exceptions.

* Increase internal transaction lock timeout on followers during cluster write
  operations. Although writes to the same keys on followers should be serialized
  by the key locks held on the leader, it is still possible that the global
  transaction lock striped mutex is a source of contention and that concurrent
  write operations time out while waiting to acquire this global mutex. The lock
  timeout on followers is now significantly increased to make this very
  unlikely.

* Improve validation for variables used in the `KEEP` part of AQL COLLECT
  operations. Previously referring to a variable that was introduced by the
  COLLECT itself from out of the KEEP part triggered an internal error. The case
  is detected properly now and handled with a descriptive error message.

* Added startup option `--rocksdb.transaction-lock-stripes` to configure the
  number of lock stripes to be used by RocksDB transactions. The option defaults
  to the number of available cores, but is bumped to a value of 16 if the number
  of cores is lower.

* Fix deadlocked shard synchronisations when planned shard leader has not yet
  taken over leadership.

* Added an IO heartbeat which checks that the underlying volume is writable with
  reasonable performance. The test is done every 15 seconds and can be
  explicitly switched off. New metrics to give visibility if the test fails:
    - `arangodb_ioheartbeat_delays_total`: total number of delayed io heartbeats
    - `arangodb_ioheartbeat_duration`: histogram of execution times [us]
    - `arangodb_ioheartbeat_failures_total`: total number of failures
  These metrics are only populated, if `--database.io-heartbeat` is set to
  `true` (which is currently the default).

* Fix lock order in Agent::advanceCommitIndex for State's _logLock and Agent's
  _waitForCV.

* BugFix (enterprise-only): (BTS-787) In a hybrid disjoint SmartGraph, having
  more than one relation, if you add a new vertex collection to a Smart -> Smart
  edge relation this vertex collection was rejected with "has to be satellite"
  error.
  Now the collection is created as a SmartVertexCollection as desired.

* Resync follower shard after a follower restart immediately and not lazily.


v3.9.1 (2022-04-04)
-------------------

* Updated ArangoDB Starter to 0.15.4.

* Improve parallelism in arangorestore in case new data format is used.

* Remove error handling fetching license information to improve user
  experience. To display the license informationn in the UI is only
  informational. It disturbs the user experience to know somthing went wrong
  and doesn't provide any important information for the user.

* Added new server option: --icu-language. Used instead of --default-language to
  set pure ICU collator.
  For example, in Sweden language("sv") lowercase letters should precede
  uppercase ones. You can achieve it using following options when server starts:

    `--icu-language sv`

* No longer put document writes from replication into the audit log by default.
  Same with low priority authentication like internal UI requests to .html files
  for the UI. This solves a performance problem for shards getting in sync with
  audit log switched on.

* Updated OpenSSL to 1.1.1n and OpenLDAP to 2.6.1.

* Fixed an assertion failure which could occur when there was an error in the
  HTTP header, so that the message body was not actually read.

* Fixed a crash which could occur when there was an error in the HTTP header
  parsing.

* Bug-Fix: Resolve BTS-673/Issue #15107, a spliced subquery could return too few
  results.

* Speed up initial sync (in case there is already data present) by prefetching
  data from leader.

* Fixed ES-1078: The REST API endpoint for handling `/_api/user/${user}/config`
  did not work properly. The supplied data by sending a PUT request has not been
  stored to the correct location. The Web UI uses this endpoint to store its
  graph properties for storing the visualization properties. As this endpoint
  did not work as expected, the graph visualization properties did not get
  persisted as well. This is now resolved.

* Fix counts and file size sum in hotbackup META files. Do no longer count
  directories.

* Optimize further RocksDB throttle to allow for no change on any given
  calculation cycle.

* Fix UI to only fetch license info as admin user.

* Added `disableIndex` index hint for AQL FOR loops. This index hint disables
  the usage of any index (except geo or full text indexes) and will cause a full
  scan over the collection.
  In some circumstances a full scan can be more efficient than an index scan,
  for example if the index scan produces many matches (close to the number of
  documents in the collection) and the index is not fully covering the query.
  The `disableIndex` hint can be given per FOR loop in the query, e.g.:

      FOR doc IN collection OPTIONS { disableIndex: true }
        RETURN doc.value

  The default value of `disableIndex` is `false`.
  In case a different index hint is provided, `disableIndex: true` takes
  precendence and produces a warning about the ambiguous settings.

* Added `maxProjections` hint for AQL FOR loops. This hint can be used to set
  the maximum number of document attributes that are taken into account for
  using projections.

  For example, in the following query, no projections will be used because the
  number of potential projection attributes (`value1`, value2`, `value3`) is
  higher than the maximum number of projection attributes set via the
  `maxProjections` option:

      FOR doc IN collection OPTIONS { maxProjections: 2 }
        RETURN [ doc.value1, doc.value2, doc.value3 ]

  The default value for `maxProjections` is `5`, which is compatible with the
  previous hard-coded default value.

* Avoid a deadlock in agency, when Inception::gossip() acquired a mutex, then
  could call Agent under the mutex, and Agent could finally could call
  Inception::signalConditionVar(), which would try to acquire the mutex again.

* Avoid multiple parallel SIGHUP requests to be handled at the same time.
  Now collapse multiple incoming SIGHUP requests into a single one, which can be
  executed race-free.

* Upgraded JavaScript "i" module from 0.3.6 to 0.3.7.

* Removed internal JavaScript dependencies "mocha" and "chalk". We recommend
  always bundling your own copy of third-party modules, even ones listed as
  public.

* Reduce memory usage of inner joins if they were performed by the IndexExecutor
  with dynamic index lookup expressions that needed to be recomputed for input
  from the outer loop.

  For example, in the query
  ```
  FOR i IN 1..1000
    FOR doc IN collection
      FILTER doc.indexAttribute == i
      RETURN doc
  ```
  the inner loop will be executed 1000 times. The IndexExecutor in the inner
  loop needed to rebuild the index lookup attribute from the value of `i` 1000
  times as well. The memory for index lookup attributes came from the Ast's
  memory allocator and was not freed until the end of the query. In this query,
  it would mean that up to 1000 lookup values were held in memory. With larger
  inputs even more memory would be used.

  Now the memory for index lookup values is freed when a new lookup value is
  computed, i.e. only a single lookup value is held in memory.
  This drastically reduces peak memory usage for queries that use index lookups
  in inner loops and that get lots of different inputs from outer loops.

* Harden validator for binary VelocyPack against additional types of malicious
  inputs.

* Bugfix: DC-2-DC Disjoint-SmartGraphs and Hybrid-SmartGraphs are now replicated
  to the follower data-center keeping their sharding intact.

* As we are now in constant stall regime, stall onset and warnings are demoted
  to DEBUG.

* Shorten the license grace period to the documented 3 hours.

* Replaced internal JS dependency xmldom with @xmldom/xmldom.

* Fixed BTS-750: Fixed the issue restricted to cluster mode in which queries 
  containing the keywords UPDATE or REPLACE together with the keyword WITH and
  the same key value would result in an error. For example: 
  `UPDATE 'key1' WITH {_key: 'key1'} IN Collection`
  because the same key used to update was provided in the object to update the
  document with.

* Replaced internal JS dependency ansi-html with ansi-html-community.

* In an attempt to make the performance of the RocksDB throttle much more
  consistent and predictable the default compaction slow down trigger is lowered
  to 128kB.

* Bug-Fix: AQL WINDOW statement if applied within a subquery could accidentally
  skip over some subquery results. This did only show up if the subquery fills
  exactly one internal batch before it is completed, so it is rather unlikely.

* Fix null pointer access when using WINDOW operation with a COUNT/LENGTH
  aggregate function without any arguments.

* Reintroduce shard synchronization cancellation check that was disabled before.

* Fixed BTS-621 Fixed rare case of segfault in cluster during database recovery
  if DBServer is in upgrade mode in the same time.

* Fixed PRESUPP-445: Foxx queues: Some Jobs are never run in case of multiple
  Coordinators.

* Fixed a race detected with chaos tests, where a db server could have
  momentarily lost leadership, just when it was about to drop a follower to a
  shard.

* Fixed issue #15501: Regression when using "exclusive" query option?
  This fixes a regression in AQL query execution when exclusive locks are used
  for a query and the query also uses the DOCUMENT() AQL function. In this case,
  when there were more concurrent requests to the underlying collection than
  available scheduler threads for the low priority queue, the query start
  successfully acquired the exclusive lock, but could not get its follow-up
  requests through and starve while holding the exclusive lock.

* Disable optimizer rule "optimize-cluster-single-document-operations" when a
  collection is accessed in exclusive mode, because the optimized query would
  use a slightly different mode of locking then.


v3.9.0 (2022-02-07)
-------------------

* Convert v3.9.0-rc.1 into v3.9.0 (GA).


v3.9.0-rc.1 (2022-02-03)
------------------------

* Fix potential access to dangling reference in cancellation of shard
  synchronization.

* Fixed BTS-740 (no released version infected) fixed Smart<->Sat SmartEdgeCollections
  determining the shard in SingleRemoteModification Nodes was incorrect. E.g. this could
  be triggered, by viewing the details of an edge in the UI. Only alpha/beta of 3.9.0
  contained this bug

* Fixed BTS-729 (no released version infected): Some conditions in a Hybrid
  Smart Graph could led to wrong shard location calculation and therefore to
  wrong graph query results. Only alpha/beta of 3.9.0 contained this bug.

* Fixed minDepth handling of weighted Traversals. When using a minDepth of 3, also paths of length
  2 have been returned, on all locally executed variants (SingleServer, OneShard, DisjointSmart).

* Fixed BTS-728 (no released version infected) fixed: for DisjointSmartGraphs, that include
  a satellite vertex collection, valid disjoint path were not always followed, if one of the
  satellites has a connection to two (or more) vertices that have different shardValues that
  by chance are routed to the same shard.

* Fix creation of satellite graphs with a `numberOfShards` value != 1.

* Fixed BTS-712: Collation analyzer now always produces valid UTF-8 sequence.

* Updated Enterprise license behavior: now there will be a 48 hour period for a
  new deployment and upgrade decision to provide the license. After that period,
  the read-only mode will be enforced.
  Upgrade procedure for a deployment without license will not take upgrade
  period into account for the read-only mode will enforcement.

* Fixed a bug that hotbackup upload could miss files (fixes BTS-734).

* BTS-590: When creating a new database in Web UI the value of the write concern
  has to be smaller or equal to the replication factor. Otherwise an error
  message will be displayed and no database will be created.

* Fixed potentially undefined behavior for exit code of arangodump.

* Ignore signals such as SIGPIPE in client tools.

* Validate that selected `writeConcern` value is less or equal to the selected
  `replicationFactor` value when creating new databases.

* Fixed issue #15476: FATAL {crash} occurs on a simple query.


v3.9.0-beta.1 (2022-01-06)
--------------------------

* Fixed ES-1025: fixed a performance regression caused by different hash
  calculation for primitive types like `uint64_t` and new the `Identifier`
  wrapper and derived types.

* BTS-707: rename "hardened" option value for `--server.support-info-api`
  startup option to "admin".

* APM-292: Added new AQL function SHARD_ID.

* Extend timeouts for caching collection counts and index selectivity estimates
  on coordinators from 15s/90s to 180s. This change will cause less requests to
  be made from coordinators to DB servers to refresh info about collection
  counts and index estimates as part of AQL queries. The cached info is used in
  cluster query execution plans only and is not required to be fully up-to-date.

* Improved performance in replication dump protocol by inserting arrays of
  documents instead of one document at a time and also not retrieving the
  document revision field when not needed.

* APM-78: Added startup security option `--foxx.allow-install-from-remote` to
  allow installing Foxx apps from remote URLs other than Github. The option is
  turned off by default.

* Fixed BTS-693: Sort-limit rule now always ensures proper LIMIT node placement
  to avoid possible invalid results in the fullCount data

* Updated OpenSSL to 1.1.1m and OpenLDAP to 2.6.0.

* Updated arangosync to 2.7.0.

* Fixed PRESUPP-439: In arangoimport, for CSV and TSV files, it could happen
  that a buffer containing only the header would be sent to the server, and also
  batches would contain the documents equivalent to the csv rows in them, but
  not the header, which should be sent together with the documents.

* Changed various default values for RocksDB to tune operations for different
  typical scenarios like gp2 type volumes and gp3 type volumes and locally
  attached SSDs with RAID0:
  - `--rocksdb.level0-slowdown-trigger` has been decreased from 20 to 16
  - `--rocksdb.level0-stop-trigger` has been increased from 36 to 256
  - `--rocksdb.max-background-jobs` has been increased to the number of cores
    and is no longer limited to 8
  - `--rocksdb.enabled-pipelined-write` is now `true` by default instead of
    `false`
  - `--rocksdb.throttle-frequency` has been decreased from 60000ms down to
    1000ms per iteration, which makes the RocksDB throttle react much quicker
  - `--rocksdb.pending-compactions-slowdown-trigger` has been decreased from 64
    GB down to 8 GB
  - `--rocksdb.pending-compactions-stop-trigger` has been decreased from 256 GB
    down to 16 GB
  - `--rocksdb.throttle-slots` has been increased from 63 to 120
  - `--rocksdb.encryption-hardware-acceleration` is now `true` by default,
    which helps performance and should not create any problems, since we
    require sandybridge anyway.
  Combined, these changes help ArangoDB/RocksDB to react quicker to a backlog of
  background jobs and thus to prevent catastrophic stops which abort data
  ingestion or lead to cluster internal timeouts.

* Adjust default value for startup option `--rocksdb.max-subcompactions` from 1
  to 2. This allows compactions jobs to be broken up into disjoint ranges which
  can be processed in parallel.

* Added startup options to adjust previously hard-coded parameters for RocksDB's
  behavior:

  - `--rocksdb.pending-compactions-bytes-slowdown-trigger` controls RocksDB's
    setting `soft_pending_compaction_bytes_limit`, which controls how many
    pending compaction bytes RocksDB tolerates before it slows down writes.
  - `--rocksdb.pending-compactions-bytes-stop-trigger` controls RocksDB's
    setting `hard_pending_compaction_bytes_limit`, which controls how many
    pending compaction bytes RocksDB tolerates before it stops writes entirely.
  - `--rocksdb.throttle-lower-bound-bps`, which controls a lower bound for the
    bandwidth restriction on RocksDB writes the throttle imposes.

* Allow initial, full dump shard synchronization to abort prematurely if it
  turns out that the follower was removed from the plan as a follower (e.g. if
  there are enough other in-sync followers).

* Set the limit for ArangoSearch segment size to 256MB during recovery to avoid
  OOM kill in rare cases.

* Cancel ongoing RocksDB compactions on server shutdown.

* Updated Enterprise license behavior: now there will be a one hour period for
  a new deployment to provide the license. After that period, the read-only mode
  will be enforced.

* Added startup options to adjust previously hard-coded parameters for the 
  RocksDB throttle:
  - `--rocksdb.throttle-frequency`: frequency for write-throttle calculations
    (in milliseconds, default value is 60000, i.e. 60 seconds).
  - `--rocksdb.throttle-slots`: number of historic measures to use for throttle
    value calculation (default value is 63).
  - `--rocksdb.throttle-scaling-factor`: adaptiveness scaling factor for write-
    throttle calculations (default value is 17). There is normally no need to
    change this value.
  - `--rocksdb.throttle-max-write-rate`: maximum write rate enforced by the
    throttle (in bytes per second, default value is 0, meaning "unlimited").
    The actual write rate will be the minimum of this value and the value the
    throttle calculation produces.
  - `--rocksdb.throttle-slow-down-writes-trigger`: number of level 0 files whose
    payload is not considered in throttle calculations when penalizing the 
    presence of L0 files. There is normally no need to change this value.

  All these options will only have an effect if `--rocksdb.throttle` is enabled
  (which is the default). The configuration options introduced here use the
  previously hard-coded settings as their default values, so there should not be
  a change in behavior if the options are not adjusted.

* Improve visibility in case of potential data corruption between primary index
  and actual document store in documents column family.

* Fixed BTS-611: In some cases AQL queries, in particular in a cluster, reported
  the wrong fullCount when the optimizer rule(s)`late-document-materialization`
  and/or `sort-limit` were active.

* Fix BTS-535: TakeoverShardLeadership waits properly for Current data in
  ClusterInfo. This avoids a fake warning and fake test failure.

* Fix potential read inconsistency for single document operations.
  When reading a single document that is concurrently being updated or replaced,
  the read operation could erroneously return a "document not found" error
  although the document actually existed. This only happened for single document
  operations, i.e., no transactions or AQL queries.

* APM-256: make arangoexport escape potential formulae in CSV exports.
  This addresses a potential security issue when exporting specially crafted
  documents to CSV, opening the CSV file in MS Excel or OpenOffice and then
  clicking links in any of the tainted cells.
  
  This change also adds a new option `--escape-csv-formulae` to toggle the
  escaping behavior for potential formulae values. The option is turned on by
  default.

* Second step of hotbackup transfer job cleanup. Now locks are also cleaned up
  as well as old, seemingly unfinished jobs.

* Fix GitHub issue #15084. Fixed a potential use-after-free on Windows for
  queries that used the NeighborsEnumerator (though other PathEnumerators
  might have been affected as well).

* BTS-624: The `move-calculations-up` optimization rule is now also applied to
  subqueries, when they don't have dependencies on the outer nodes, don't have
  modification nodes and don't read their own writes. This fixed the execution
  of a query without the splicing-subqueries option being faster than the
  execution of a query with this option (after version 3.8, this option cannot
  be switched off). 

* Added the following metrics for revision trees:
  - `arangodb_revision_tree_hibernations_total`: number of times a revision tree
    was compressed in RAM and set to hibernation
  - `arangodb_revision_tree_resurrections_total`: number of times a revision
    tree was resurrected from hibernation and uncompressed in RAM
  - `arangodb_revision_tree_memory_usage`: total memory usage (in bytes) by all
    active revision trees

* Added metric `rocksdb_wal_sequence` to track the current tip of the WAL's
  sequence number.


v3.9.0-alpha.1 (2021-11-30)
---------------------------

* Cleanup and properly fail hotbackup upload and download jobs if a dbserver
  fails or is restarted during the transfer. This gets rid of upload and
  download blockages in these circumstances.

* Make `db.<collection>.figures(true)` operate on the same snapshot when
  counting the number of documents in the documents column family and the
  indexes. This ensures consistency for the results of a single figures result.

* Upgraded bundled version of RocksDB to 6.27.

* Improved sync protocol to commit after each chunk and get rid of potentially
  dangerous NO_INDEXING optimization.

* Removed an invalid assertion that could be triggered during chaos testing in
  maintainer mode.

* Simplify the tagging of EnumerateCollectionNodes and IndexNodes with the
  "read-own-writes" flag. Previously the tagging only happened after all query
  optimizations were completed, making the tag unavailable to the optimizer.
  Now the tag is set early on, so it is accessible by the query optimizer.

* APM-187: The "Rebalance Shards" button now is displayed in a new tab, and it
  is displayed for any database in cluster mode. There is also a new flag for
  arangod, `--cluster.max-number-of-move-shards` (default = 10), which limits
  the amount of move shards operations each time the button is clicked to
  rebalance shards. When the button is clicked, the number of move shards
  operations scheduled is shown, or that no operation was scheduled if the flag
  `--cluster.max-number-of-move-shards` has a value of 0.

* Make the `--version` and `--version-json` commands usable in arangobackup when
  no positional argument (operation type) was specified. Previously,
  arangobackup insisted on specifying the operation type alongside the
  `--version` or `--version-json` commands.

* Fixed an issue in old incremental sync protocol with document keys that
  contained special characters (`%`). These keys could be send unencoded in the
  incremental sync protocol, leading to wrong key ranges being transfered
  between leader and follower, and thus causing follow-up errors and preventing
  getting in sync.

* APM-209: Histogram displaying is now switched off by default. For displaying
  it, the new flag `histogram.generate` must be set to true. Its default value
  is false for compatibility with other versions and also for complying with the
  histogram not being displayed by default. If this flag is not set to true, but
  other histogram flags are addressed, e.g. `--histogram.interval-size 500`,
  everything will still run normally, but a warning message will be displayed
  saying that the histogram is switched off and setting this flag would not be
  of use. When the flag is set to true, the histogram is displayed before the
  summary in the output.

* In the shards overview the list of servers to move the leader shard to, now
  also contains the current followers. This means that from now on also active
  follower servers can be nominated as the leading server for that specific
  shard.

* Extend Windows minidumps with memory regions referenced from CPU registers or
  the stack to provide more contextual information in case of crashes.

* Fix issues during rolling upgrades from 3.8.0 to 3.8.x (x >= 1) and from 3.7.x
  (x <= 12) to 3.8.3. The problem was that older versions did not handle
  following term ids that are sent from newer versions during synchronous
  replication operations.

* Increase default stack size on Windows from 1MB to 4MB. This should allow
  execution of larger queries without overflowing the stack.

* Make background calculation of SHA hashes for RocksDB .sst files less
  intrusive. The previous implementation frequently iterated over all files in
  the database directory to check if it needed to ad-hoc calculate the SHA
  hashes for .sst files it previously missed. The procedure it used was to
  iterate over all files in the database directory and check if there were
  matching pairs of .sst files and .sha files. This was expensive, because a
  full directory iteration was performed and a lot of temporary strings were
  created for filenames and used in comparisons. This was especially expensive
  for larger deployments with lots of .sst files.
  The expensive iteration of files in the directory is now happening less
  frequently, and will not be as expensive as before if it runs.

* Close a potential gap during shard synchronization when moving from the
  initial sync step to the WAL tailing step. In this small gap the leader could
  purge some of the WAL files that would be required by the following WAL
  tailing step. This was possible because at the end of the initial sync step,
  the snapshot on the leader is released, and there is a small window of time
  before the follower will issue its first WAL tailing request.

* Improve Shards overview in web UI: the number of currently syncing shards is
  now displayed per collection. Additionally, shards on failed servers are now
  displayed in a different color.

* Fixed BTS-637: Slow SynchronizeShard jobs which need to copy data could block
  quick SynchronizeShard jobs which have the data and only need to resync.

* DEVSUP-899: Fixed Subquery execution in a very rare case a subquery, nested in
  another subquery, was not executed, which is fixed now.
  Technical details:
  If we have two subqueries: `Outer` and `Nested` the Outer will define the
  input for Nested. And Outer has the pattern: 1 input, subqueryDone, 1 input,
  subqueryDone [...] and our internal batching did cut a batch like this:
  [<...>, input (A)] | [subqueryDone, input (B), subqueryDone, <...>] than
  Nested on input (B) was not executed. As soon as we have more than 1 input per
  Outer, or a different cutting position, all was good.

* When enabling the cluster supervision maintenance mode via the web UI, there
  is now the possibility to select a duration for the maintenance mode.
  Previous versions of ArangoDB always enabled the maintenance mode for one
  hour, without allowing any choice here.

* Stop calling unnecessary `/_api/wal/open-transactions` REST API before
  starting the continuous synchronization in active fail and single server
  replication. This request is unnecessary with the RocksDB storage engine.

* Fixed potential undefined behavior in edge cache during cache migration tasks.
  There was a short window of time in which an already freed Table could be used
  by concurrently running edge lookups.

* BTS-623: The audit log messages, when written, were not showing the log level
  of the message, as in the example:
    `2021-10-21T02:28:42Z | hostname | audit-authentication | n/a | _system |
     127.0.0.1:52490 | n/a | credentials missing | /_admin/aardvark/favicon.ico`
  With the new flag `--audit.display-log-level`, the level of the audit log
  message can be displayed in the log text.  When set to true, this behavior is
  expected, as in the example:
    `2021-10-21T02:28:42Z | DEBUG | hostname | audit-authentication | n/a |
     _system | 127.0.0.1:52490 | n/a | credentials missing |
     /_admin/aardvark/favicon.ico`
  The default value for the flag is false for compatibility with former
  versions. When this flag is not used, it is considered to have the default
  behavior (that is, set to false).

* Fixed SEARCH-261: Fix possible race between file creation and directory
  cleaner (ArangoSearch).

* Fixed SEARCH-260: Fix invalid sorting order of stored features in presence of
  primary sort (ArangoSearch).

* Change error message for queries that use too much memory from "resource limit
  exceeded" to "query would use more memory than allowed".

* When using Indexes within traversals (e.g. [_from, date]) and filter based on
  a function (e.g. FILTER path.edges[0].date <= DATE_ADD(@now, 5, "day")) this
  function was passed through to the index. The index cannot evaluate this
  function and returned incorrect results. Now all functions are evaluted before
  looking into the index. (Fixes BTS-407)

* Old license mechanism for docker containers removed.

* arangorestore: Fix the order (regarding distributeShardsLike) in which
  collections are being created during restore, which could result in an error
  and make manual intervention necessary.

* Single server license output checking fixed.

* Updated ArangoDB Starter to 0.15.3.

* Fix caching of collection counts and index selectivity estimates in cluster.
  The cache values expired too early in previous versions, making the cache
  ineffective.

* Add better error message for replication request failures in case requests are
  retried.

* Make background statistics gathering more efficient by avoiding one AQL query
  every 10 seconds that fetched the most recent stats entry. Instead, buffer the
  entry in value after we have written it. Also spread out the statistics calls
  by different servers more randomly, so that request spikes are avoided for
  cluster with many coordinators that used to run their statistics queries at
  about the same time when the instances were started simultaneously.

* Fix a potential overwhelm situation on DB servers that can lead to no further
  tasks being pulled from a DB servers queue even though there would still be
  processing capacity and idle threads available.

* Fixed compilation and linking when using glibc 2.34.

* Fuerte: don't fall back to identity encoding in case of unknown encoding.

* Fixed ES-881: Fixed LDAP global options. This needs to use the first active
  provider, not just the first provider and it should be globally disabled.

* Web UI: Fixes the loading of map tiles which are being used to display the
  query output based on a world map when using SSL encryption. This lead to not
  displaying some world map tiles correctly (OASIS-590).

* Web UI - Added missing HTML escaping inside the file upload plugin used in the
  section of deploying a new Foxx application when uploading a zip file.

* Now, arangoimport supports merging of attributes. When importing data from a
  file into a collection, a document attribute can be comprised of merging
  attributes from the file into it, with separators and other literal strings.
  The new document attribute will result in the concatenation of the literal
  strings, the values of the attributes and the separators, as in the example:

      arangoimport --merge-attributes fullName=[firstName]:[lastName]

* Do not use an edge index for range queries, i.e. with the comparison operators
  `>`, `>=`, `<` or `<=`, but only for equality lookups using the `==` and `IN`
  comparison operators.
  The edge index is not fully ordered, so while using it for range queries may
  produce _some_ documents, it is possible that other documents from the range
  would be skipped.

* Do not rename the arangod process to "arangod [shutting down]" during the
  server shutdown. The renaming can cause issues with tools that look for
  the exact process name "arangod".

* Remove internal AQL query option `readCompleteInput` that controled if all
  input for a modification operation (UPDATE / REPLACE / REMOVE) are read into
  memory first. This was a necessity with the MMFiles storage engine in cases
  when a query read from a collection and wrote into it in the same query
  afterwards. With the RocksDB engine and its snapshots, we never need to read
  the entire input into memory first.

* Fix windows installer PATH manipulation issue by replaceng the NSIS plugin
  (BTS-176).

* Fixed counting of all read transaction as aborted. Added a new metric to count
  read transactions.

* Fixed potential issues with revision trees and document counters getting out
  of sync with the underlying collection data.

* Fix race in RocksDB throttle listener, when it was getting started lazily
  during server shutdown.

* Extended the Views web UI by letting it capture View properties that are
  immutable once created.

* Fixed BTS-602 by not starting license feature is upgrade mode.

* APM-173: Now, arangobench, arangodump and arangorestore support multiple
  coordinators, so the flag `--server.endpoint` can be used multiple times, as
  in the example below:

  arangobench \
    --server.endpoint tcp://[::1]::8529 \
    --server.endpoint tcp://[::1]::8530 \
    --server.endpoint tcp://[::1]::8531

  This does not compromise the use of the other client tools, which preserve the
  behavior of having one coordinator.

* The server now has two flags to control the escaping control and Unicode
  characters in the log. The flag `--log.escape` is now deprecated and, instead,
  the new flags `--log.escape-control-chars` and `--log.escape-unicode-chars`
  should be used. 

  - `--log.escape-control-chars`: this flag applies to the control characters,
    which have hex code below `\x20`, and also the character DEL, with hex code
    of `\x7f`. When its value is set to false, the control character will be
    retained, and its actual value will be displayed when it is a visible
    character, or a space ` ` character will be displayed if it is not a visible
    character. The same will happen to `DEL` character (code `\xF7`), even
    though it is not a control character, because it is not visible. For
    example, control characer `\n` is visible, so a `\n` will be displayed in
    the log, and control character `BEL` is not visible, so a space ` ` would be
    displayed. When its value is set to true, the hex code for the character is
    displayed, for example, `BEL` character would be displayed as its hex code,
    `\x07`.
    The default value for this flag is `true` for compatibility with previous
    versions.

  - `--log.escape-unicode-chars`: when its value is set to false, the unicode
    character will be retained, and its actual value will be displayed. For
    example, `犬` will be displayed as `犬`. When its value is set to true, the
    character is escaped, and the hex code for the character is displayed. For
    example, `犬` would be displayed as its hex code, `\u72AC`.
    The default value for this flag is `false` for compatibility with previous
    versions.

* Fixed BTS-582: ArangoDB client EXE package for Windows has incorrect metadata.

* Fixed BTS-575: Windows EXE installer doesn't replace service during upgrade in
  silent (non-UI) mode.

* APM-121: allow the UPSERT query to have indexHint as an extra parameter for
  OPTIONS. It will be used as a hint by the inner FOR loop that is performed as
  part of the UPSERT query, and would help in cases such as UPSERT not picking
  the best index automatically for lookup.

* Fix issue #14819: Query: AQL: missing variable # for node #... location
  RestCursorHandler.cpp.

* Added enterprise licensing support including (only for Enterprise version):
    - additional API endpoint `_admin/license(GET/PUT)?force=true <text>`
    - arangosh functions: `setLicense(<text>)`, `getLicense()`
    - new error codes and metrics support

* Fix issue #14807: Fix crash during optimization of certain AQL queries during
  the remove-collect-variables optimizer rule, when a COLLECT node without
  output variables (this includes RETURN DISTINCT) occured in the plan.

* Update iresearch library to the upstream. Fixed TSan/ASan detected issues.

* Added new ArangoSearch analyzer type 'collation'.

* Add basic overload control to arangod.
  This change adds the `x-arango-queue-time-seconds` header to all responses
  sent by arangod. This header contains the most recent request dequeing time
  (in seconds) as tracked by the scheduler. This value can be used by client
  applications and drivers to detect server overload and react on it.
  The new startup option `--http.return-queue-time-header` can be set to `false`
  to suppress these headers in responses sent by arangod.

  In addition, client applications and drivers can optionally augment their
  requests sent to arangod with a header of the same name. If set, the value of
  the header should contain the maximum queuing time (in seconds) that the
  client is willing to accept. If the header is set in an incoming request,
  arangod will compare the current dequeing time from its scheduler with the
  maximum queue time value contained in the request. If the current dequeing
  time exceeds the value set in the header, arangod will reject the request and
  return HTTP 412 (precondition failed) with the new error code 21004 (queue
  time violated).

  There is also a new metric `arangodb_scheduler_queue_time_violations_total`
  that is increased whenever a request is dropped because of the requested queue
  time not satisfiable.

* Fixed a bug for array indexes on update of documents (BTS-548).

* Prevent some possible deadlocks under high load regarding transactions and
  document operations, and also improve performance slightly.

* Hide help text fragment about VST connection strings in client tools that do
  not support VST.

* Added REST API endpoint `/_admin/debug/failat/all` to retrieve the list of
  currently enabled failure points. This API is available only if failure
  testing is enabled, but not in production.

* APM-60: optionally allow special characters and Unicode characters in database
  names.

  This feature allows toggling the naming convention for database names from the
  previous strict mode, which only allowed selected ASCII characters in database
  names, to an extended, more relaxed mode. The extended mode allows additional
  ASCII characters in database names as well as non-ASCII UTF-8 characters.
  The extended mode can be enabled by setting the new startup option
  `--database.extended-names-databases` to true. It is turned off by default and
  requires an explicit opt-in, simply because some drivers and client
  applications may not be ready for it yet. The arangod server, the ArangoDB web
  interface and the following bundled client tools are prepared and ready for
  using the extended database names:
  - arangobench
  - arangodump
  - arangoexport
  - arangoimport
  - arangorestore
  - arangosh
  More tools and the drivers shipped by ArangoDB will be added to the list in
  the future. 

  Please note that the extended names for databases should not be turned on
  during upgrades from previous versions, but only once the upgrade has been
  completed successfully. In addition, the extended names should not be used in
  environments that require extracting data into a previous version of ArangoDB,
  or when database dumps may be restored into a previous version of ArangoDB.
  This is because older versions may not be able to handle the extended database
  names. Finally, it should not be turned on in environments in which drivers
  are in use that haven't been prepared to work with the extended naming
  convention.

  Warning: turning on the `--database.extended-names-databases` option for a
  deployment requires it to stay enabled permanently, i.e. it can be changed
  from `false` to `true` but not back. When enabling it, it is also required to
  do this consistently on all coordinators and DB servers.

  The extended names for databases will be enabled by default in one of the
  future releases of ArangoDB, once enough drivers and other client tools have
  had the chance to adapt.

  Naming conventions for collections, views, analyzers, and document keys
  (`_key` values) are not affected by this feature and will remain as in
  previous versions of ArangoDB.

* Prevent stealing of values from AQL const value registers. This fixes an issue
  for queries that produce constant results (known at query compile time) when
  he queries are executed directly on a DB server in a cluster (which is not
  supported, but may happen for troubleshooting).

* Fixed BTS-562: reduce-extraction-to-projection optimization returns null for
  one attribute if nested attributes are named the same.

* Add `--datatype` startup option to arangoimport, in order to hard-code the
  datatype (null/boolean/number/string) for certain attributes in the CSV/TSV
  import.
  For example, given the following input file:

      key,price,weight,fk
      123456,200,5,585852
      864924,120,10,9998242
      9949,70,11.5,499494
      6939926,2130,5,96962612

  When invoking arangoimport with the startup options

      --datatype key=string 
      --datatype price=number 
      --datatype weight=number 
      --datatype fk=string

  it will turn the numeric-looking values in "key" into strings (so that they
  can be used in the `_key` attribute), but treat the attributes "price" and
  "weight" as numbers. The values in attribute "fk" finally will be treated as
  strings again (potentially because they are used for linking to other "_key"
  values).

* Avoid the acquisition of a recursive read lock on server shutdown, which could
  in theory lead to shutdown hangs at least if a concurrent thread is trying to
  modify the list of collections (very unlikely and never observed until now).

* Fixed display of unicode characters in Windows console.

* Fixed issue BTS-531 "Error happens during EXE package installation if
  non-ASCII characters are present in target path".
  
* Fix active failover, so that the new host actually has working Foxx services
  (BTS-558).

* Fixed issue #14720: Bulk import ignores onDuplicate in 3.8.0.
  The "onDuplicate" attribute was ignored by the `/_api/import` REST API when
  not specifying the "type" URL parameter.

* Updated OpenSSL to 1.1.1l and OpenLDAP to 2.4.59.

* APM-70: allow PRUNE condition to be stored in a variable.

  This feature allows the PRUNE condition to be stored in a variable, and this
  variable can be used as a condition for some other statement, such as FILTER.

* Allow startup of arangod with an existing database directory that was missing
  the ZkdIndex column family.

* Truncate must not trigger intermediate commits while in a streaming
  transaction, because that would be against the assumption that streaming
  transactions never do intermediate commits.

* Added ArangoSearch condition optimization: STARTS_WITH is merged with
  LEVENSHTEIN_MATCH if used in the same AND node and field name and prefix
  matches.

* Hybrid (Disjoint) SmartGraphs (Enterprise Edition):
  SmartGraphs have been extended with a new option to create Hybrid SmartGraphs.
  Hybrid SmartGraphs are capable of using SatelliteCollections within their
  graph definition. You can now select some VertexCollections to be satellites,
  and therefore available on all DBServers. The SmartGraph can make use of those
  to collections to increase the traversal performance by larger local
  components.

* Added multidimensional indexes which can be used to efficiently intersect
  multiple range queries. They are currently limited to IEEE-754 double values.
  Given documents of the form {x: 12.9, y: -284.0, z: 0.02} one can define a
  multidimensional index using the new type 'zkd' on the fields ["x", "y", "z"].

  The AQL optimizer will then consider this index when doing queries on multiple
  ranges, for example:

    FOR p IN points
        FILTER x0 <= p.x && p.x <= x1
        FILTER y0 <= p.y && p.y <= y1
        FILTER z0 <= p.z && p.z <= z1
        RETURN p

  The index implements the relation <=, == and >= natively. Strict relations are
  emulated using post filtering. Ranges can be unbounded on one or both sides.

* No runtime limits for shard move and server cleanout jobs, instead possibility
  to cancel them.

* Fix cluster-internal network protocol to HTTP/1 for now. Any other protocol
  selected via the startup option `--network.protocol` will automatically be
  switched to HTTP/1. The startup option `--network.protocol` is now deprecated
  and hidden by default. It will be removed in a future version of arangod.
  The rationale for this change is to move towards a single protocol for
  cluster-internal communication instead of 3 different ones.

* Disable RTTI when compiling Snappy. RTTI used to be disabled previously, up
  until some Merkle tree improvement PR was merged about one month ago, which
  turned on RTTI for compiling Snappy.

* (EE only) Bug-fix: If you created a ArangoSearch view on satellite collections
  only and then join with a collection only having a single shard the
  cluster-one-shard-rule was falsely applied and could lead to empty view
  results. The Rule will now detect the situation properly, and not trigger.

* (EE only) If you have a query using only satellite collections, now the
  cluster-one-shard-rule can be applied to improve query performance.

* (Enterprise Edition only): added query option `forceOneShardAttributeValue`
  to explicitly set a shard key value that will be used during query snippet
  distribution to limit the query to a specific server in the cluster.

  This query option can be used in complex queries in case the query optimizer
  cannot automatically detect that the query can be limited to only a single
  server (e.g. in a disjoint smart graph case).
  When the option is set to the correct shard key value, the query will be
  limited to the target server determined by the shard key value. It thus
  requires that all collections in the query use the same distribution (i.e.
  `distributeShardsLike` attribute via disjoint SmartGraphs).

  Limiting the query to a single DB server is a performance optimization and may
  make complex queries run a lot faster because of the reduced setup and
  teardown costs and the reduced cluster-internal traffic during query
  execution.

  If the option is set incorrectly, i.e. to a wrong shard key value, then the
  query may be shipped to a wrong DB server and may not return results (i.e.
  empty result set). It is thus the caller's responsibility to set the
  `forceOneShardAttributeValue` correctly or not use it.

  The `forceOneShardAttributeValue` option will only honor string values.
  All other values as well as the empty string will be ignored and treated as if
  the option is not set.

  If the option is set and the query satisfies the requirements for using the
  option, the query's execution plan will contain the "cluster-one-shard"
  optimizer rule.

* SEARCH-238: Improved SortNodes placement optimization in cluster so
  late materialization could cover more cases

* Fix some memory leaks after adding optimization rule for AqlAnalyzer.

* Fix internal iterator states after intermediate commits in write transactions.
  Iterators could point to invalid data after an intermediate commit, producing
  undefined behavior.

* Fix read-own-write behavior in different scenarios:
    - in some cases writes performed by an AQL query could be observed within
      the same query. This was not intended and is fixed now.
    - AQL queries in streaming transactions could observe their own writes in
      even more cases, which could potentially result in an endless loop when
      the query iterates over the same collection that it is inserting documents
      into.
    - UPSERT did not find documents inserted by a previous iteration if the
      subquery relied on a non-unique secondary index.
    - disabled intermediate commits for queries with UPSERTs, because
      intermediate commits can invalidate the internal read-own-write iterator
      required by UPSERT. Previously, UPSERTs that triggered intermediate
      commits could have produced unexpected results (e.g., previous inserts
      that have been committed might not be visible) or even crashes.
  To achieve the correct read-own-write behavior in streaming transactions, we
  sometimes have to copy the internal WriteBatch from the underlying RocksDB
  transaction. In particular, the copy is created whenever an AQL query with
  modification operations (INSERT/REMOVE/UPDATE/UPSERT/REPLACE) is executed in
  the streaming transaction. If there have not been any other modifications so
  far (queries/document operations), then the WriteBatch is empty and creating
  the copy is essentially a no-op. However, if the transaction already contains
  a lot of modifications, creating the WriteBatch copy might incur some
  overhead that can now lead to decreased performance.

* Fix rare case of invalid data that could be inserted into the ArangoSearch
  index if several clients concurrently insert data and use custom analyzer
  with non-string return type.

* Fix a rare shutdown race in RocksDBShaCalculatorThread.

* Added "Analyzers" view to web UI to let manage ArangoSearch analyzers
  creation.

* Add pseudo log topic "all" to set the log levels for all log topics at once.
  For example, this can be used when starting a server with trace or debug
  logging enabled for all log topics, e.g.

  `--log.level all=debug`
  `--log.level all=trace`

  This is very coarse and should only be used for such use cases.

* Change the default value for the `--threads` startup parameter of the
  following client tools from previously 2 to the maximum of 2 and the number of
  available CPU cores:
  - arangodump
  - arangoimport
  - arangorestore

* Remove old fixPrototypeChain agency migration, which was introduced in 3.2 and
  is no longer necessary. This will make it impossible to upgrade directly from
  a version < 3.2 to a version >= 3.9, provided one has a chain of
  `distributeShardsLike` collections.

* Added metrics for the number of errors and warnings logged:
  - `arangodb_logger_warnings_total`: total number of warnings (WARN messages)
    logged since server start
  - `arangodb_logger_errors_total`: total number of errors (ERR messages)
    logged since server start

* Added REST API `/_admin/support-info` to retrieve deployment information.
  As this API may reveal sensitive data about the deployment, it can only be
  accessed from inside the system database. In addition, there is a policy
  control startup option `--server.support-info-api` that determines if and to
  whom the API is made available. This option can have the following values:
  - `disabled`: support info API is disabled.
  - `jwt`: support info API can only be accessed via superuser JWT.
  - `hardened`: if `--server.harden` is set, the support info API can only be
    accessed via superuser JWT. Otherwise it can be accessed by admin users
    only.
  - `public`: everyone with access to `_system` database can access the support
    info API.

* Send a keystroke to arangod's stdin when a shutdown command is received via
  the REST API `/_admin/shutdown` and the server is started with the `--console`
  argument. The keystroke will exit the blocking read loop that is waiting on
  console input and that otherwise blocks the shutdown.
  The implementation is based on ioctl and is thus only present on Linux and
  macOS.

* Some AQL queries erroneously reported the "access after data-modification"
  error for queries in which there was a read attempt from a collection _before_
  a data-modification operation. Such access is legal and should not trigger
  said error anymore. Accessing a collection _after_ in a query a
  data-modification in the same query is still disallowed.

* Make AQL modification operations in a cluster asynchronous. This allows to
  free the thread for other work until both the write and synchronous
  replication are complete.

* Fixed: /_api/transaction/begin called on edge collections of disjoint
  SmartGraphs falsely returned CollectionNotFound errors.

* Bugfix: In more complex queries there was a code-path where a (Disjoint-)Smart
  graph access was not properly optimized.

* Add ReplicatedLogs column family.

* Add optimization rule for AqlAnalyzer.

* Change optimization level for debug builds back to `-O0` (from `-Og`) because
  `-Og` seems to cause debuggability issues in some environments.

* Automatically extend web UI sessions while they are still active.
  The web UI can now call a backend route to renew its JWT, so there will not be
  any rude logouts in the middle of an active session.

  Active web UI sessions (here: sessions with user activity within the last 90
  minutes) will automatically renew their JWT if they get close to the JWT
  expiry date.

* Reduce memory usage for in-memory revision trees. Previously, a revision tree
  instance for a non-empty collection/shard was using 4 MB of memory when
  uncompressed. Trees that were unused for a while were compressed on the fly to
  use less memory, and later uncompressed again when needed.
  Now the uncompressed in-memory version of the revision tree will dynamically
  allocate memory as needed. This allows the initial version of the trees to get
  away with just 64 KB of memory. Memory usage will grow lazily when more parts
  of the trees get populated. The compression of unused in-memory tree data is
  still in place.

* Refactored arangobench:
  - Updated testcases to show description of them when beginning execution
  - Fixed testcase histogram with time measures when batch size > 0
  - Integrated testcases with Velocypack for simplification
  - Deprecated some testcases
  - Internal changes for performance optimization

* Add 3 AQL functions: COSINE_SIMILARITY, L1_DISTANCE and L2_DISTANCE.

* Honor the value of startup option `--rocksdb.sync-interval` on Windows, too.
  Previously, the value was ignored and WAL syncing on Windows was using a
  different code paths than on the other supported platforms. Now syncing is
  unified across all platforms, and they all call RocksDB's `SyncWAL()`.

* APM-132: Clean up collection statuses.
  Removes collection statuses "new born", "loading", "unloading" and "unloaded".
  These statuses were last relevant with the MMFiles storage engine, when it was
  important to differentiate which collections are present in main memory and
  which aren't. With the RocksDB storage engine, all that was automatically
  handled anyway, and the statuses were not important anymore.

  The change removes the "Load" and "Unload" buttons for collections from the
  web interface. All collections in the web interface will be marked as "loaded"
  permanently.

  This change also obsoletes the `load()` and `unload()` calls for collections
  as well as their HTTP API equivalents. The APIs will remain in place for now
  but are changed to no-ops. They will removed eventually in a future version of
  ArangoDB. This will be announced separately.

* Reduce default value for max-nodes-per-callstack to 200 for macOS, because on
  macOS worker threads have a stack size of only 512kb.

* Slightly increase internal AQL query and transaction timeout on DB servers
  from 3 to 5 minutes.
  Previously, queries and transactions on DB servers could expire quicker, which
  led to spurious "query ID not found" or "transaction ID not found" errors on
  DB servers for multi-server queries/transactions with unbalanced access
  patterns for the different participating DB servers.
  The timeouts on coordinators remain unchanged, so any queries/transactions
  that are abandoned will be aborted there, which will also be propagated to DB
  servers. In addition, if a particpating server in an AQL query becomes
  unavailable, the coordinator is now notified of that and will terminate the
  query more eagerly.

* Add hard-coded complexity limits for AQL queries, in order to prevent
  programmatically generated large queries from causing trouble (too deep
  recursion, enormous memory usage, long query optimization and distribution
  passes etc.).
  This change introduces 2 limits:
  - a recursion limit for AQL query expressions. An expression can now be up to
    500 levels deep. An example expression is `1 + 2 + 3 + 4`, which is 3 levels
    deep `1 + (2 + (3 + 4))`.
    The expression recursion is limited to 500 levels.
  - a limit for the number of execution nodes in the initial query execution
  plan.
    The number of execution nodes is limited to 4,000.

* Remove _msg/please-upgrade handler.

* Adapt various places related to handling of execution plans non-recursive in
  order to avoid stack overflows. This allows us now to execute much larger
  queries.

* Fix locking of AQL queries write queries on DB servers.

* APM-112: invalid use of OPTIONS in AQL queries will now raise a warning in the
  query.
  The feature is useful to detect misspelled attribute names in OPTIONS, e.g.

      INSERT ... INTO collection
        OPTIONS { overwrightMode: 'ignore' } /* should be 'overwriteMode' */

  It is also useful to detect the usage of valid OPTIONS attribute names that
  are used for a wrong query part, e.g.

      FOR doc IN collection
        FILTER doc.value == 1234
        INSERT doc INTO other
          OPTIONS { indexHint: 'myIndex' } /* should be used above for FOR */

  In case a wrong option attribute is used, a warning with code 1575 will be
  raised.
  By default, warnings are reported but do not lead to the query being aborted.
  This can be toggled by the startup option `--query.fail-on-warnings` or the
  per-query runtime option `failOnWarnings`.

* Added new command line-option `--version-json`. This will return the version
  information as json object.

* Fix ArangoAgency::version(), which always returned an empty string instead of
  the agency's correctly reported version. This also fixes the agency version in
  the startup log messages of the cluster.

* Fix potential memleak in Pregel conductor garbage collection.

* Added garbage collection for finished and failed Pregel conductors.
  Previously, Pregel executions that finished successfully or unsuccessfully
  remained in memory until being explicitly canceled. This prevented a cleanup
  of abandoned jobs. Such jobs are now automatically cleaned about 10 minutes
  after finalization. The time-to-live values can be overriden per Pregel job by
  passing a "ttl" value.

* Revive startup parameter `--server.session-timeout` to control the timeout for
  web interface sessions and other sessions that are based on JWTs created by
  the `/_open/auth` API.

  This PR also changes the default session timeout for web interface sessions to
  one hour. Older versions of ArangoDB had longer session timeouts.

* Removed redirects from /_admin/cluster* to /_admin/cluster/*. Adjusted
  internal requests to use the new url.

* Fix potential stack overflow when executing large queries. This is achieved by
  splitting the callstack and moving part of the execution to a separate thread.
  The number of execution nodes after which such a callstack split should be
  performed can be configured via the query option `maxNodesPerCallstack` and
  the command line option `--query.max-nodes-per-callstack`; the default is 250.

* Fixed invalid shard synchronization for documents not added via INSERT with
  `overwriteMode` set to `ignore`. In this case, if a document with the given
  key already exists, it is not changed on the leader (i.e. no write happens on
  the leader). However, a write was replicated to the follower, which was wrong.
  This write is now suppressed, which can only make such insert operations
  faster.

* Web UI: Disables the hover tooltip within the statistics view of the memory
  consumption chart.

* Add 3 AQL functions: DECAY_GAUSS, DECAY_EXP and DECAY_LINEAR.

* Implemented an optimization for Traversals. If you apply a POST filter on the
  vertex and/or edge result this filter will now be applied during the traversal
  to avoid generating the full output for AQL. This will have positive effect if
  you filter on the vertex/edge but return the path, this way the system does
  only need to produce a path that is allowed to be passed through. E.g.

    FOR v,e,p IN 10 OUTBOUND @start GRAPH "myGraph"
      FILTER v.isRelevant == true
      RETURN p

  can now be optimized, and the traversal statement will only produce paths
  where the last vertex has `isRelevant == true`.

* Fix BTS-446: When finding a not yet fully initialized agency, do not
  immediately fatal exit. Keep trying for (very generous) 5 minutes.

* Reduced the agency store public members, for simpler support long-term.

* Added a number of tests for the Agency Store public members.

* Updated bundled version of Snappy library to 1.1.9.

* Introduce a new internal error code for cases where a call cannot succeed
  because the server startup phase is still in progress. This error will be
  mapped to the HTTP status code 503 (service unavailable).
  One example where this can happen is when trying to authenticate a request,
  but the _users collection is not yet available in the cluster.

* Fix DEVSUP-749: Fix potential deadlock when executing concurrent view/link
  DDL operations and index DDL operations on the same collection.

* Fixed issue #14122: when the optimizer rule "inline-subqueries" is applied,
  it may rename some variables in the query. The variable renaming was however
  not carried out for traversal PRUNE conditions, so the PRUNE conditions
  could still refer to obsolete variables, which would make the query fail with
  errors such as

      Query: AQL: missing variable ... for node ... while planning registers

* Fixed the error response if the HTTP version is not 1.0 or 1.1 and if
  the Content-Length is too large (> 1 GB).

* Add a connection cache for internal replication requests.

* Improve legibility of size values (by adding KB, MB, GB, TB suffixes) to
  output generated by client tools.

* Timely updates of rebootId / cluster membership of DB servers and
  coordinators in ClusterInfo. Fixes BTS-368 detected in chaos tests.

* Guarded access only to ActionBase::_result.

* Fixed proper return value in sendRequestRetry if server is shutting down.

* Fixed internal issue #798: In  rare case when remove request
  completely cleans just consolidated segment commit could be cancelled
  and documents removed from collection may be left dangling in the ArangoSearch index.
  Also fixes ES-810 and  BTS-279.

* Retry if an ex-leader can no longer drop a follower because it is no longer
  leading.

* Fixed a small problem in fuerte which could lead to an assertion failure.

* Upgrade jemalloc version to latest stable dev.

* Fixed issue BTS-373: ASan detected possible heap-buffer-overflow at
  arangodb::transaction::V8Context::exitV8Context().

* Allow to specify a fail-over LDAP server. Instead of "--ldap.OPTION" you need
  to specify "--ldap2.OPTION". Authentication / Authorization will first check
  the primary LDAP server. If this server cannot authenticate a user, it will
  try the secondary one. It is possible to specify a file containing all users
  that the primary (or secondary) LDAP server is handling by specifying the
  option "--ldap.responsible-for". This file must contain the usernames
  line-by-line.

* Make the time-to-live (TTL) value of a streaming cursor only count after
  the response has been sent to the client.

* Improve performance of batch CRUD operations (insert, update, replace,
  remove) if some of the documents in the batch run into write-write conflicts.
  Rolling back partial operations in case of a failure is very expensive
  because it requires rebuilding RocksDB write batches for the transaction
  from scratch. Rebuilding write batches takes time proportional to the number
  of operations in the batch, and for larger batches the cost can be
  prohibitive.
  Now we are not rolling back write batches in some situations when this is
  not required, so that in many cases running into a conflict does not have
  that high overhead. There can still be issues when conflicts happen for index
  entries, but a lot of previously problematic cases should now work better.

* Allow AQL variable names starting with an underscore, as stated in the docs.

* Fix crashes during arangorestore operations due to usage of wrong pointer
  value for updating user permissions.

* Added option `--query-max-runtime` to arangoexport, in order to control
  maximum query runtime.

* Fix BTS-340: AQL expressions similar to `x < 3 || x` are no longer erroneously
  be reduced to `x < 3` by the optimizer rule remove-redundant-or.

* Changed default value of arangodump's `--envelope` option from `true` to
  `false`. This allows using higher parallelism in arangorestore when
  restoring large collection dumps. As a side-effect, this will also decrease
  the size of dumps taken with arangodump, and should slightly improve dump
  speed.

* Improve parallelism capabilities of arangorestore.

  arangorestore can now dispatch restoring data chunks of a collection to idle
  background threads, so that multiple restore requests can be in flight for
  the same collection concurrently.

  This can improve restore speed in situations when there are idle threads
  left (number of threads can be configured via arangorestore's `--threads`
  option) and the dump file for the collection is large.

  The improved parallelism is only used when restoring dumps that are in the
  non-enveloped format. This format has been introduced with ArangoDB 3.8.
  The reason is that dumps in the non-enveloped format only contain the raw
  documents, which can be restored independent of each other, i.e. in any
  order. However, the enveloped format may contain documents and remove
  operations, which need to be restored in the original order.

* Fix BTS-374: thread race between ArangoSearch link unloading and storage
  engine WAL flushing.

* Fix thread race between ArangoSearch link unloading and storage engine
  WAL flushing.

* Add value of `_key` to more insert/update/replace/remove error messages so it
  is easier to figure out which document caused unique constraint violations
  and/or write-write conflict during a multi-document write operation.

* Don't display obsoleted startup options and sections in `--help` and
  `--help-.` commands. Also rename "global" to "general" options.

* Removed assertion for success of a RocksDB function. Throw a proper exception
  instead.

* Experimentally switch to wyhash (from xxhash) for velocypack. This is an
  experiment in devel to check if it produces any observable speedups.

* Remove deprecated HTTP REST API `/_api/export`. This API was deprecated in a
  previous version because it was not supported in clusters and was also covered
  completely by streaming AQL queries for the RocksDB storage engine.

* Added enterprise-build-repository and oskar-build-repository to `--version` as
  `enterprise-build-repository` and `oskar-build-repository`.

* Clean up replication code and remove a 3.2-compatibility mode that was only
  useful when replicating from a leader < ArangoDB version 3.3.

* Obsolete option `--database.old-system-collections`. This option has no
  meaning in ArangoDB 3.9, as old system collections will not be created anymore
  in this version. The option was deprecated in 3.8 and announced to be
  obsoleted.

* Upgrade velocypack to latest, C++17-only version.

* Make arangovpack more powerful, by supporting different input and output
  formats (json and vpack, plain or hex-encoded).
  The arangovpack options `--json` and `--pretty` have been removed and have
  been replaced with separate options for specifying the input and output types:
    - `--input-type` ("json", "json-hex", "vpack", "vpack-hex")
    - `--output-type` ("json", "json-pretty", "vpack", "vpack-hex")
  The previous option `--print-non-json` has been replaced with the option
  `--fail-on-non-json` which makes arangovpack fail when trying to emit non-JSON
  types to JSON output.

* Remove obsolete API endpoint /_admin/repair/distributeShardsLike`. This API
  was intended to correct some bad state introduced before 3.2.12 or 3.3.4,
  respectively. It had to be invoked manually by callers and there was never any
  driver support for it.

* Remove now-unused SubqueryExecutor. This is an internal change only and should
  not have any effect on queries, as from 3.8 onwards only spliced subqueries
  should be used in query execution plans and during query execution.

* Switched to GCC 10 as the default compiler and use Sandy Bridge as the default
  required architecture (Linux, macOS binaries).

* Removed obsolete metrics in new v2 metric API. Those metrics' values were
  identical to the sum value of histograms.

* Fix potentially undefined behavior when creating a
  CalculationTransactionContext for an ArangoSearch analyzer. An uninitialized
  struct member was passed as an argument to its base class. This potentially
  had no observable effects, but should be fixed.


v3.8.1 (2021-08-27)
-------------------

* Reduce internal priority of AQL execution. This prevents possible deadlocks
  with modification operations in a cluster and replicationFactor >= 2, and can
  also improve responsiveness under high load of AQL queries.

* Updated arangosync to 2.6.0.

* Added protocol specific metrics: histogram about request body size, total
  number of HTTP/2 connections and total number of VST connections.

* Fix a potential multi-threading issue in index creation on coordinators, when
  an agency callback was triggered at the same time the method
  `ensureIndexCoordinatorInner` was left.

* Append physical compaction of log collection to every Raft log compaction
  (BTS-542).

* Preselect "create index in background" option when creating indexes in the web
  UI. The "create index in background" option can be less intrusive because it
  allows other write operations on the collection to proceed.

* Do not block a scheduler thread on the coordinator while an index is being
  created. Instead, start a background thread for the actual index fill-up work.
  The original thread can then be relinquished until the index is completely
  filled or index creation has failed.
  The default index creation timeout on coordinators has also been extended from
  1 hour to 4 days, but it is still configurable via the startup parameter
  `--cluster.index-create-timeout` in case this is necessary.

* Fixed: getResponsibleShard call on disjoint Smart Graphs if you asked for the
  responsible shard on a disjoint edge collection where the _from and _to differ
  (invalid), the server would respond with "DATASOURCE_NOT_FOUND". This is now
  fixed to "BAD_PARAMETER" to emphasize that the collection is fine but the
  input is invalid.

* Fixed: _api/transaction/begin called on edge collections of disjoint
  SmartGraphs falsely returned CollectionNotFound errors.

* Bug-Fix: In more complex queries there was a code-path where a (Disjoint-)
  Smart graph access was not properly optimized.

* Fix wrong assertion in fuerte and move it to where the TLA+ model says i
  should be. This fixes a unit test failure occurring on newer Macs with a
  certain clang version.

* When creating Pregel memory-mapped files, create them with O_TMPFILE attribute
  on Linux so that files are guaranteed to vanish even if a process dies.

* Improve log messages for Pregel runs by giving them more context.

* Fixed issue BTS-536 "Upgrading without rest-server is aborted by error".
  Now stating `--server.rest-server false` does not require the additional
  `--console` argument for upgrading a server.

* Fixed issue #14592:  IS_NULL(@x) isn't recognized as a constant expression.

* Fixed issue BTS-539 "Unsynchronized query kill while it's being finalized in
  another thread was uncovered through `test-kill.js` of `communication_ssl`
  suite". Fixed possible (but unlikely) crash when killing an AQL query.

* Fixed various problems in GEO_INTERSECTS: wrong results, not implemented cases
  and numerically unstable behaviour. In particular, the case of the
  intersection of two polygons in which one is an S2LngLatRect is fixed
  (BTS-475).

* Fixed ES-867 and ES-922: removed eslint from NPM packages descriptions and
  updated netmask package to non-vulnerable version.

* Web UI: Fixes the loading of map tiles which are being used to display the
  query output based on a world map when using SSL encryption. This lead to not
  displaying some world map tiles correctly (OASIS-590).

* Timely update of database server list on health check fixes BTS-505.

* Updated JavaScript dependencies, including breaking changes to non-public
  modules. We recommend always bundling your own copy of third-party modules,
  even ones listed as public.

  - accepts: 1.3.5 -> 1.3.7
  - ansi_up: 4.0.3 -> 5.0.1
  - content-type: (added) -> 1.0.4
  - error-stack-parser: 2.0.2 -> 2.0.6
  - highlight.js: 9.15.6 -> 10.7.3
  - http-errors: 1.7.2 -> 1.8.0
  - iconv-lite: 0.4.24 -> 0.6.3
  - js-yaml: 3.13.1 -> 3.14.1
  - lodash: 4.17.13 -> 4.17.21
  - marked: 0.6.2 -> removed
  - mime-types: 2.1.22 -> 2.1.31
  - mocha: 6.1.3 -> 6.2.3
  - netmask: 1.0.6 -> 2.0.2
  - qs: 6.7.0 -> 6.10.1
  - range-parser: 1.2.0 -> 1.2.1
  - semver: 6.0.0 -> 7.3.5
  - sinon: 1.17.6 -> 1.17.7
  - timezone: 1.0.22 -> 1.0.23
  - type-is: 1.6.16 -> 1.6.18
  - underscore: 1.9.1 -> 1.13.1
  - xmldom: 0.1.27 -> 0.6.0

* Updated ArangoDB Starter to 0.15.1.

* Fix BTS-453: Download of a HotBackup from remote source doesn't work on macOS.

* Web UI: Fixes a logical error which occured after re-visiting the logs view
  which lead to not displaying the logs view and its entries correctly
  (BTS-507).

* Raised the versions of the node modules `node-sass` and `sass-loader` to be
  able to build the Web UI with Node v16+.

* Suppress repeated warnings when settings LDAP options which turn out to be
  unsupported on the target system. This avoids logging the same warnings
  repeatedly.

* Make `--javascript.copy-installation` also copy the `node_modules` sub
  directory. This is required so we have a full copy of the JavaScript
  dependencies and not one that excludes some infrequently changed modules.
  In addition, file copying now intentionally excludes .map files as they are
  not needed.

* Fixed a bug, where Coordinators handled plan changes for databases in
  heartbeat thread in wrong order. Databases could be listed, but not used.

* Include K_SHORTEST_PATHS and SHORTEST_PATH execution nodes in AQL query memory
  usage accounting. The memory used by these execution node types was previously
  not tracked against the configured query memory limit (BTS-411).

* Lower log level to warning, when take over shard leadership finds an agency
  Current entry is missing the server taking over.

* Fixed BTS-408: treat positive or negative signed numbers as constants
  immediately during AQL query parsing.
  Previously, a value of `-1` was parsed initially as `unary minus(value(1))`,
  which was not treated in the same way as a constant value `value(-1)`.
  The value was later optimized to just `value(-1)`, but this only happened
  during constant-folding after parsing. Any operations that referred to the
  unfolded values during parsing thus did not treat such values as constants.

* Fix startup issues with encryption-at-rest enabled when there were empty (0
  byte) RocksDB WAL files present. Such empty files caused RocksDB to abort the
  startup, reporting corruption. However, empty WAL files are possible in case
  of server crashes etc. Now, if a WAL file is completely empty, there will be
  no attempt to read the encryption meta data from it, so the startup succeeds
	(BTS-392).

* Fixes a bug in the maintenance's error-handling code. A shard error would
  result in log messages like
    ```
    WARNING [ceb1a] {maintenance} caught exception in Maintenance shards error
    reporting: Expecting Object
    ERROR [c9a75] {maintenance} Error reporting in current: Expecting Object
    ```
  and also prevent the maintenance from reporting the current state to the
  agency, which in turn can prevent cluster-wide progress of various actions.

* APM-107: Added metric "rocksdb_read_only" to determine whether RocksDB is
  currently in read-only mode due to a background error. The metric will have a
  value of "1" if RocksDB is in read-only mode and "0" if RocksDB is in normal
  operations mode. If the metric value is "1" it means all writes into RocksDB
  will fail, so inspecting the logfiles and acting on the actual error situation
  is required.

* Fix numeric overflow in AQL WINDOW node cost estimation if the number of
  preceding rows was set to `unbounded`.

* Added a retry loop for arangorestore during the initial connection phase. The
  number of retries defaults to 3 and can be configured using
  --initial-connect-retries (BTS-491).

* Add following term ids, which prevents old synchronous replication requests
  to be accepted after a follower was dropped and has gotten in sync again.
  This makes the chaos tests which delay synchronous replication requests more
  reliable and prevent inconsistent shard replicas under bad network conditions.

* Enable process metrics on agent instances by default. Previously, some metrics
  (including the metrics starting with `arangodb_process` prefix) were not
  returned by agent instances.

* Add prefix parameter to LEVENSHTEIN_MATCH function in ArangoSearch
  (DEVSUPP-753).

* Bug-fix: Pregel WCC algorithm could yield incorrect results if a part of the
  connected component was only attached via OUTBOUND edges.
  The underlying algorithm is now modified to properly retain INBOUND edges for
  the runtime of the execution. This uses more RAM for the algorithm but
  guarantees correctness.

* Fix serialization of query shutdown error code when sending it to DB servers.
  The error code is numeric, but it was sent to the `/_api/aql/finish` API as a
  string. This led to the DB servers always assuming the default error code
  TRI_ERROR_INTERAL (error code 4). This was not a problem for normal query
  operations, but it could have led to warnings being logged stating "please
  contact ArangoDB support". Now the actual error code is using during query
  shutdown.

* Fix display of running and slow queries in web UI when there are multiple
  coordinators. Previously, the display order of queries was undefined, which
  could lead to queries from one coordinator being display on top once and then
  the queries from another. That made using this UI harder than necessary.

  Now queries are sorted for display, according to their query IDs.

* Fixed an issue in index selection, when the selectivty estimate of another
  prefix index was used without checking if the other index covered the FILTER
  condition.

  For example, given the following indexes:

  - index 1: ["e", "a", "b", "c"]
  - index 2: ["e", "a", "b"]
  - index 3: ["d", "e", "f", "g"]

  and the FILTER condition `d == 1 && e == 2 && f == 3`, then the best index to
  pick would be index 3. However, the optimizer may have picked index 1 here.
  All indexes are valid candidates for this FILTER condition, but none of the
  indexes covered all attributes of the FILTER condition. So the index
  selectivity estimates were (correctly) not used directly to determine the best
  index.
  The actual bug happened when comparing the usefulness of the candidate
  indexes, when figuring out that even though the selectivity estimate for index
  1 could not be used, but that there existed a prefix index of index 1 (index
  2). The selecivity estimate of this index was taken _without_ checking that
  prefix index actually satisfied the FILTER condition fully.
  The prefix index' selectivity estimate must only be used if it fully satisfies
  the FILTER condition, which was not the case here.

* Fixed DEVSUP-799: unique vertex getter may point to invalid memory after being
  resetted, resulting in undefined behavior for traversals returning unique
  vertices from inner FOR loops.

* Improve usability of hidden options: `--help` mentions that these exist and
  how to display them.

* Fixed ES-863: reloading of users within the Cluster.
  If a Coordinator is asked to reload its users (e.g. by the UserManager in
  Foxx, it is also possible to do via API, but this is internal and on purpose
  not documented, so unlikely that it is used), in concurrency with user
  management updates there is a chance that the reload is not correctly
  performed on this coordinator. It may have missed the last update locally,
  causing one user to have an older state. It will be fixed on the next
  modification of any other users/permissions. Unfortunately this bug can
  cascade and when hit again, the coordinator can now be off by two updates.
  In DC2DC this situation is more likely to happen on the target datacenter,
  causing this datacenter to have other users/permissions than the source one.

* Fix BTS-446: When finding a not yet fully initialized agency, do not
  immediately fatal exit. Keep trying for (very generous) 5 minutes.

* Only build actually used subattributes of traversal paths, i.e. "vertices",
  "edges" or "weights". If any of the paths subcomponents is not used, the
  optimizer will try to save these components from being built for each result
  item.

* Backport bugfix from upstream rocksdb repository for calculating the free disk
  space for the database directory. Before the bugfix, rocksdb could
  overestimate the amount of free space when the arangod process was run as
  non-privileged users.

* Fixed a problem with active failover, where a failover could take 5 mins
  because the follower was caught in a bad state during replication. This fixes
  BTS-425.

* Add soft coordinator shutdown: This is a new option `soft=true` for the
  DELETE /_admin/shutdown API. Has only meaning for coordinators, otherwise
  ignored. A number of things are allowed to finish but no new things are
  allowed when in soft coordinator shutdown:
    - AQL cursors
    - transactions
    - asynchronous operations
    - Pregel runs
  Once all of the ongoing operations of these have finished and all requests on
  the low priority queue have been executed, the coordinator shuts down the
  normal way. This is supposed to make a coordinator restart less intrusive for
  clients.

* Fix BTS-398: Cannot force index hint for primary index if FILTER has multiple
  OR conditions that require different indexes.

* Bug-fix (macOs): in macOs there is an upper bound for descriptors defined by
  the system, which is independend of the settings in `ulimit -n`. If the hard
  limit is set above this upper bound value ArangoDB tries to raise the soft
  limit to the hard limit on boot. This will fail due to the system limit. This
  could cause ArangoDB to not start, asking you to lower the minimum of required
  file descriptors. The system set upper bound is now honored and the soft limit
  will be set to either hard limit or system limit whichever is lower.

* Implemented APM-86: add query option `fillBlockCache` to control population of
  RocksDB block cache with data read by the query. The default value for this
  per-query option is `true`, which mimics the previous behavior.
  Setting the option to off allows not storing data in RocksDB's block cache for
  queries that are known to read only semi-relevant or unimportant data.


v3.8.0 (2021-07-14)
-------------------

* Always remove blocker object for revision trees in case of replication
  failures.

* Fix invalid assertion for insert/removal buffers positioning and internals of
  `hasBlockerUpTo` function.

* Updated ArangoDB Starter to 0.15.0-1.

* Updated arangosync to 2.4.0.

* For cluster AQL queries, let the coordinator determine the query id to be used
  on DB servers. This allows the coordinator to roll back AQL query setup
  requests via the query id. Previously, the DB servers each generated a local
  query id and returned it to the coordinator, who would then keep track of them
  for later use. The problem with this was that if an AQL query setup request
  timed out, the coordinator had no way to roll it back.

  In addition, if setting up a query takes long on a DB server so that the
  coordinator sends a rollback request, there are some measures in place for the
  unlikely case in which the rollback request overtakes the setup request. In
  this case, the rollback request will not find a query yet, but will register a
  tombstone for it. Once the query gets registered by the delayed request, it
  will (correctly) fail because of the tombstone.

* Removed a special case for empty document update operations (i.e. update
  requests in which no attributes were specified to be updated) were handled in
  a special way without performing any writes. The problem was that such updates
  did not update the local state, but could have been replicated to followers.
  This special empty update case is now removed and update operations that do
  not update any attributes are treated as normal write operations both locally
  and in the replication.

* Fix partial cleanup of internal write batches for multi-document operations of
  which one or multiple failed. The previous implementation had an unreleased
  performance optimization that wouldn't clean up the write batch completely.
  That could have led to a wrong sequence of events being accumulated in the
  write batch, which may have confused the WAL tailing API later. This bug was
  only present in 3.8 RCs.

* Fix some occurrences in which Merkle trees could silently apply the same
  change multiple times, which led to data drift between the Merkle tree and the
  underlying collection's data. This bug was only present in 3.8 but no earlier
  versions.

* On a failure during synchronous replication, do not remove the failed follower
  from the list of known servers in the transaction.
  If we do, we would not be able to send the commit/abort to the follower later.
  However, we still need to send the commit/abort to the follower at transaction
  end, because the follower may be responsible  for _other_ shards as well.

  This change also removes dangling transactions that could stay around on
  followers until they expired after the transaction idle timeout (180 seconds),
  and that could prevent a follower from getting back in sync during this
  period.

* Added more context to "dropping follower" messages, so it is easier to analyze
  what exactly went wrong.

* Fixed invalid shard synchronization for documents not added via INSERT with
  `overwriteMode` set to `ignore`. In this case, if a document with the given
  key already exists, it is not changed on the leader (i.e. no write happens on
  the leader). However, a write was replicated to the follower, which was wrong.
  This write is now suppressed, which can only make such insert operations
  faster.

* Fix DEVSUP-753: now it is safe to call visit on exhausted disjunction
  iterator.

* Slightly improve specific warning messages for better readability.

* Fix URL request parsing in case data is handed in in small chunks.
  Previously the URL could be cut off if the chunk size was smaller than the URL
  size.

* Fix BTS-430: Added missing explain output about indexes for SHORTEST_PATH,
  K_SHORTEST_PATHS and K_PATHS.

* Added check to utils/generateAllMetricsDocumentation.py to check that the file
  name and the value of the name attribute are the same in the metrics
  documentation snippets. Correct a few such names.

v3.8.0-rc.2 (2021-06-07)
------------------------

* Updated arangosync to 2.3.0.

* Fix BTS-456, BTS-457: Make geo intersection between point and rectangle
  symmetrical.

* Fix BTS-450: RandomGenerator caught assertion during a value generation within
  `dump_maskings` testsuite. Ensure correct conversion between 64 and 32bit.

* Added check for data type compatibility between members of pipeline
  ArangoSearch analyzer.


v3.8.0-rc.1 (2021-05-26)
------------------------

* Fix BTS-442: a query with fullCount on a sharded collection hangs indefinitely
  when LIMIT is less then number of available documents.

* Removed unused documentation snippets (non-Rest DocuBlocks) as well as the
  documentation about the long deprecated features Simple Queries and
  JavaScript-based graph traversal. Also removed the descriptions of the JS API
  methods `collection.range()`, `collection.closedRange()`,
  `cursor.setBatchSize()` and `cursor.getBatchSize()`. All the functionality is
  superseded by AQL.

* Fixed ES-881: ensure that LDAP options for async, referrals and restart set
  the off value correctly. Otherwise, this can result in an "operations error".

* Improve Merkle tree memory usage and allow left-growth of trees, too. This can
  help with insertions of arbitrarily old data.

* Added metric `arangodb_sync_rebuilds_total` to track the full rebuild of a
  shard follower after too many subsequent shard synchronization failures. This
  metric should always have a value of 0. Everything else indicates a serious
  problem.

* Fixed BTS-422: SingleRemoteModification in AQL behaves different.

  This disables the optimizer rule `optimize-cluster-single-document-operations`
  for array inputs, e.g.

      INSERT [...] INTO collection
      REMOVE [...] IN collection

  For the cases, the optimization is not pulled off, and the normal insert/
  update/replace/remove behavior is executed, which will fail because of an
  array being used as input.

* Fix BTS-409: return error 1948 when a negative edge was detected during or was
  used as default weight in a SHORTEST_PATH or a K_SHOTRTEST_PAHS traversal.

* Fixed issue BTS-424: fix invalid input row handling in WINDOW execution.

* Added 2 options to allow HTTP redirection customization for root ("/") call of
  HTTP API:

  `--http.permanently-redirect-root`: if true (default), use a permanent
  redirection (use HTTP 301 code), if false fall back to temporary redirection
  (use HTTP 302 code);
  `--http.redirect-root-to`: redirect of root URL to a specified path (redirects
  to "/_admin/aardvark/index.html" if not set (default)).

* Fixed DEVSUP-764 (SEARCH-7): inconsistent BM25 scoring for LEVENSHTEIN_MATCH
  function.

* Rename two metrics with previously Prometheus-incompatible names:
  - `arangodb_aql_global_query_memory_limit_reached` was renamed to
    `arangodb_aql_global_query_memory_limit_reached_total`
  - `arangodb_aql_local_query_memory_limit_reached` was renamed to
    `arangodb_aql_local_query_memory_limit_reached_total`

  These metrics were introduced in 3.8 so there is no migration for these
  metrics.

* Return error 1948 when a negative edge was detected during a weighted
  traversal or was used as default weight.

* Fixes BTS-417. In some cases an index did not consider both bounds (lower and
  upper) for a close range scan if both bounds are expressed using the same
  operator, e.g., `FILTER doc.beginDate >= lb AND ub >= doc.beginDate`.

* Fix various issues related to the new WINDOW operation (see BTS-402)
  - Improved explain output for ISO 8601 duration strings and fixed missing week
    component.
  - Improved validation of input data and error messages.
  - Prevent FILTERs from being moved beyond a WINDOW.

* Fixes BTS-416. During shutdown, a Shard leader wrongly reported that it could
  not drop a shard follower instead of correctly indicating the shutdown as the
  reason.

* Fixes pregel lifetime management. Previously shutting down the server while a
  pregel job was still running could result in a segfault or a shutdown hanger.

* Improve error reporting for Merkle tree operations and improve memory usage
  for unused trees by hibernating them. In addition, add some backoff to shard
  synchronization in case there are repeated sync failures for the same shard.

* Fixed BTS-403: Hot restores must also clear relevant `Current` keys. The
  overriding of the `Plan` entries needs to be reflected in `Current` to avoid
  conflicts in maintenance jobs.

* Log a proper message if an unexpected state is encountered when taking over
  shard leadership. In addition, make the change to the internal followerinfo
  state atomic so that it cannot be semi-changed.

* Fixed two bugs in fuerte with HTTP/2 and VST connections.
  One could lead to ordered timeouts not being honoured. The other could lead to
  an ordered callback be called multiple times.

* Improve "Shards" view in web UI so that the shards of individual collections
  can be expanded and collapsed without affecting the display of any other
  shards. Also added a "Toggle all" button the web UI to expand/collapse the
  shards for all collections.

* Improve exception safety for maintenance thread and shard unlock operations.

* Fixed issue #14122: when the optimizer rule "inline-subqueries" is applied, it
  may rename some variables in the query. The variable renaming was however not
  carried out for traversal PRUNE conditions, so the PRUNE conditions could
  still refer to obsolete variables, which would make the query fail with errors
  such as

    Query: AQL: missing variable ... for node ... while planning registers

* Improve performance of batch CRUD operations (insert, update, replace, remove)
  if some of the documents in the batch run into write-write conflicts.
  Rolling back partial operations in case of a failure is very expensive because
  it requires rebuilding RocksDB write batches for the transaction from scratch.
  Rebuilding write batches takes time proportional to the number of operations
  in the batch, and for larger batches the cost can be prohibitive.
  Now we are not rolling back write batches in some situations when this is not
  required, so that in many cases running into a conflict does not have that
  high overhead. There can still be issues when conflicts happen for index
  entries, but a lot of previously problematic cases should now work better.

  This change also reduces the RocksDB-internal lock timeout for writing to keys
  locked by another transaction from 1s to 1ms. This will mean that operations
  that ran into a write-write conflict may fail quicker than before, and not
  wait and retry to acquire the locked key(s).

* Fix response when isBuilding could not be removed from newly created
  collection, when agency precondition fails. This can happen, when own rebootId
  increment has triggered plan entry to be removed.

* Fixed issue BTS-354: Assertion related to getCollection.

* Fix DEVSUP-749: Fix potential deadlock when executing concurrent view/link DDL
  operations and index DDL operations on the same collection.

* When writing to starting shard leader respond with specific 503.
  Fixes BTS-390.

* Fixed a use after free bug in the connection pool.

* Show peak memory usage in AQL query profiling output.

* Fixed various issues (mainly data races) reported by ThreadSanitizer.

* Fixed bug in error reporting when a database create did not work, which lead
  to a busy loop reporting this error to the agency.

* Fixed the error response if the HTTP version is not 1.0 or 1.1 and if the
  Content-Length is too large (> 1 GB).

* Guarded access only to ActionBase::_result.

* Updated arangosync to 2.2.0.

* Fixed proper return value in sendRequestRetry if server is shutting down.

* Fixed internal issue #798: In  rare case when remove request completely cleans
  just consolidated segment commit could be cancelled and documents removed from
  collection may be left dangling in the ArangoSearch index.
  Also fixes ES-810 and BTS-279.

* Fixed a small problem in fuerte which could lead to an assertion failure.

* Retry if an ex-leader can no longer drop a follower because it is no longer
  leading.

* Fixed issue BTS-373: ASan detected possible heap-buffer-overflow at
  arangodb::transaction::V8Context::exitV8Context().

* Fix a potential buffer overflow in RestReplicationHandler.

* Make the time-to-live (TTL) value of a streaming cursor only count after the
  response has been sent to the client.

* Added option `--query-max-runtime` to arangoexport, in order to control
  maximum query runtime.

* Fix BTS-340: AQL expressions similar to `x < 3 || x` are no longer erroneously
  be reduced to `x < 3` by the optimizer rule remove-redundant-or.

* Change arangosh client behavior:
  - *_RAW methods will never add a `body` to HEAD responses
  - *_RAW methods will now always return velocypack-typed responses in Buffers
  - `--server.force-json` will now be applied as default, overrideable
    by user code


v3.8.0-beta.1 (2021-04-20)
--------------------------

* Fix BTS-374: thread race between ArangoSearch link unloading and storage
  engine WAL flushing.

* Improve parallelism capabilities of arangorestore.

  arangorestore can now dispatch restoring data chunks of a collection to idle
  background threads, so that multiple restore requests can be in flight for the
  same collection concurrently.

  This can improve restore speed in situations when there are idle threads left
  (number of threads can be configured via arangorestore's `--threads` option)
  and the dump file for the collection is large.

  The improved parallelism is only used when restoring dumps that are in the
  non-enveloped format. This format has been introduced with ArangoDB 3.8.
  The reason is that dumps in the non-enveloped format only contain the raw
  documents, which can be restored independent of each other, i.e. in any order.
  However, the enveloped format may contain documents and remove operations,
  which need to be restored in the original order.

* Fix crashes during arangorestore operations due to usage of wrong pointer
  value for updating user permissions.

* Fixed BTS-360 and ES-826: sporadic ArangoSearch error `Invalid RL encoding in
  'dense_fixed_offset_column_key'`.

* Add HTTP REST API endpoint POST `/_api/cursor/<cursor-id>` as a drop-in
  replacement for PUT `/_api/cursor/<cursor-id>`. The POST API is functionally
  equivalent to the existing PUT API. The benefit of using the POST API is that
  HTTP POST requests will not be considered as idempotent, so proxies may not
  retry them if they fail. This was the case with the existing PUT API, as HTTP
  PUT requests can be considered idempotent according to the HTTP specification.

  The POST API is not used internally by ArangoDB's own requests in this
  version. This means that compatibility to older versions of ArangoDB that do
  not provide the new API is ensured.

* Timely updates of rebootId / cluster membership of DB servers and coordinators
  in ClusterInfo. Fixes BTS-368 detected in chaos tests.

* Fix cluster internal retry behavior for network communications. In particular
  retry on 421 (leader refuses operation). This leads to the cluster letting
  less internal errors out to clients.

* Fixed CPPCHECK warning or added suppression.

* Web UI - Added missing HTML escaping inside the file upload plugin used in the
  section of deploying a new Foxx application when uploading a zip file.

* Allow to specify a fail-over LDAP server. Instead of "--ldap.OPTION" you need
  to specify "--ldap2.OPTION". Authentication / Authorization will first check
  the primary LDAP server. If this server cannot authenticate a user, it will
  try the secondary one. It is possible to specify a file containing all users
  that the primary (or secondary) LDAP server is handling by specifying the
  option "--ldap.responsible-for". This file must contain the usernames
  line-by-line.

* Fix BTS-352: removed assertion for success of a RocksDB function and throw a
  proper exception instead.

* Added option `--query.require-with` to make AQL in single server mode also
  require `WITH` clauses where the cluster would need them.
  The option is turned off by default, but can be turned on in single servers to
  remove this behavior difference between single servers and clusters, making
  later a transition from single server to cluster easier.

* Fixed a problem in document batch operations, where errors from one shard were
  reported multiple times, if the shard is completely off line.

* Fixed issue #13169: arangoimport tsv conversion of bools and null, although
  switched off by `--convert false`.

  Importing unquoted `null`, `false` and `true` literals from delimited files
  get imported as strings now if `convert` is explicitly turned off. It
  previously affected unquoted numbers only.

* Web UI: Highlight binary and hexadecimal integer literals in AQL queries.

* Prevent arangod from terminating with "terminate called without an active
  exception" (SIGABRT) in case an out-of-memory exception occurs during creating
  an ASIO socket connection.

* Micro improvements for Pregel job API and documentation:
  - Added a few useful attributes to Pregel HTTP API docs.
  - Added "parallelism" attribute to the result of Pregel job status responses,
    so that the effective parallelism is reported back.
  - Make sure "computationTime" in Pregel job status response does not underflow
    in case of errors.

* Fix BTS-350, BTS-358: Fixed potential startup errors due to global replication
  applier being started before end of database recovery procedure.
  Also fixed potential shutdown errors due to global replication applier being
  shut down in parallel to a concurrent shut down attempt.

* Fix BTS-357: Fix processing of analyzer with return type by TOKENS function.

* Fix BTS-346: Improved handling of AQL query kill command in unlikely places,
  before the query starts to execute and after the query is done but the result
  is still being written. Now the cleanup of queries works more reliably. This
  unreliable kill time windows were very short and unlikely to hit, although if
  one was hit transactions were not aborted, and collection locks could be
  lingering until query timeout.

* Updated ArangoDB Starter to 0.15.0.

* Fix BTS-357: Fix processing of analyzer with return type by TOKENS function.

* Added error handling for figures command in cluster. Previously errors
  returned by shards were ignored when aggregating the individual responses.

* Remove CMake control variable `UNCONDITIONALLY_BUILD_LOG_MESSAGES`.

* Fix undefined behavior in dynarray constructor when running into
  an out-of-memory exception during construction. In arangod, this can only
  happen during metrics objects construction at program start.

* Added option `--headers-file` to arangoimport, to optionally read CSV/TSV
  headers from a separate file.

* Fixed issue BTS-353: memleak when running into an out-of-memory situation
  while repurposing an existing AqlItemBlock.

* Change metrics' internal `low()` and `high()` methods so that they  return by
  value, not by reference.

* Fix logging of urls when using `--log.level requests=debug`. There was an
  issue since v3.7.7 with the wrong URL being logged in request logging if
  multiple requests were sent over the same connection. In this case, the
  request logging only reported the first URL requested in the connection, even
  for all subsequent requests.

* Added startup option `--query.allow-collections-in-expressions` to control
  whether collection names can be used in arbitrary places in AQL expressions,
  e.g. `collection + 1`. This was allowed before, as a collection can be seen as
  an array of documents. However, referring to a collection like this in a query
  would materialize all the collection's documents in RAM, making such
  constructs prohibitively expensive for medium-size to large-size collections.

  The option can now be set to `false` to prohibit accidental usage of
  collection names in AQL expressions. With that setting, using a collection
  inside an arbitrary expression will trigger the error `collection used as
  expression operand` and make the query fail.
  Even with the option being set to `false`, it is still possible to use
  collection names in AQL queries where they are expected, e.g. `FOR doc IN
  collection RETURN doc`.

  The option `--query.allow-collections-in-expressions` is introduced with a
  default value of `true` in 3.8 to ensure downwards-compatibility, but the
  default value will change to `false` in 3.9. Furthermore, the option will be
  deprecated in 3.9 and removed in later versions, in addition to making
  unintended usage of collection names always an error.

* Deprecate option `--rocksdb.exclusive-writes`, which was meant to serve only
  as a stopgap measure while porting applications from the MMFiles storage
  engine to RocksDB.

* Fix errors caused by creating some log messages in log level DEBUG in log
  topics PREGEL and GRAPHS. Setting the log level to DEBUG for these topics
  could lead to errors when running some Pregel jobs or SmartGraph traversals.


v3.8.0-alpha.1 (2021-03-29)
---------------------------

* Updated ArangoDB Starter to 0.15.0-preview-2.

* Updated OpenSSL to 1.1.1k and OpenLDAP to 2.4.58.

* Updated arangosync to 2.0.1.

* Fix connectionTime statistic. This statistic should provide the distribution
  of the connection lifetimes, but in previous versions the tracking was broken
  and no values were reported.

* When using connections to multiple endpoints and switching between them,
  arangosh can now reuse existing connections by referring to an internal
  connection cache. This helps for arangosh scripts that repeatedly connect to
  multiple endpoints, and avoids wasting lots of ephemeral TCP ports remaining
  in CLOSE_WAIT state.
  This change is transparent to any arangosh scripts or commands that do not
  reconnect to other endpoints than the one specified at arangosh start.

* Add an option for locking down all endpoints in the `/_admin/cluster` REST API
  for callers without a proper JWT set in the request. There is a new startup
  option `--cluster.api-jwt-policy` that allows *additional* checks for a valid
  JWT in requests to sub-routes of `/_admin/cluster`. The possible values for
  the startup option are:

  - "jwt-all": requires a valid JWT for all accesses to `/_admin/cluster` and
    its sub-routes. If this configuration is used, the "Cluster" and "Nodes"
    sections of the web interface will be disabled, as they are relying on the
    ability to read data from several cluster APIs.
  - "jwt-write": requires a valid JWT for write accesses (all HTTP methods
    except HTTP GET) to `/_admin/cluster`. This setting can be used to allow
    privileged users to read data from the cluster APIs, but not to do any
    modifications. All existing permissions checks for the cluster API routes
    are still in effect with this setting, meaning that read operations without
    a valid JWT may still require dedicated other permissions (as in 3.7).
  - "jwt-compat": no *additional* access checks are in place for the cluster
    APIs. However, all existing permissions checks for the cluster API routes
    are still in effect with this setting, meaning that all operations may still
    require dedicated other permissions (as in 3.7).

  The default value for the option is `jwt-compat`, which means this option will
  not cause any extra JWT checks compared to 3.7.

* UI builds are now using the yarn package manager instead of the previously
  used node package manager.

* Fix shortName labels in metrics, in particular for agents.

* The old metrics api contains the following gauges which should actually be
  counters:
    - arangodb_scheduler_jobs_dequeued
    - arangodb_scheduler_jobs_submitted
    - arangodb_scheduler_jobs_done
  Therefore the new v2 metric api adds the following counters:
    - arangodb_scheduler_jobs_dequeued_total
    - arangodb_scheduler_jobs_submitted_total
    - arangodb_scheduler_jobs_done_total
  These counters are only visible in the new v2 metric and replace the old
  metrics which are suppressed for v2.

* Fix implicit capture of views in a context of JS transaction.

* Introduce metrics for AQL query memory limit violations:
  - `arangodb_aql_global_query_memory_limit_reached`: Total number of times the
    global query memory limit was violated.
  - `arangodb_aql_local_query_memory_limit_reached`: Total number of times a
    local query memory limit was violated.

* Set the default value for `--query.global-memory-limit` to around 90% of RAM,
  so that a global memory limit is now effective by default.

  The default global memory limit value is calculated by a formula depending on
  the amount of available RAM and will result in the following values for
  common RAM sizes:

  RAM:            0      (0MiB)  Limit:            0   unlimited, %mem:  n/a
  RAM:    134217728    (128MiB)  Limit:     33554432     (32MiB), %mem: 25.0
  RAM:    268435456    (256MiB)  Limit:     67108864     (64MiB), %mem: 25.0
  RAM:    536870912    (512MiB)  Limit:    255013683    (243MiB), %mem: 47.5
  RAM:    805306368    (768MiB)  Limit:    510027366    (486MiB), %mem: 63.3
  RAM:   1073741824   (1024MiB)  Limit:    765041049    (729MiB), %mem: 71.2
  RAM:   2147483648   (2048MiB)  Limit:   1785095782   (1702MiB), %mem: 83.1
  RAM:   4294967296   (4096MiB)  Limit:   3825205248   (3648MiB), %mem: 89.0
  RAM:   8589934592   (8192MiB)  Limit:   7752415969   (7393MiB), %mem: 90.2
  RAM:  17179869184  (16384MiB)  Limit:  15504831938  (14786MiB), %mem: 90.2
  RAM:  25769803776  (24576MiB)  Limit:  23257247908  (22179MiB), %mem: 90.2
  RAM:  34359738368  (32768MiB)  Limit:  31009663877  (29573MiB), %mem: 90.2
  RAM:  42949672960  (40960MiB)  Limit:  38762079846  (36966MiB), %mem: 90.2
  RAM:  68719476736  (65536MiB)  Limit:  62019327755  (59146MiB), %mem: 90.2
  RAM: 103079215104  (98304MiB)  Limit:  93028991631  (88719MiB), %mem: 90.2
  RAM: 137438953472 (131072MiB)  Limit: 124038655509 (118292MiB), %mem: 90.2
  RAM: 274877906944 (262144MiB)  Limit: 248077311017 (236584MiB), %mem: 90.2
  RAM: 549755813888 (524288MiB)  Limit: 496154622034 (473169MiB), %mem: 90.2

* Increase default idle timeout in streaming transactions from 10 seconds to
  60 seconds, and make the timeout configurable via a startup parameter
  `--transaction.streaming-idle-timeout`.

* Use RebootTracker to abort cluster transactions on DB servers should the
  originating coordinator die or be rebooted. The previous implementation left
  the coordinator's transactions open on DB servers until they timed out there.
  Now, the coordinator's unavailability or reboot will be detected as early as
  it is reported by the agency, and all open transactions from that coordinator
  will be auto-aborted on DB servers.

* Update the Web UI's list of built-in AQL functions for proper syntax
  highlighting in the query editor.

* Fix a crash caused by returning a result produced by ANALYZER function.

* Fix a race in LogAppender::haveAppenders.
  `haveAppenders` is called as part of audit logging. It accesses internal maps
  but previously did not hold a lock while doing so.

* Bug-fix in the case of very rare network issues there was a chance that an AQL
  query could get stuck during a cleanup and after a commit.
  This would cause the client to receive a timeout, and the Coordinator blocking
  a Scheduler thread. This situation is sorted out and the thread will not be
  blocked anymore. We also added logs in case the query could not successfully
  be cleaned up, which would leave locks on shards behind.

* Fix an assertion failure that occurred when restoring view definitions from a
  cluster into a single server.

* Added new ArangoSearch analyzer type "stopwords".

* Fix error message in case of index unique constraint violations. They were
  lacking the actual error message (i.e. "unique constraint violated") and only
  showed the index details. The issue was introduced only in devel in Feb.

* Removed old metrics in new v2 metric api. Those metric endpoints were
  identical to the sum value of histograms.

* Allow process-specific logfile names.

  This change allows replacing '$PID' with the current process id in the
  `--log.output` and `--audit.output` startup parameters.
  This way it is easier to write process-specific logfiles.

* Backport a bugfix from upstream RocksDB for opening encrypted files with small
  sizes. Without the bugfix, the server may run into assertion failures during
  recovery.

* Fix duplicate leaving of V8 contexts when returning streaming cursors.
  The `exitContext` call done on query shutdown could previously try to exit the
  V8 context multiple times, which would cause undefined behavior. Now we are
  tracking if we already left the context to prevent duplicate invocation.

* In a cluster, do not create the collections `_statistics`, `_statistics15` and
  `statisticsRaw` on DB servers. These collections should only be created by the
  coordinator, and should translate into 2 shards each on DB servers. But there
  shouldn't be shards named `_statistics*` on DB servers.

* Fixed two bogus messages about hotbackup restore:
  - Coordinators unconditionally logged the message "Got a hotbackup restore
    event, getting new cluster-wide unique IDs..." on shutdown. This was not
    necessarily related to a hotbackup restore.
  - DB servers unconditionally logged the message "Strange, we could not
    unregister the hotbackup restore callback." on shutdown, although this was
    meaningless.

* Rename "save" return attribute to "dst" in AQL functions `DATE_UTCTOLOCAL` and
  `DATE_LOCALTOUTC`.

* Fix potentially undefined behavior when creating a
  CalculationTransactionContext for an arangosearch analyzer. An uninitialized
  struct member was passed as an argument to its base class. This potentially
  had no observable effects, but should be fixed.

* Retry a cluster internal network request if the connection comes from the pool
  and turns out to be stale (connection immediately closed). This fixes some
  spurious errors after a hotbackup restore.

* Fix progress reporting for arangoimport with large files on Windows.
  Previously, progress was only reported for the first 2GB of data due to an int
  overflow.

* Log the actual signal instead of "control-c" and also include the process id
  of the process that sent the signal.

* Fixed Github issue #13665: Improve index selection when there are multiple
  candidate indexes.

* When dropping a collection or an index with a larger amount of documents, the
  key range for the collection/index in RocksDB gets compacted. Previously, the
  compaction was running in foreground and thus would block the deletion
  operations.
  Now, the compaction is running in background, so that the deletion operations
  can return earlier.
  The maximum number of compaction jobs that are executed in background can be
  configured using the new startup parameter
  `--rocksdb.max-parallel-compactions`, which defaults to 2.

* Put Sync/LatestID into hotbackup and restore it on hotbackup restore if it is
  in the backup. This helps with unique key generation after a hotbackup is
  restored to a young cluster.

* Fixed a bug in the index count optimization that doubled counted documents
  when using array expansions in the fields definition.

* Don't store selectivity estimate values for newly created system collections.

  Not storing the estimates has a benefit especially for the `_statistics`
  system collections, which are written to periodically even on otherwise idle
  servers. In this particular case, the actual statistics data was way smaller
  than the writes caused by the index estimate values, causing a disproportional
  overhead just for maintaining the selectivity estimates.
  The change now turns off the selectivity estimates for indexes in all newly
  created system collections, and for new user-defined indexes of type
  "persistent", "hash" or "skiplist", there is now an attribute "estimates"
  which can be set to `false` to disable the selectivity estimates for the
  index.
  The attribute is optional. Not setting it will lead to the index being created
  with selectivity estimates, so this is a downwards-compatible change for
  user-defined indexes.

* Added startup option `--query.global-memory-limit` to set a limit on the
  combined estimated memory usage of all AQL queries (in bytes).
  If this option has a value of `0`, then no memory limit is in place.
  This is also the default value and the same behavior as in previous versions
  of ArangoDB.
  Setting the option to a value greater than zero will mean that the total
  memory usage of all AQL queries will be limited approximately to the
  configured value.
  The limit is enforced by each server in a cluster independently, i.e. it can
  be set separately for coordinators, DB servers etc. The memory usage of a
  query that runs on multiple servers in parallel is not summed up, but tracked
  seperately on each server.
  If a memory allocation in a query would lead to the violation of the
  configured global memory limit, then the query is aborted with error code 32
  ("resource limit exceeded").
  The global memory limit is approximate, in the same fashion as the per-query
  limit provided by the option `--query.memory-limit` is. Some operations,
  namely calls to AQL functions and their intermediate results, are currently
  not properly tracked.
  If both `--query.global-memory-limit` and `--query.memory-limit` are set, the
  former must be set at least as high as the latter.

  To reduce the cost of globally tracking the memory usage of AQL queries, the
  global memory usage counter is only updated in steps of 32 kb, making this
  also the minimum granularity of the global memory usage figure.
  In the same fashion, the granularity of the peak memory usage counter inside
  each query was also adjusted to steps of 32 kb.

* Added startup option `--query.memory-limit-override` to control whether
  individual AQL queries can increase their memory limit via the `memoryLimit`
  query option. This is the default, so a query that increases its memory limit
  is allowed to use more memory.
  The new option `--query.memory-limit-override` allows turning this behavior
  off, so that individual queries can only lower their maximum allowed memory
  usage.

* Added metric `arangodb_aql_global_memory_usage` to expose the total amount of
  memory (in steps of 32 kb) that is currently in use by all AQL queries.

* Added metric `arangodb_aql_global_memory_limit` to expose the memory limit
  from startup option `--query.global-memory-limit`.

* Allow setting path to the timezone information via the `TZ_DATA` environment
  variable, in the same fashion as the currently existing `ICU_DATA` environment
  variable. The `TZ_DATA` variable is useful in environments` that start arangod
  from some unusual locations, when it can't find its `tzdata` directory
  automatically.

* Fixed a bug in query cost estimation when a NoResults node occured in a
  spliced subquery. This could lead to a server crash.

* Fix slower-than-necessary arangoimport behavior:
  arangoimport has a built-in rate limiter, which can be useful for importing
  data with a somewhat constant rate. However, it is enabled by default and
  limits imports to 1MB per second. These settings are not useful.

  This change turns the rate limiting off by default, and sets the default chunk
  size to 8MB (up from 1MB) as well. This means that arangoimport will send
  larger batches to the server by default. The already existing `--batch-size`
  option can be used to control the maximum size of each batch.

  The new parameter `--auto-rate-limit` can now be used to toggle rate limiting.
  It defaults to off, whereas previously rate limiting was enabled by default
  unless `--batch-size` was specified when arangoimport was invoked.

* The cluster dashboard charts in the web UI are now more readable during the
  initialization phase. Additionally, the amount of agents are now displayed
  there as well. An agent failure will also appear here in case it exists.

* Added more useful information during the SmartGraph creation in the web UI
  in case the current database is a OneShard database.

* Add support for building with Zen 3 CPU when optimizing for the local
  architecture.

* The web UI's node overview now displays also agent information (cluster only).

* The statistics view in the web UI does now provide more system specific
  information in case the Metrics API is enabled. Different statistics may be
  visible depending on the operating system.

* Added metrics documentation snippets and infrastructure for that.

* Added a new cluster distribution view to the web UI. The view includes general
  details about cluster-wide distribution in general as well as more detailed
  shard distribution specific information.

* Reasonably harden MoveShard against invalid VelocyPack input.

* Removed older reference to VelocyPackDumper.

* Added `--documents-per-batch` option to arangoexport.
  This option allows to control the number of documents to be returned by each
  server-side batch. It can be used to limit the number of documents per batch
  when exporting collections with large documents.

* Added a new metrics view to the web UI. This view can be used in a clustered
  environment as well as in a single instance. Metrics are displayed either in
  a tabular format or as plain text (Prometheus Text-based format).
  Additionally, the metrics can be downloaded there.

* Added a new maintenance mode tab to the web UI in cluster mode.
  The new tab shows the current state of the cluster supervision maintenance and
  allows to enable/disable the maintenance mode from there. The tab will only be
  visible in the `_system` database. The required privileges for displaying the
  maintenance mode status and/or changing it are the as for using the REST APIs
  for the maintenance mode.

* Added ability to display Coordinator and DBServer logs from inside the Web UI
  in a clustered environment when privileges are sufficient.
  Additionally, displayed log entries can now be downloaded from the web UI in
  single server and in cluster mode.

* The Web UI's info view of a collection now displays additional properties and
  statistics (e.g. RocksDB related figures, sharding information and more).

* Improve progress reporting for shard synchronization in the web UI.
  The UI will now show how many shards are actively syncing data, and will
  provide a better progress indicator, especially if there is more than one
  follower for a shard.

* Added `--shard` option to arangodump, so that dumps can be restricted to one
  or multiple shards only.

* Add optional hostname logging to log messages.
  Whether or not the hostname is added to each log message can be controlled via
  the new startup option `--log.hostname`. Its default value is the empty
  string, meaning no hostname will be added to log messages.
  Setting the option to an arbitrary string value will make this string be
  logged in front of each regular log message, and inside the `hostname`
  attribute in case of JSON-based logging. Setting the option to a value of
  `auto` will use the hostname as returned by `gethostbyname`.

* Added list-repeat AIR primitive that creates a list containing n copies of the
  input value.

* Prevent arangosh from trying to connect after every executed command.
  This fixes the case when arangosh is started with default options, but no
  server is running on localhost:8529. In this particular case, arangosh will
  try to connect on startup and after every executed shell command. The connect
  attempts all fail and time out after 300ms.
  In this case we now don't try to reconnect after every command.

* Added 'custom-query' testcase to arangobench to allow execution of custom
  queries.
  This also adds the options `--custom-query` and `--custom-query-file` for
  arangobench.

* Addition to the internal Refactoring of K_PATHS feature: K_PATHS queries are
  now being executed on the new refactored graph engine in a clustered
  environment. This change should not have any visible effect on users.

* Reduce memory footprint of agency Store in Node class.

* On Windows create a minidump in case of an unhandled SEH exception for
  post-mortem debugging.

* Add JWT secret support for arangodump and arangorestore, i.e. they now also
  provide the command-line options `--server.ask-jwt-secret` and
  `--server.jwt-secret-keyfile` with the same meanings as in arangosh.

* Add optional hyperlink to program option sections for information purposes,
  and add optional sub-headlines to program options for better grouping.
  These changes will be visible only when using `--help`.

* For Windows builds, remove the defines
  `_SILENCE_ALL_CXX17_DEPRECATION_WARNINGS` and `_ENABLE_ATOMIC_ALIGNMENT_FIX`
  that were needed to build Boost components with MSVC in older versions of
  Boost and MSVC.
  Both of these defines are obsolete nowadays.

* Database initial sync considers document count on leader for estimating
  timeouts when over 1 million docs on leader.

* Fixed issue #13117: Aardvark: Weird cursor offsets in query editor.

  Disabled font ligatures for Ace editor in Web UI to avoid rare display issue.

* Make all AQL cursors return compact result arrays.

  As a side-effect of this change, this makes profiling (i.e. using
  `db._profileQuery(...)` work for streaming queries as well. Previously,
  profiling a streaming query could have led to some internal errors, and even
  query results being returned, even though profiling a query should not return
  any query results.

* Try to raise file descriptors limit in local start scripts (in `scripts/`
  directory - used for development only).

* Fixed replication bug in MerkleTree sync protocol, which could lead to data
  corruption. The visible effect was that shards could no longer get in sync
  since the counts would not match after sync, even after a recount.
  This corruption only happened if there were large amounts of differences (at
  least 65537) and the destination side had newer revisions for some keys than
  the source side.

* Simplify the DistributeExecutor and avoid implicit modification of its input
  variable. Previously the DistributeExecutor could update the input variable
  in-place, leading to unexpected results (see #13509).
  The modification logic has now been moved into three new _internal_ AQL
  functions (MAKE_DISTRIBUTE_INPUT, MAKE_DISTRIBUTE_INPUT_WITH_KEY_CREATION, and
  MAKE_DISTRIBUTE_GRAPH_INPUT) and an additional calculation node with an
  according function call will be introduced if we need to prepare the input
  data for the distribute node.

* Added new REST APIs for retrieving the sharding distribution:

  - GET `/_api/database/shardDistribution` will return the number of
    collections, shards, leaders and followers for the database it is run
    inside. The request can optionally be restricted to include data from only a
    single DB server, by passing the `DBserver` URL parameter.

    This API can only be used on coordinators.

  - GET `/_admin/cluster/shardDistribution` will return global statistics on the
    current shard distribution, showing the total number of databases,
    collections, shards, leaders and followers for the entire cluster.
    The results can optionally be restricted to include data from only a single
    DB server, by passing the `DBserver` URL parameter.
    By setting the `details` URL parameter, the response will not contain
    aggregates, but instead one entry per available database will be returned.

    This API can only be used in the `_system` database of coordinators, and
    requires admin user privileges.

* Decrease the size of serialized index estimates, by introducing a compressed
  serialization format. The compressed format uses the previous uncompressed
  format internally, compresses it, and stores the compressed data instead. This
  makes serialized index estimates a lot smaller, which in turn decreases the
  size of I/O operations for index maintenance.

* More improvements for logging:

  - Added new REST API endpoint GET `/_admin/log/entries` to return log entries
    in a more intuitive format, putting each log entry with all its properties
    into an object. The API response is an array with all log message objects
    that match the search criteria.
    This is an extension to the already existing API endpoint GET `/_admin/log`,
    which returned log messages fragmented into 5 separate arrays.

    The already existing API endpoint GET `/_admin/log` for retrieving log
    messages is now deprecated, although it will stay available for some time.

  - Truncation of log messages now takes JSON format into account, so that the
    truncation of oversized JSON log messages still keeps a valid JSON structure
    even after the truncation.

  - The maximum size of in-memory log messages was doubled from 256 to 512
    chars, so that longer parts of each log message can be preserved now.

* Fix `/_admin/cluster/removeServer` API.
  This often returned HTTP 500 with an error message "Need open Array" due to an
  internal error when setting up agency preconditions.

* Remove logging startup options `--log.api-enabled` and `--log.keep-logrotate`
  for all client tools (arangosh, arangodump, arangorestore etc.), as these
  options are only meaningful for arangod.

* Extend the "move-calculations-up" optimizer rule so that it can move
  calculations out of subqueries into the outer query.

* Don't allocate ahead-of-time memory for striped PRNG array in arangod, but
  instead use thread-local PRNG instances. Not only does this save a few
  megabytes of memory, but it also avoids potential (but unlikely) sharing of
  the same PRNG instance by multiple threads.

* Remove undocumented CMake variable `USE_BACKTRACE`, and remove define
  `ARANGODB_ENABLE_BACKTRACE`. Both were turned off by default before, and when
  turned on allow to produce backtraces from within the executable in case debug
  symbols were available, working and the build was also compiled with
  `USE_MAINTAINER_MODE=On`. Some code in this context was obviously unreachable,
  so now it has all been removed.
  To log a backtrace from within arangod, it is now possible to call
  `CrashHandler::logBacktrace()`, which will log a backtrace of the calling
  thread to the arangod log. This is restricted to Linux builds only.

* Fix warnings about suggest-override which can break builds when warnings aret
  reated as errors.

* Turn off option `--server.export-read-write-metrics` for now, until there is
  certainty about the runtime overhead it introduces.

* Remove unsafe query option `inspectSimplePlans`. This option previously
  defaulted to `true`, and turning it off could make particular queries fail.
  The option was ignored in the cluster previously, and turning it off only had
  an effect in single server, there making very simple queries (queries not
  containing any FOR loops) not going through the optimizer's complete pipeline
  as a performance optimization. However, the optimization was only possible for
  a very small number of queries and even had adverse effects, so it is now
  removed entirely.

* On Linux and MacOS, require at least 8192 usable file descriptors at startup.
  If less file descriptors are available to the arangod process, then the
  startup is automatically aborted.

  Even the chosen minimum value of 8192 will often not be high enough to store
  considerable amounts of data. However, no higher value was chosen in order to
  not make too many existing small installations fail at startup after
  upgrading.

  The required number of file descriptors can be configured using the startup
  option `--server.descriptors-minimum`. It defaults to 8192, but it can be
  increased to ensure that arangod can make use of a sufficiently high number of
  files. Setting `--server.descriptors-minimum` to a value of `0` will make the
  startup require only an absolute minimum limit of 1024 file descriptors,
  effectively disabling the change.
  Such low values should only be used to bypass the file descriptors check in
  case of an emergency, but this is not recommended for production.

* Added metric `arangodb_transactions_expired` to track the total number of
  expired and then garbage-collected transactions.

* Allow toggling the document read/write counters and histograms via the new
  startup option `--server.export-read-write-metrics false`. This option
  defaults to `true`, so these metrics will be exposed by default.

* Upgraded bundled version of libunwind to v1.5.

* Added startup option `--javascript.tasks` to allow turning off JavaScript
  tasks if not needed. The default value for this option is `true`, meaning
  JavaScript tasks are available as before.
  However, with this option they can be turned off by admins to limit the amount
  of JavaScript user code that is executed.

* Only instantiate a striped PRNG instance for the arangod server, but not for
  any of the client tools (e.g. arangosh, arangodump, arangorestore).
  The client tools do not use the striped PRNG, so we can save a few MBs of
  memory for allocating the striped PRNG instance there, plus some CPU time for
  initializing it.

* Improve shard synchronization protocol by only transferring the required parts
  of the inventory from leader to follower. Previously, for each shard the
  entire inventory was exchanged, which included all shards of the respective
  database with all their details.
  In addition, save 3 cluster-internal requests per shard in the initial shard
  synchronization protocol by reusing already existing information in the
  different steps of the replication process.

* Added metric `arangodb_scheduler_low_prio_queue_last_dequeue_time` that
  provides the time (in millseconds) it took for the most recent low priority
  scheduler queue item to bubble up to the queue's head. This metric can be used
  to estimate the queuing time for incoming requests.
  The metric will be updated probabilistically when a request is pulled from the
  scheduler queue, and may remain at its previous value for a while if only few
  requests are coming in or remain permanently at its previous value if no
  further requests are incoming at all.

* Allow {USER} placeholder string also in `--ldap.search-filter`.

* Fixed some wrong behavior in single document updates. If the option
  ignoreRevs=false was given and the precondition _rev was given in the body but
  the _key was given in the URL path, then the rev was wrongly taken as 0,
  rather than using the one from the document body.

* Improved logging for error 1489 ("a shard leader refuses to perform a
  replication operation"). The log message will now provide the database and
  shard name plus the differing information about the shard leader.

* Add shard-parallelism to arangodump when dumping collections with multiple
  shards.
  Previously, arangodump could execute a dump concurrently on different
  collections, but it did not parallelize the dump for multiple shards of the
  same collection.
  This change should speed up dumping of collections with multiple shards.
  When dumping multiple shards of the same collection concurrently, parallelism
  is still limited by all these threads needing to serialize their chunks into
  the same (shared) output file.

* Add option `--envelope` for arangodump, to control if each dumped document
  should be wrapped into a small JSON envelope (e.g.
  `{"type":2300,"data":{...}}`). This JSON envelope is not necessary anymore
  since ArangoDB 3.8, so omitting it can produce smaller (and slightly faster)
  dumps.
  Restoring a dump without these JSON envelopers is handled automatically by
  ArangoDB 3.8 and higher. Restoring a dump without these JSON envelopes into
  previous versions (pre 3.8) however is not supported. Thus the option should
  only be used if the client tools (arangodump, arangorestore) and the arangod
  server are all using v3.8 or higher, and the dumps will never be stored into
  earlier versions.
  The default value for this option is `true`, meaning the JSON wrappers will be
  stored as part of the dump. This is compatible with all previous verions.

* Make AQL optimizer rule "splice-subqueries" mandatory, in the sense that it
  cannot be disabled anymore. As a side effect of this change, there will no
  query execution plans created by 3.8 that contain execution nodes of type
  `SubqueryNode`. `SubqueryNode`s will only be used during query planning and
  optimization, but at the end of the query optimization phase will all have
  been replaced with nodes of types `SubqueryStartNode` and `SubqueryEndNode`.
  The code to execute non-spliced subqueries remains in place so that 3.8 can
  still execute queries planned on a 3.7 instance with the "splice-subqueries"
  optimizer rule intentionally turned off. The code for executing non-spliced
  subqueries can be removed in 3.9.

* AQL query execution plan register usage optimization.

  This is a performance optimization that may positively affect some AQL queries
  that use a lot of variables that are only needed in certain parts of the
  query.
  The positive effect will come from saving registers, which directly translates
  to saving columns in AqlItemBlocks.

  Previously, the number of registers that were planned for each depth level of
  the query never decreased when going from one level to the next. Even though
  unused registers were recycled since 3.7, this did not lead to unused
  registers being completely dismantled.

  Now there is an extra step at the end of the register planning that keeps
  track of the actually used registers on each depth, and that will shrink the
  number of registers for the depth to the id of the maximum register. This is
  done for each depth separately.
  Unneeded registers on the right hand side of the maximum used register are now
  discarded. Unused registers on the left hand side of the maximum used register
  id are not discarded, because we still need to guarantee that registers from
  depths above stay in the same slot when starting a new depth.

* Added metric `arangodb_aql_current_query` to track the number of currently
  executing AQL queries.

* Internal refactoring of K_PATH feature, with the goal to have all graph
  algorithms on the same framework. This change should not have any visible
  effect on users.

* Removed server-side JavaScript object `ArangoClusterComm`, so it cannot be
  used from inside JavaScript operations or Foxx.
  The `ArangoClusterComm` object was previously used inside a few internal
  JavaScript operations, but was not part of the public APIs.

* Restrict access to functions inside JavaScript objects `ArangoAgency` and
  `ArangoAgent` to JavaScript code that is running in privileged mode, i.e. via
  the server's emergency console, the `/_admin/execute` API (if turned on) or
  internal bootstrap scripts.

* Added startup option `--javascript.transactions` to allow turning off
  JavaScript transactions if not needed. The default value for this option is
  `true`, meaning JavaScript transactions are available as before.
  However, with this option they can be turned off by admins to limit the amount
  of JavaScript user code that is executed.

* Introduce a default memory limit for AQL queries, to prevent rogue queries
  from consuming the entire memory available to an arangod instance.

  The limit is introduced via changing the default value of the option
  `--query.memory-limit` from previously `0` (meaning: no limit) to a
  dynamically calculated value.
  The per-query memory limits defaults are now:

    Available memory:            0      (0MiB)  Limit:            0   unlimited, %mem:  n/a
    Available memory:    134217728    (128MiB)  Limit:     33554432     (32MiB), %mem: 25.0
    Available memory:    268435456    (256MiB)  Limit:     67108864     (64MiB), %mem: 25.0
    Available memory:    536870912    (512MiB)  Limit:    201326592    (192MiB), %mem: 37.5
    Available memory:    805306368    (768MiB)  Limit:    402653184    (384MiB), %mem: 50.0
    Available memory:   1073741824   (1024MiB)  Limit:    603979776    (576MiB), %mem: 56.2
    Available memory:   2147483648   (2048MiB)  Limit:   1288490189   (1228MiB), %mem: 60.0
    Available memory:   4294967296   (4096MiB)  Limit:   2576980377   (2457MiB), %mem: 60.0
    Available memory:   8589934592   (8192MiB)  Limit:   5153960755   (4915MiB), %mem: 60.0
    Available memory:  17179869184  (16384MiB)  Limit:  10307921511   (9830MiB), %mem: 60.0
    Available memory:  25769803776  (24576MiB)  Limit:  15461882265  (14745MiB), %mem: 60.0
    Available memory:  34359738368  (32768MiB)  Limit:  20615843021  (19660MiB), %mem: 60.0
    Available memory:  42949672960  (40960MiB)  Limit:  25769803776  (24576MiB), %mem: 60.0
    Available memory:  68719476736  (65536MiB)  Limit:  41231686041  (39321MiB), %mem: 60.0
    Available memory: 103079215104  (98304MiB)  Limit:  61847529063  (58982MiB), %mem: 60.0
    Available memory: 137438953472 (131072MiB)  Limit:  82463372083  (78643MiB), %mem: 60.0
    Available memory: 274877906944 (262144MiB)  Limit: 164926744167 (157286MiB), %mem: 60.0
    Available memory: 549755813888 (524288MiB)  Limit: 329853488333 (314572MiB), %mem: 60.0

  As previously, a memory limit value of `0` means no limitation.
  The limit values are per AQL query, so they may still be too high in case
  queries run in parallel. The defaults are intentionally high in order to not
  stop any valid, previously working queries from succedding.

* Added startup option `--audit.queue` to control audit logging queuing behavior
  (Enterprise Edition only):

  The option controls whether audit log messages are submitted to a queue and
  written to disk in batches or if they should be written to disk directly
  without being queued.
  Queueing audit log entries may be beneficial for latency, but can lead to
  unqueued messages being lost in case of a power loss or crash. Setting this
  option to `false` mimics the behavior from 3.7 and before, where audit log
  messages were not queued but written in a blocking fashion.

* Added metric `arangodb_server_statistics_cpu_cores` to provide the number of
  CPU cores visibile to the arangod process. This is the number of CPU cores
  reported by the operating system to the process.
  If the environment variable `ARANGODB_OVERRIDE_DETECTED_NUMBER_OF_CORES` is
  set to a positive value at instance startup, this value will be returned
  instead.

* `COLLECT WITH COUNT INTO x` and `COLLECT var = expr WITH COUNT INTO x` are now
  internally transformed into `COLLECT AGGREGATE x = LENGTH()` and
  `COLLECT var = expr AGGREGATE x = LENGTH()` respectively. In addition, any
  argument passed to the `COUNT`/`LENGTH` aggregator functions are now optimized
  away. This not only simplified the code, but also allows more query
  optimizations:
    - If the variable in `COLLECT WITH COUNT INTO var` is not used, the implicit
      aggregator is now removed.
    - All queries of the form `COLLECT AGGREGATE x = LENGTH()` are now executed
      using the count executor, which can result in significantly improved
      performance.

* Added AQL timezone functions `DATE_TIMEZONE` and `DATE_TIMEZONES`.

* Make DB servers report storage engine health to the agency, via a new "health"
  attribute in requests sent to Sync/ServerStates/<id>.
  The supervision can in the future check this attribute if it is posted, and
  mark servers as BAD or FAILED in case an unhealthy status is reported.
  DB server health is currently determined by whether or not the storage engine
  (RocksDB) has reported a background error, and by whether or not the free disk
  space has reached a critical low amount. The current threshold for free disk
  space is set at 1% of the disk capacity (only the disk is considered that
  contains the RocksDB database directory).
  The minimum required free disk space percentage can be configured using the
  new startup option `--rocksdb.minimum-disk-free-percent`, which needs to be
  between 0 and 1 (including). A value of 0 disables the check.
  The minimum required free disk space can also be configured in bytes using the
  new startup option `--rocksdb.minimum-disk-free-bytes`. A value of 0 disables
  this check, too.

* Failed servers are now reported consistently in the web interface, at
  approximately the same time in the navigation bar and in the nodes view.
  Previously these two places had their own, independent poll mechanism for the
  nodes' health, and they were updated independently, which could cause an
  inconsistent view of the nodes' availability.
  Using only one poll mechanism instead also saves some period background
  requests for the second availability check.

* Stabilize a Foxx cleanup test.

* Drop a pair of braces {} in /_admin/metrics in case of empty labels, which
  makes the API adhere better to the official Prometheus syntax.

* Add some more metrics to the ConnectionPool.

* Reduce overhead of audit logging functionality if audit logging is turned off.

* Add several more attributes to audit-logged queries, namely query execution
  time and exit code (0 = no error, other values correspond to general ArangoDB
  error codes).

* Fixed a bug in maintainer mode sorting followerinfo lists the wrong way.

* Limit value of `--rocksdb.block-cache-size` to 1 GB for agent instances to
  reduce agency RAM usage, unless configured otherwise. In addition, limit the
  value of `--rocksdb.total-write-buffer-size` to 512 MB on agent instances for
  the same reason.

* Added new `rocksdb_write_stalls` and `rocksdb_write_stops` counter metrics,
  which should be more accurate than existing metrics related to the underlying
  conditions.

* Increased the default value of `--rocksdb.min-write-buffer-number-to-merge` in
  some cases when we have allocated a sufficient amount of memory to the write
  buffers for this to make sense. The increased value should help prevent
  compaction-induced write stalls/stops, and should only be enabled when under
  conditions such that it shouldn't greatly increase the chance of flush-induced
  write stalls/stops.

* Changed the default values for `--rocksdb.cache-index-and-filter-blocks` and
  `--rocksdb.cache-index-and-filter-blocks-with-high-priority` to true to
  improve control over memory usage.

* Lowered the minimum allowed value for `--rocksdb.max-write-buffer-number` from
  9 to 4 to allow more fine-grained memory usage control.

* Added new ArangoSearch view option 'countApproximate' for customizing view
  count strategy.

* Views on SmartGraph Edge collections do not contain some documents twice.

* Fixed issue #12248: Web UI - Added missing HTML escaping in the setup script
  section of a Foxx app.

* The scheduler will now run a minimum of 4 threads at all times, and the
  default and minimal value for `--server.maximal-threads` has been lowered from
  64 to the greater of 32 and twice the number of detected cores.

* Throttle work coming from low priority queue, according to a constant and to
  an estimate taking into account fanout for multi-shard operations.

* Move to 4 priority levels "low", "medium", "high" and "maintenance" in
  scheduler to ensure that maintenance work and diagnostics is always possible,
  even in the case of RocksDB throttles. Do not allow any RocksDB work on
  "maintenance".

* Commit replications on high priority queue.

* Essentially get rid of timeout in replication to drop followers. This is now
  entirely handled via reboot and failure tracking. The timeout has now a
  default minimum of 15 minutes but can still be configured via options.

* Additional metrics for all queue lengths and low prio ongoing work.

* New metric for number and total time of replication operations.

* New metrics for number of internal requests in flight, internal request
  duration, and internal request timeouts

* Fix `Gauge` assignment operators.

* Add cluster support for collection.checksum() method to calculate CRC
  checksums for collections.

* Make all Pregel HTTP and JavaScript APIs also accept stringified execution
  number values, in addition to numeric ones.

  This allows passing larger execution numbers as strings, so that any data loss
  due to numeric data type conversion (uint32 => double) can be avoided.

  The change also makes the Pregel HTTP and JavaScript APIs for starting a run
  return a stringified execution number, e.g. "12345" instead of 12345.

* Turn off `StatisticsWorker` thread on DB servers.
  This thread was previously only running queries on the local RocksDB instance,
  but using the cluster-wide collection names. So effectively it did nothing
  except use a bit of background CPU. In this case it is better to turn off the
  background thread entirely on the DB servers.

* Avoid the usage of std::regex when constructing date/time string values for
  log messages. This is a performance optimization only.

* Increase background garbage-collection interval for cluster transactions from
  1 second to 2 seconds. This change should reduce the amount of background task
  activity a tiny bit (though hardly measurable on an otherwise idle server).

* Make the audit log honor the configured logging date/time output format (i.e.
  `--log.time-format` option). Previously the audit logging always created a
  time value in the server's local time, and logged it in format
  YYYY-MM-DDTHH:MM:SS.

  From 3.8 onwards, the audit logger will honor the date/time format specified
  via the `--log.time-format` option, which defaults to `utc-datestring`. The
  means the audit logging will by default log all dates/times in UTC time. To
  restore the pre-3.8 behavior, please set the option `--log.time-format` to
  `local-datestring`, which will make the audit logger (and all other server log
  messages) use the server's local time.

* Added metrics for the system CPU usage:
  - `arangodb_server_statistics_user_percent`: Percentage of time that the
    system CPUs have spent in user mode
  - `arangodb_server_statistics_system_percent`: Percentage of time that the
    system CPUs have spent in kernel mode
  - `arangodb_server_statistics_idle_percent`: Percentage of time that the
    system CPUs have been idle
  - `arangodb_server_statistics_iowait_percent`: Percentage of time that the
    system CPUs have been waiting for I/O

  These metrics resemble the overall CPU usage metrics in `top`. They are
  available on Linux only.

* Fix log topic of general shutdown message from "cluster" to general.

* Automatically add "www-authenticate" headers to server HTTP 401 responses, as
  required by the HTTP specification.

* Enable HTTP request statistics and provide metrics even in case
  `--server.statistics-history` is set to `false` (this option will set itself
  to off automatically on agency instances on startup if not explicitly set).
  This change provides more metrics on all server instances, without the need to
  persist them in the instance's RocksDB storage engine.

* Remove extra CMake option `DEBUG_SYNC_REPLICATION` and use the already
  existing `USE_FAILURE_TESTS` options for its purpose.

* Updated bundled version of Snappy compression/decompression library to 1.1.8.

* Added support of `GEO_DISTANCE`, `GEO_CONTAINS`, `GEO_INTERSECTS`,
  `GEO_IN_RANGE` to ArangoSearch.

* Added new `GeoJSON` ArangoSearch analyzer.

* Added new `GeoPoint` ArangoSearch analyzer.

* Added new `GEO_IN_RANGE` AQL function.

* Added new 'aql' type for ArangoSearch analyzers.

* Obsoleted the startup options `--database.throw-collection-not-loaded-error`
  and `--ttl.only-loaded-collection`.

  These options were meaningful for the MMFiles storage engine only, but for the
  RocksDB storage engine they did not make any difference. Using these startup
  options is still possible, but will have no effect other than generating a
  warning at server startup.

* Added CMake option `USE_MINIMAL_DEBUGINFO`.
  This option is turned off by default. If turned on, the created binaries
  will contain only a minimum amount of debug symbols, reducing the size of the
  executables. If turned off (which is the default), the binaries will contain
  full debug information, which will make them bigger in size unless the debug
  information is later stripped again.

* Modified the returned error code for calling the `shards()` function on a
  collection in single-server from "internal error" (error number 4) to "shards
  API is only available in cluster" and error number 9, HTTP status code 501.

* Added WINDOW keyword to AQL to allow aggregations on related rows.

* Added new graph method K_PATHS to AQL. This will enumerate all paths between a
  source and a target vertex that match the given length.
  For example, the query
  ```
    FOR path IN 2..4 OUTBOUND K_PATHS "v/source" TO "v/target" GRAPH "g"
      RETURN path
  ```
  will yield all paths in format
    {
      vertices: [v/source, ... , v/target],
      edges: [v/source -> v/1, ...,  v/n -> v/target
    }
  that have length exactly 2 or 3 or 4, start at v/source and end at v/target.
  The order of those paths in the result set is not guaranteed.

* Fixed issue BTS-195: AQL update queries using the `keepNull` option set to
  false had an inconsistent behavior. For example, given a collection `test`
  with an empty document with just key `testDoc`, the following query would
  return different results when running for the first time or the second time:

    UPDATE 'testDoc'
        WITH {test: {sub1: true, sub2: null}} IN test
    OPTIONS { keepNull: false, mergeObjects: true }

  For its first run, the query would return

      {
        "_key": "testDoc",
        "test": {
          "sub1": true,
          "sub2": null
        }
      }

  (with the `null` attribute value not being removed). For all subsequent runs,
  the same query would return

      {
        "_key": "testDoc",
        "test": {
          "sub1": true,
        }
      }

  (with the `null` value removed as requested).

  This inconsistency was due to how the `keepNull` attribute was handled if the
  attribute already existed in the to-be-updated document or not. The behavior
  is now consistent, so `null` values are now properly removed from
  sub-attributes even if in the to-be-updated document the target attribute did
  not yet exist. This makes such updates idempotent again.

  This a behavior change compared previous versions, but it will only have
  effect when `keepNull` is set to `false` (the default value is `true`
  however), and only when just-inserted object sub-attributes contained `null`
  values.

* Optimization of empty append entries.

* Remove any special handling for obsoleted collection attributes
  `indexBuckets`, `journalSize`, `doCompact` and `isVolatile`. These attributes
  were meaningful only with the MMFiles storage engine and have no meaning with
  the RocksDB storage engine. Thus any special handling for these attributes can
  be removed in the internal code.
  Client applications and tests that rely on the behavior that setting any of
  these attributes produces an error when using the RocksDB engine may need
  adjustment now.

* Added a --continue option to arangorestore. arangorestore now keeps track of
  the progress and can continue the restore operation when some error occured.

* Don't respond with misleading error in smart vertex collections.

  When inserting a document with a non-conforming key pattern into a smart
  vertex collection, the response error code and message are 1466
  (ERROR_CLUSTER_MUST_NOT_SPECIFY_KEY) and "must not specify _key for this
  collection".
  This is misleading, because it is actually allowed to specify a key value for
  documents in such collection. However, there are some restrictions for valid
  key values (e.g. the key must be a string and contain the smart graph
  attribute value at the front, followed by a colon.
  If any of these restrictions are not met, the server currently responds with
  "must not specify key for this collection", which is misleading. This change
  rectifies it so that the server responds with error 4003
  (ERROR_KEY_MUST_BE_PREFIXED_WITH_SMART_GRAPH_ATTRIBUTE) and message "in smart
  vertex collections _key must be a string and prefixed with the value of the
  smart graph attribute". This should make it a lot easier to understand what
  the actual problem is.

* Fix an issue in arangoimport improperly handling filenames with less than 3
  characters. The specified input filename was checked for a potential ".gz"
  ending, but the check required the filename to have at least 3 characters.
  This is now fixed.

* Fix for BTS-191: Made transaction API database-aware.

* Minor clean up of and less verbosity in agent callbacks.

* Speed up initial replication of collections/shards data by not wrapping each
  document in a separate `{"type":2300,"data":...}` envelope. In addition, the
  follower side of the replication will request data from leaders in VelocyPack
  format if the leader is running at least version 3.8.
  Stripping the envelopes and using VelocyPack for transfer allows for smaller
  data sizes when exchanging the documents and faster processing, and thus can
  lead to time savings in document packing and unpacking as well as reduce the
  number of required HTTP requests.

* Added metric `arangodb_agency_callback_registered counter` for tracking the
  total number of agency callbacks that were registered.

* Added weighted traversal. Use `mode: "weighted"` as option to enumerate paths
  by increasing weights. The cost of an edge can be read from an attribute which
  can be specified using `weightAttribute` option.

* Fixed issue ES-696: SEARCH vs FILTER lookup performance.
  Consolidation functionality for ArangoSearch view links was able to hit non-
  mergable enormous amount of segments due to improper scheduling logic.

* Make scheduler react and start new threads slightly faster in case a lot of
  new work arrives.

* Added new ArangoSearch "pipeline" analyzer type.

* Added replication metrics `arangodb_replication_initial_sync_bytes_received`
  for the number of bytes received during replication initial sync operations
  and `arangodb_replication_tailing_bytes_received` for the number of bytes
  received for replication tailing requests.
  Also added `arangodb_replication_failed_connects` to track the number of
  connection failures or non-OK response during replication.

* Added metrics `rocksdb_free_inodes` and `rocksdb_total_inodes` to track the
  number of free inodes and the total/maximum number of inodes for the file
  system the RocksDB database directory is located in. These metrics will always
  be 0 on Windows.

* Fixed slightly wrong log level for authentication and also added login event
  to the standard log.

* Added new metrics for the total and the free disk space for the mount used for
  the RocksDB database directory:

  - `arangodb_rocksdb_free_disk_space`: provides the free disk space for the
    mount, in bytes
  - `arangodb_rocksdb_total_disk_space`: provides the total disk space of the
    mount, in bytes

* Apply user-defined idle connection timeouts for HTTP/2 and VST connections.
  The timeout value for idle HTTP/2 and VST connections can now be configured
  via the configuration option `--http.keep-alive-timeout` in the same way as
  for HTTP/1 connections.
  HTTP/2 and VST connections that are sending data back to the client are now
  closed after 300 seconds or the configured idle timeout (the higher of both
  values is used here).
  Before this change, the timeouts for HTTP/2 and VST connections were
  hardcoded to 120 seconds, and even non-idle connections were closed after this
  timeout.

* Added new metrics for replication:
  - `arangodb_replication_dump_requests`: number of replication dump requests
    made.
  - `arangodb_replication_dump_bytes_received`: number of bytes received in
    replication dump requests.
  - `arangodb_replication_dump_documents`: number of documents received in
    replication dump requests.
  - `arangodb_replication_dump_request_time`: wait time for replication dump
    requests.
  - `arangodb_replication_dump_apply_time`: time required for applying data from
    replication dump responses.
  - `arangodb_replication_initial_sync_keys_requests`: number of replication
    initial sync keys requests made.
  - `arangodb_replication_initial_sync_docs_requests`: number of replication
    initial sync docs requests made.
  - `arangodb_replication_initial_sync_docs_requested`: number of documents
    requested via replication initial sync requests.
  - `arangodb_replication_initial_sync_docs_inserted`: number of documents
    inserted by replication initial sync.
  - `arangodb_replication_initial_sync_docs_removed`: number of documents
    inserted by replication initial sync.
  - `arangodb_replication_initial_chunks_requests_time`: wait time histogram for
    replication key chunks determination requests.
  - `arangodb_replication_initial_keys_requests_time`: wait time for replication
    keys requests.
  - `arangodb_replication_initial_docs_requests_time`: time needed to apply
    replication docs data.
  - `arangodb_replication_initial_insert_apply_time`: time needed to apply
    replication initial sync insertions.
  - `arangodb_replication_initial_remove_apply_time`: time needed to apply
    replication initial sync removals.
  - `arangodb_replication_initial_lookup_time`: time needed for replication
    initial sync key lookups.
  - `arangodb_replication_tailing_requests`: number of replication tailing
    requests.
  - `arangodb_replication_tailing_follow_tick_failures`: number of replication
    tailing failures due to missing tick on leader.
  - `arangodb_replication_tailing_markers`: number of replication tailing
    markers processed.
  - `arangodb_replication_tailing_documents`: number of replication tailing
    document inserts/replaces processed.
  - `arangodb_replication_tailing_removals`: number of replication tailing
    document removals processed.
  - `arangodb_replication_tailing_bytes_received`: number of bytes received for
    replication tailing requests.
  - `arangodb_replication_tailing_request_time`: wait time for replication
    tailing requests.
  - `arangodb_replication_tailing_apply_time`: time needed to apply replication
    tailing markers.

* Allow calling of REST APIs `/_api/engine/stats`, GET `/_api/collection`, GET
  `/_api/database/current` and GET `/_admin/metrics` on followers in active
  failover deployments. This can help debugging and inspecting the follower.

* Support projections on sub-attributes (e.g. `a.b.c`).

  In previous versions of ArangoDB, projections were only supported on top-level
  attributes. For example, in the query

      FOR doc IN collection
        RETURN doc.a.b

  the projection that was used was just `a`. Now the projection will be `a.b`,
  which can help reduce the amount of data to be extracted from documents, when
  only some sub-attributes are accessed.

  In addition, indexes can now be used to extract the data of sub-attributes for
  projections. If for the above example query an index on `a.b` exists, it will
  be used now. Previously, no index could be used for this projection.

  Projections now can also be fed by any attribute in a combined index. For
  example, in the query

    FOR doc IN collection
      RETURN doc.b

  the projection can be satisfied by a single-attribute index on attribute `b`,
  but now also by a combined index on attributes `a` and `b` (or `b` and `a`).

* Remove some JavaScript files containing testsuites and test utilities from our
  official release packages.

* Show optimizer rules with highest execution times in explain output.

* Renamed "master" to "leader" and "slave" to "follower" in replication
  messages.
  This will change the contents of replication log messages as well the string
  contents of replication-related error messages.

  The messages of the error codes 1402, 1403 and 1404 were also changed
  accordingly, as well as the identifiers:
  - `TRI_ERROR_REPLICATION_MASTER_ERROR` renamed to
    `TRI_ERROR_REPLICATION_LEADER_ERROR`
  - `TRI_ERROR_REPLICATION_MASTER_INCOMPATIBLE` renamed to
    `TRI_ERROR_REPLICATION_LEADER_INCOMPATIBLE`
  - `TRI_ERROR_REPLICATION_MASTER_CHANGE` renamed to
    `TRI_ERROR_REPLICATION_LEADER_CHANGE`

  This change also renames the API endpoint `/_api/replication/make-slave` to
  `/_api/replication/make-follower`. The API is still available under the old
  name, but using it is deprecated.

* Make optimizer rule "remove-filters-covered-by-index" remove FILTERs that were
  referring to aliases of the collection variable, e.g.

      FOR doc IN collection
        LET value = doc.indexedAttribute
        FILTER value == ...

  Previously, FILTERs that were using aliases were not removed by that optimizer
  rule.
  In addition, the optimizer rule "remove-unnecessary-calculations" will now run
  again in case it successfully removed variables. This can unlock further
  removal of unused variables in sequences such as

      FOR doc IN collection
        LET value = doc.indexedAttribute
        LET tmp1 = value > ...
        LET tmp2 = value < ...

  when the removal of `tmp1` and `tmp2` makes it possible to also remove the
  calculation of `value`.

* Fixed issue BTS-168: Fixed undefined behavior that did trigger segfaults on
  cluster startups. It is only witnessed for macOS based builds. The issue could
  be triggered by any network connection.
  This behavior is not part of any released version.

* Hard-code returned "planVersion" attribute of collections to a value of 1.
  Before 3.7, the most recent Plan version from the agency was returned inside
  "planVersion".
  In 3.7, the attribute contained the Plan version that was in use when the
  in-memory LogicalCollection object was last constructed. The object was always
  reconstructed in case the underlying Plan data for the collection changed or
  when a collection contained links to arangosearch views.
  This made the attribute relatively useless for any real-world use cases, and
  so we are now hard-coding it to simplify the internal code. Using the
  attribute in client applications is also deprecated.

* Don't prevent concurrent synchronization of different shards from the same
  database. Previously only one shard was synchronized at a time per database.

* Wait until restore task queue is idle before shutting down.

* Fix a race problem in the unit tests w.r.t. PlanSyncer.

* Errors with error code 1200 (Arango conflict) will now get the HTTP response
  code 409 (Conflict) instead of 412 (Precondition failed), unless "if-match"
  header was used in `_api/document` or `_api/gharial`.

* Keep the list of last-acknowledged entries in Agency more consistent.
  During leadership take-over it was possible to get into a situation that the
  new leader does not successfully report the agency config, which was
  eventually fixed by the Agent itself. Now this situation is impossible.

* Added support `db._engineStats()` API in coordinator. Previously calling this
  API always produced an empty result. Now it will return the engine statistics
  as an object, with an entry for each individual DB-Server.

* Added option `--log.use-json-format` to switch log output to JSON format.
  Each log message then produces a seperate line with JSON-encoded log data,
  which can be consumed by applications.

* Added option `--log.process` to toggle the logging of the process id (pid) in
  log messages. Logging the process id is useless when running arangod in Docker
  containers, as the pid will always be 1. So one may as well turn it off in
  these contexts.

* Added option `--log.in-memory` to toggle storing log messages in memory, from
  which they can be consumed via the `/_admin/log` and by the web UI. By
  default, this option is turned on, so log messages are consumable via API and
  the web UI. Turning this option off will disable that functionality and save a
  tiny bit of memory for the in-memory log buffers.


v3.7.10 (2021-03-14)
--------------------

* Reasonably harden MoveShard against unexpected VelocyPack input.

* Follower DB servers will now respond with error code
  `TRI_ERROR_CLUSTER_SHARD_FOLLOWER_REFUSES_OPERATION`
  to any read request. This fixes inadequate HTTP 404 responses from followers,
  e.g. during chaos tests.

* Fixed Github issue #13632: Query Fails on Upsert with Replace_nth.

* Updated arangosync to 1.2.3.

* Backported AQL sort performance improvements from devel.
  This change can improve the performance of local sorts operations, e.g.

  Baseline (3.7.9):

      Query String (94 chars, cacheable: false):
       FOR i IN 1..500000 LET value = CONCAT('testvalue-to-be-sorted', i) SORT value ASC RETURN value

      Execution plan:
       Id   NodeType            Calls    Items   Runtime [s]   Comment
        1   SingletonNode           1        1       0.00003   * ROOT
        2   CalculationNode         1        1       0.00003     - LET #2 = 1 .. 500000   /* range */   /* simple expression */
        3   EnumerateListNode     500   500000       0.08725     - FOR i IN #2   /* list iteration */
        4   CalculationNode       500   500000       0.22722       - LET value = CONCAT("testvalue-to-be-sorted", i)   /* simple expression */
        5   SortNode              500   500000       2.05180       - SORT value ASC   /* sorting strategy: standard */
        6   ReturnNode            500   500000       0.02911       - RETURN value

      Query Statistics:
       Writes Exec   Writes Ign   Scan Full   Scan Index   Filtered   Exec Time [s]
                 0            0           0            0          0         2.39644

  With sort optimization (3.7.10):

      Query String (94 chars, cacheable: false):
       FOR i IN 1..500000 LET value = CONCAT('testvalue-to-be-sorted', i) SORT value ASC RETURN value

      Execution plan:
       Id   NodeType            Calls    Items   Runtime [s]   Comment
        1   SingletonNode           1        1       0.00002   * ROOT
        2   CalculationNode         1        1       0.00003     - LET #2 = 1 .. 500000   /* range */   /* simple expression */
        3   EnumerateListNode     500   500000       0.08755     - FOR i IN #2   /* list iteration */
        4   CalculationNode       500   500000       0.26161       - LET value = CONCAT("testvalue-to-be-sorted", i)   /* simple expression */
        5   SortNode              500   500000       1.36070       - SORT value ASC   /* sorting strategy: standard */
        6   ReturnNode            500   500000       0.02864       - RETURN value

      Query Statistics:
       Writes Exec   Writes Ign   Scan Full   Scan Index   Filtered   Exec Time [s]
                 0            0           0            0          0         1.73940

* Fixed a problem that coordinators would vanish from the UI and the Health API
  if one switched the agency Supervision into maintenance mode and kept left
  that maintenance mode on for more than 24h.

* Fixed a bug in the web interface that displayed the error "Not authorized to
  execute this request" when trying to create an index in the web interface in a
  database other than `_system` with a user that does not have any access
  permissions for the `_system` database.
  The error message previously displayed error actually came from an internal
  request made by the web interface, but it did not affect the actual index
  creation.

* Fixed issue BTS-309: The Graph API (Gharial) did not respond with the correct
  HTTP status code when validating edges. It now responds with 400 (Bad Request)
  as documented and a new, more precise error code (1947) and message if a
  vertex collection referenced in the _from or _to attribute is not part of the
  graph.


v3.7.9 (2021-03-01)
-------------------

* Fix issue #13476: The Java driver v6.9.0 (and older) has bad performance when
  iterating over AQL cursor results in certain cases. This works around this.
  This workaround will no longer be available in 3.8.

* Enable statistics in web UI in non-`_system` databases in cluster mode.
  In cluster mode, the web UI dashboard did not display statistics properly when
  not being logged in to the `_system` database. For all other databases than
  `_system`, no statistics were displayed but just some "No data..."
  placeholders.
  Statistics for non-`_system` databases were not properly displayed since
  3.7.6 due to an internal change in the statistics processing.

  In addition, a new startup option `--server.statistics-all-databases` controls
  whether cluster statistics are displayed in the web interface for all
  databases (if the option is set to `true`) or just for the system database (if
  the option is set to `false`).
  The default value for the option is `true`, meaning statistics will be
  displayed in the web interface for all databases.

* Updated OpenSSL to 1.1.1j and OpenLDAP to 2.4.57.

* Cleanup old HotBackup transfer jobs in agency.

* Added logging of elapsed time of ArangoSearch commit/consolidation/cleanup
  jobs.

* Fix too early stop of replication, when waiting for keys in large
  collections/shards.

* Fixed issue BTS-268: fix a flaky Foxx self-heal procedure.

* Fixed issue DEVSUP-720: Within an AQL query, the "COLLECT WITH COUNT INTO"
  statement could lead to a wrong count output when used in combination with an
  index which has been created with an array index attribute.

* Fix profiling of AQL queries with the `silent` and `stream` options sets in
  combination. Using the `silent` option makes a query execute, but discard all
  its results instantly. This led to some confusion in streaming queries, which
  can return the first query results once they are available, but don't
  necessarily execute the full query.
  Now, `silent` correctly discards all results even in streaming queries, but
  this has the effect that a streaming query will likely be executed completely
  when the `silent` option is set. This is not the default however, and the
  `silent` option is normally not set. There is no change for streaming queries
  if the `silent` option is not set.

  As a side-effect of this change, this makes profiling (i.e. using
  `db._profileQuery(...)` work for streaming queries as well. Previously,
  profiling a streaming query could have led to some internal errors, and even
  query results being returned, even though profiling a query should not return
  any query results.

* Improved the wording for sharding options displayed in the web interface.

  Instead of offering `flexible` and `single`, now use the more intuitive
  `Sharded` and `OneShard` options, and update the help text for them.

* EE only bugfix: On DisjointSmartGraphs that are used in anonymous way, there
  was a chance that the query could fail, if non-disjoint collections were part
  of the query. Named DisjointSmartGraphs have been save to this bug.
  Example:
  DisjointSmartGraph (graph) on vertices -edges-> vertices
  Query:

  WITH vertices, unrelated
  FOR out IN 1 OUTBOUND "v/1:1" edges
    FOR u IN unrelated
    RETURN [out, u]

  The "unrelated" collection was pulled into the DisjointSmartGraph, causing the
  AQL setup to create erroneous state.
  This is now fixed and the above query works.
  This query:

  WITH vertices, unrelated
  FOR out IN 1 OUTBOUND "v/1:1" GRAPH "graph"
    FOR u IN unrelated
    RETURN [out, u]

  was not affected by this bug.

* Avoid a potential deadlock when dropping indexes.

  A deadlock could theoretically happen for a thread that is attempting to drop
  an index in case there was another thread that tried to create or drop an
  index in the very same collection at the very same time. We haven't managed to
  trigger the deadlock with concurrency tests, so it may have been a theoretical
  issue only. The underlying code was changed anyway to make sure this will not
  cause problems in reality.

* Make dropping of indexes in cluster retry in case of precondition failed.

  When dropping an indexes of a collection in the cluster, the operation could
  fail with a "precondition failed" error in case there were simultaneous index
  creation or drop actions running for the same collection. The error was
  returned properly internally, but got lost at the point when
  `<collection>.dropIndex()` simply converted any error to just `false`.
  We can't make `dropIndex()` throw an exception for any error, because that
  would affect downwards-compatibility. But in case there is a simultaneous
  change to the collection indexes, we can just retry our own operation and
  check if it succeeds then. This is what `dropIndex()` will do now.

* Improve incremental sync replication for single server and cluster to cope
  with multiple secondary index unique constraint violations (before this was
  limited to a failure in a single unique secondary index). This allows
  replicating the leader state to the follower in basically any order, as any
  *other* conflicting documents in unique secondary indexes will be  detected
  and removed on the follower.

* Fix potential undefined behavior when iterating over connected nodes in an
  execution plan and calling callbacks for each of the nodes: if the callbacks
  modified the list of connected nodes of the current that they were called
  from, this could lead to potentially undefined behavior due to iterator
  invalidation. The issue occurred when using a debug STL via `_GLIBCXX_DEBUG`.

* Fixed a RocksDB bug which could lead to an assertion failure when compiling
  with STL debug mode -D_GLIBCXX_DEBUG.

* Fixed a rare internal buffer overflow around ridBuffers.

* Issue #13141: The `move-filters-into-enumerate` optimization, when applied to
  an EnumerateCollectionNode (i.e. full collection scan), did not do regular
  checks for the query being killed during the filtering of documents, resulting
  in the maxRuntime option and manual kill of a query not working timely.

* Do not create index estimator objects for proxy collection objects on
  coordinators and DB servers. Proxy objects are created on coordinators and DB
  servers for all shards, and they also make index objects available. In order
  to reduce the memory usage by these objects, we don't create any index
  estimator objects for indexes in those proxy objects. Index estimators usually
  take several KB of memory each, so not creating them will pay out for higher
  numbers of collections/shards.

* Improvements for logging. This adds the following startup options to arangod:

  - `--log.max-entry-length`: controls the maximum line length for individual
    log messages that are written into normal logfiles by arangod (note: this
    does not include audit log messages).
    Any log messages longer than the specified value will be truncated and the
    suffix '...' will be added to them. The purpose of this parameter is to
    shorten long log messages in case there is not a lot of space for logfiles,
    and to keep rogue log messages from overusing resources.
    The default value is 128 MB, which is very high and should effectively mean
    downwards-compatiblity with previous arangod versions, which did not
    restrict the maximum size of log messages.

  - `--audit.max-entry-length`: controls the maximum line length for individual
    audit log messages that are written into audit logs by arangod. Any audit
    log messages longer than the specified value will be truncated and the
    suffix '...' will be added to them.
    The default value is 128 MB, which is very high and should effectively mean
    downwards-compatiblity with previous arangod versions, which did not
    restrict the maximum size of log messages.

  - `--log.in-memory-level`: controls which log messages are preserved in
    memory. The default value is `info`, meaning all log messages of types
    `info`, `warning`, `error` and `fatal` will be stored by an instance in
    memory (this was also the behavior in previous versions of ArangoDB).
    By setting this option to `warning`, only `warning` log messages will be
    preserved in memory, and by setting the option to `error` only error
    messages will be kept.
    This option is useful because the number of in-memory log messages is
    limited to the latest 2048 messages, and these slots are by default shared
    between informational, warning and error messages.

* Honor the value of startup option `--log.api-enabled` when set to `false`.
  The desired behavior in this case is to turn off the REST API for logging, but
  was not implemented. The default value for the option is `true`, so the REST
  API is enabled. This behavior did not change, and neither did the behavior
  when setting the option to a value of `jwt` (meaning the REST API for logging
  is only available for superusers with a valid JWT token).

* Fix error reporting in the reloadTLS route.

* Split the update operations for the _fishbowl system collection with Foxx apps
  into separate insert/replace and remove operations. This makes the overall
  update not atomic, but as removes are unlikely here, we can now get away with
  a simple multi-document insert-replace operation instead of a truncate and an
  exclusive transaction, which was used before.


v3.7.8 (2021-02-16)
-------------------

* Fixed ES-784 regression related to encryption cipher propagation to
  ArangoSearch data.


v3.7.7 (2021-02-05)
-------------------

* Added metrics for document read and write operations:

  - `arangodb_document_writes: Total number of document write operations
    (successful and failed) not performed by synchronous replication.
  - `arangodb_document_writes_replication`: Total number of document write
    operations (successful and failed) by cluster synchronous replication.
  - `arangodb_collection_truncates`: Total number of collection truncate
    operations (successful and failed) not performed by cluster synchronous
    replication.
  - `arangodb_collection_truncates_replication`: Total number of collection
    truncate operations (successful and failed) by synchronous replication.
  - `arangodb_document_read_time`: Execution time histogram of all document
    primary key read operations (successful and failed) [s]. Note: this does not
    include secondary index lookups, range scans and full collection scans.
  - `arangodb_document_insert_time`: Execution time histogram of all document
    insert operations (successful and failed) [s].
  - `arangodb_document_replace_time`: Execution time histogram of all document
    replace operations (successful and failed) [s].
  - `arangodb_document_remove_time`: Execution time histogram of all document
    remove operations (successful and failed) [s].
  - `arangodb_document_update_time`: Execution time histogram of all document
    update operations (successful and failed) [s].
  - `arangodb_collection_truncate_time`: Execution time histogram of all
    collection truncate operations (successful and failed) [s].

  The timer metrics are turned off by default, and can be enabled by setting the
  startup option `--server.export-read-write-metrics true`.

* Fixed issue #12543: Unused Foxx service config can not be discarded.

* Fixed issue #12363: Foxx HTTP API upgrade/replace always enables development
  mode.

* Fixed BTS-284: upgrading from 3.6 to 3.7 in cluster enviroment.
  Moved upgrade ArangoSearch links task to later step as it needs cluster
  connection. Removed misleading error log records for failed ArangoSearch index
  creation during upgrade phase.

* Normalize user-provided input/output directory names in arangoimport,
  arangoexport, arangodump and arangorestore before splitting them into path
  components, in the sense that now both forward and backward slashes can be
  used on Windows, even interchangingly.

* Fixed some wrong behaviour in single document updates. If the option
  ignoreRevs=false was given and the precondition _rev was given in the body but
  the _key was given in the URL path, then the rev was wrongly taken as 0,
  rather than using the one from the document body.

* Allow {USER} paceholder string also in `--ldap.search-filter`.

* Make `padded` and `autoincrement` key generators export their `lastValue`
  values, so that they are available in dumps and can be restored elsewhere from
  a dump.

* Fix decoding of values in `padded` key generator when restoring from a dump.

* Fixed error reporting for hotbackup restore from dbservers back to
  coordinators. This could for example swallow out of disk errors during
  hotbackup restore.

* Fix decoding of values in `padded` key generator when restoring from a dump.

* Fixed some situations of
  [...]
  SUBQUERY
  FILTER
  LIMIT
  [...]
  in AQL queries, yielding incorrect responses. A distributed state within the
  subquery was not resetted correctly. This could also lead into "shrink" errors
  of AQL item blocks, or much higher query runtimes.
  Fixes:
  - BTS-252
  - ES-687
  - github issue: #13099
  - github issue: #13124
  - github issue: #13147
  - github issue: #13305
  - DEVSUP-665

* Fix a bug in the agency Supervision which could lead to removeFollower
  jobs constantly being created and immediately stopped again.

* Limit additional replicas in failover cases to +2.

* Prepare register planning for rolling upgrades. Previously, changes in
  register planning from 3.7 to a minor future version (i.e. 3.8) could cause
  queries executed by a 3.7 coordinator in combination with a minor future
  version (i.e. 3.8) DBServer to fail during a rolling upgrade.

* Fixed rare objectId conflict for indexes.

* Fix for OASIS-409: fixed indexing _id attribute at recovery.

* Fix some issues with key generators not properly taking into account the
  `allowUserKeys` attribute when in a cluster.

* Added the following bit handling functions to AQL:

  - BIT_AND(array): and-combined result
  - BIT_OR(array): or-combined result
  - BIT_XOR(array): xor-combined result
  - BIT_NEGATE(value, bits): bitwise negation of `value`, with a mask of `bits`
    length
  - BIT_TEST(value, index): test if bit at position `index` is set in `value`
    (indexes are 0-based)
  - BIT_POPCOUNT(value): return number of bits set in `value`
  - BIT_SHIFT_LEFT(value, shift, bits): bitwise shift-left of `value` by `shift`
    bits, with a mask of `bits` length
  - BIT_SHIFT_RIGHT(value, shift, bits): bitwise shift-right of `value` by
    `shift` bits, with a mask of `bits` length
  - BIT_CONSTRUCT(array): construct a number with bits set at the positions
    given in the array
  - BIT_DECONSTRUCT(value): deconstruct a number into an array of its individual
    set bits
  - BIT_TO_STRING(value): create a bitstring representation from numeric `value`
  - BIT_FROM_STRING(value): parse a bitstring representation into a number

  `BIT_AND`, `BIT_OR` and `BIT_XOR` are also available as aggregate functions
  for usage inside COLLECT AGGREGATE.

  All above bit operations support unsigned integer values with up to 32 bits.
  Using values outside the supported range will make any of these bit functions
  return `null` and register a warning.

* Add binary (base 2) and hexadecimal (base 16) integer literals to AQL.
  These literals can be used where regular (base 10) integer literal can used.
  The prefix for binary integer literals is `0b`, e.g. `0b10101110`.
  The prefix for hexadecimal integer literals i `0x`, e.g. `0xabcdef02`.

  Binary and hexadecimal integer literals can only be used for unsigned
  integers.
  The maximum supported value is `(2 ^ 32) - 1`, i.e. `0xffffffff` (hexadecimal)
  or `0b11111111111111111111111111111111` (binary).

* Print a version mismatch (major/minor version difference) between the arangosh
  version and the remote arangod version at arangosh startup.

* Fix a potential shutdown deadlock in AgencyCache.

* Updated arangosync to 1.2.2.

* Minor and rare AQL performance improvement, in nested subqueries:
  LET sq1 ([..] FILTER false == true LET sq2 = (<X>) [..])
  where sq1 produces no data (e.g. by the above filter) for sq2, the part <X>
  have been asked two times (second returns empty result), instead of one, if
  and only if the mainquery executes sq1 exactly one time.
  Now we get away with one call only.
  In the case sq1 has data, or sq1 is executed more often, only one call was
  needed (assuming the data fits in one batch).

* Improve internal error reporting by cluster maintenance.

* Bug-Fix: In one-shard-database setups that were created in 3.6.* and then
  upgraded to 3.7.5 the DOCUMENT method in AQL will now return documents again.


v3.7.6 (2021-01-04)
-------------------

* Updated OpenSSL to 1.1.1i and OpenLDAP to 2.4.56.

* Added new metric: "arangodb_collection_lock_sequential_mode" this will count
  how many times we need to do a sequential locking of collections. If this
  metric increases this indicates lock contention in transaction setup.
  Most likely this is caused by exlcusive locks used on collections with more
  than one shard.

* Fix for BTS-213
  Changed the transaction locking mechanism in the cluster case.
  For all installations that do not use "exclusive" collection locks this change
  will not be noticable. In case of "exclusive" locks, and collections with more
  than one shard, it is now less likely to get a LOCK_TIMEOUT (ErrorNum 18).
  It is still possible to get into the LOCK_TIMEOUT case, especially if the
  "exclusive" operation(s) are long-running.

* Fixed an endless busy loop which could happen if a coordinator tries to roll
  back a database creation, but the database has already been dropped by other
  means.

* Make internal ClusterInfo::getPlan() wait for initial plan load from agency.

* Remove HTTP "Connection" header when forwarding requests in the cluster from
  one coordinator to another, and let the internal network layer handle closing
  of connections and keep-alive.

* Prevent a write to RocksDB during recovery in the case that the database
  already exists. The write at startup is potentially blocking, and will delay
  the startup for servers that were shut down while in a write-stopped state.

* Fix recovery of "clientId" values in Agency when restarting an agent from
  persistence.

* Added "startupTime", "computationTime" and "storageTime" to Pregel result
  statistics.

* Add query execution time and query id to audit log query messages.

* Fixed issue #13238 Thread naming API on Windows are now used only if
  available in KERNEL32.DLL

* Fix for issue #772: Optimized document counting for ArangoSearch views.
  Added new ArangoSearch view option 'countApproximate' for customizing view
  count strategy.

* Fix ordering of FollowerInfo lists in maintainer mode.

* Fix AR-113. Disallow non-values in the AQL geo-index-optimizer rule.

* Added SNI support for arangosh.

* Fix agency restart with mismatching compation and log indexes.

* Improve performance and memory efficiency of agency restart from persisted
  database directory.

* Added the following agency-related metrics:
  - `arangodb_agency_client_lookup_table_size`: current number of entries in
    agency client id lookup table. This gauge is available only on agent
    instances.
  - `arangodb_agency_cache_callback_count`: current number of entries in agency
    cache callbacks table. This gauge will be effective on coordinators and DB
    servers.
  - `arangodb_agency_callback_count`: current number of agency callbacks
    registered. This gauge will be effective on coordinators and DB servers.

* Fix cluster-internal replication of documents with special keys (percent
  character, which has a special meaning when used inside URLs).

* Improvements for the Pregel distributed graph processing feature:
  - during the loading/startup phase, the in-memory edge cache is now
    intentionally bypassed. The reason for this is that any edges are looked up
    exactly once, so caching them is not beneficial, but would only lead to
    cache pollution.
  - the loading/startup phase can now load multiple collections in parallel,
    whereas previously it was only loading multiple shards of the same
    collection in parallel. This change helps to reduce load times in case there
    are many collections with few shards, and on single server.
  - the loading and result storage phases code has been overhauled so that it
    runs slightly faster.
  - for Pregel runs that are based on named graphs (in contrast to explicit
    naming of the to-be-used vertex and edge collections), only those edge
    collections are considered that, according to the graph definition, can have
    connections with the vertex. This change can reduce the loading time
    substantially in case the graph contains many edge definitions.
  - the number of executed rounds for the underlying Pregel algorithm now does
    not vary for different `parallelism` values.

* Reimplement coordshort request handler. The new implementation only runs two
  DB queries without any additional requests to other coordinators, resulting in
  reduced load on the cluster. Previously this involved requests to all
  coordinators, where each of them ran two DB queries.

* When querying the list of currently running or slow AQL queries, ignore
  not-yet created databases on other coordinators.

* Fix AQL cost estimate of spliced subqueries which could lead to overly large
  numbers in the explain output of such queries.

* Add an AQL query kill check during early pruning. Fixes issue #13141.

* Fix Windows directory creation error handling.

* Added new metrics for tracking AQL queries and slow queries:
  * `arangodb_aql_query_time`: histogram with AQL query times distribution.
  * `arangodb_aql_slow_query_time`: histogram with AQL slow query times
    distribution.

* Reduce the number of dropped followers when running larger (>= 128 MB) write
  transactions.

* Remove a case in which followers were dropped unnecessarily in streaming
  transactions that replicated to the same follower.

* Added metrics for collection locks:
  - `arangodb_collection_lock_timeouts_exclusive`: Number of lock timeouts when
    trying to acquire collection exclusive locks
  - `arangodb_collection_lock_timeouts_write`: Number of lock timeouts when
    trying to acquire collection write locks
  - `arangodb_collection_lock_acquisition_micros`: Total amount of collection
    lock acquisition time [μs]
  - `arangodb_collection_lock_acquisition_time`: Total collection lock
    acquisition time histogram [s]

* Reduce lock timeout on followers to 15 seconds.
  Rationale: we should not have any locking conflicts on followers, generally.
  Any shard locking should be performed on leaders first, which will then,
  eventually replicate changes to followers. replication to followers is only
  done once the locks have been acquired on the leader(s).

* Better tracking of memory used in AQL graph traversals, COLLECT and SORT
  operations. From this version onwards, certain AQL queries can report a higher
  memory usage than in previous versions of ArangoDB. This is not because the
  queries use more memory than before, but because the memory usage tracking has
  been improved.
  A side effect of this change is that queries with a memory limit set may now
  be aborted whereas in previous versions they ran through successfully (but
  actually violated the limit). In this case it may be necessary to adjust (i.e.
  raise) query memory limits accordingly.

* Added startup option `--foxx.force-update-on-startup` to toggle waiting for
  all Foxx services in all databases to be propagated to a coordinator before it
  completes the boot sequence.
  In case the option is set to `false` (i.e. no waiting), the coordinator will
  complete the boot sequence faster, and the Foxx services will be propagated
  lazily. Until the initialization procedure has completed for the local Foxx
  apps, any request to a Foxx app will be responded to with an HTTP 503 error
  and message

    waiting for initialization of Foxx services in this database

  This can cause an unavailability window for Foxx services on coordinator
  startup for the initial requests to Foxx apps until the app propagation has
  completed.

  When not using Foxx, this option should be set to `false` to benefit from a
  faster coordinator startup.
  Deployments relying on Foxx apps being available as soon as a coordinator is
  integrated or responding should set this option to `true` (which is the
  default value).
  The option only has an effect for cluster setups.
  On single servers and in active failover mode, all Foxx apps will be available
  from the very beginning.
  Note: ArangoDB 3.6 and 3.7 introduce this option with a default value of
  `true`. ArangoDB 3.8 changes the default value to `false`.

* Changed the server-side implementation of the following internal JavaScript
  APIs to no-ops:
  * `internal.reloadAqlFunctions()`: this is a no-op function now
  * `@arangodb/actions.buildRouting()`: this is a no-op function now
  * `@arangodb/actions.routingTree`: will return an empty object
  * `@arangodb/actions.routingList`: will return an empty object

  All the above APIs were intended to be used for internal means only. These
  APIs are deprecated now and will be removed in ArangoDB v3.9.

* Fix HTTP/1.1 status response header in fuerte responses

  This change makes fuerte return the full status header, including the numeric
  status code and the status string in the `http/1.1` header of fuerte
  responses.

  Previously, the return header lacked the numeric status code, so it looked
  like
  ```
  "http/1.1" : "Ok"
  ```
  Now, with the numeric status code, the response header will look like
  ```
  "http/1.1" : "200 Ok"
  ```
  This PR also adds a protocol() method for arango client connections in order
  to check the protocol in use. The possible return values are
  - "http" for HTTP/1.1 connections
  - "http2" for HTTP/2 connections
  - "vst" for VST connections
  - "unknown" for everyhting else
  This is needed during testing, but can also be used for other purposes.

* Fixed bug in the connection pool which could prevent connection reusage under
  high load and lead to lots of new connection creations, in particular with
  TLS.

* Added more metrics around connection pool.

* Fix a potential nullptr access in AsyncAgencyComm in case there was a specific
  error when sending an agency request.

* Clean up agency change log, cluster info caches.


v3.7.5 (2020-12-09)
-------------------

* Fixed ES-662 by introducing refactored thread pool to make more efficient
  consolidation and commit routines for links of ArangoSearch views.

  Added new command line options for fine-grained ArangoSearch maintenance
  control:
  - `--arangosearch.commit-threads` - max number of ArangoSearch commit
    threads
  - `--arangosearch.commit-threads-idle` - min number of ArangoSearch
    commit threads
  - `--arangosearch.consolidation-threads` - max number of ArangoSearch
    consolidation threads
  - `--arangosearch.consolidation-threads-idle` - min number of ArangoSearch
    consolidation threads

  Deprecated the following command line options:
  - `--arangosearch.threads` - will be used in --arangosearch.commit-threads`
    and `--arangosearch.consolidation-threads` as provided value divided by 2
    if set to non-autodetect value > 0
  - `--arangosearch.threads-limit`

* Updated ArangoDB Starter to 0.14.15-1.

* Fixed agency redirect in poll API.

* Updated arangosync to 1.2.1.

* Added support for fetching the list of currently running and slow AQL queries
  from all databases at once, by adding an `all` parameter to the following
  query APIs:

  * `require("@arangodb/aql/queries").current({ all: true })`: will return the
    currently running queries from all databases, not just the currently
    selected database.
  * HTTP GET `/_api/query/current?all=true`: same, but for the HTTP REST API.
  * `require("@arangodb/aql/queries").slow({ all: true })`: will return the slow
    query history from all databases, not just the currently selected database.
  * HTTP GET `/_api/query/slow?all=true`: same, but for the HTTP REST API.
  * `require("@arangodb/aql/queries").clearSlow({ all: true })`: will clear the
    slow query history for all databases, not just the currently selected
    database.
  * HTTP DELETE `/_api/query/slow?all=true`: same, but for the HTTP REST API.

  Using the `all` parameter is only allowed when making the call inside the
  `_system` database and with superuser privileges.

* Fixed issue #12734: Accept HTTP headers into Foxx framework.

* Fix Gauge class' assignment operators.

* Clean up callback bin and empty promises in single-host-agency.

* Fix an issue where a query would not return a result when the geo index was
  used.

* Fix the activation of the agency supervision maintenance via the REST API
  `/_admin/cluster/maintenance`. This API stored a boolean value instead of an
  (expected) maintenance period end date/time string.

* Make the cancel operation safe for asynchronoulsly started JavaScript
  transactions (via HTTP POST to `/_api/transaction` with the `x-arango-async`
  header set).

* Fixed initial population of local AgencyCache values after a server restart.
  Previously the local cache was populated from the agency using a commit index
  value of 1, whereas it should have been 0 to get the full agency snapshot.

* Updated OpenSSL to 1.1.1h.

* Make the number of network I/O threads properly configurable via the startup
  option `--network.io-threads`. This option existed before, but its configured
  value was effectively clamped to a value of `1`. ArangoDB 3.7.5 thus also uses
  a default value of `1` for this option to remain compatible in terms of
  default option values.

* Fix internal issue #777: Fixed memory access while substituting stored values
  for ArangoSearch view optimization.

* Added new metric `arangodb_network_forwarded_requests` to track the number
  of requests forwarded from one coordinator to another in a load-balancing
  context.

* Added new metric `arangodb_replication_cluster_inventory_requests` to track
  the number of requests received for cluster inventories. The cluster
  inventory API is called at the beginning of a dump process or by arangosync.

* Added new AQL metrics:
  - `arangodb_aql_total_query_time_msec": Total execution time of all AQL
    queries (ms)
  - `arangodb_aql_all_query`: total number of all AQL queries

* Added new metric `arangodb_aql_total_query_time_msec` to track the combined
  runtime of AQL queries (slow queries and non-slow queries).

* Added more scheduler metrics:

  - `arangodb_scheduler_threads_started`: Total number of scheduler threads
    started
  - `arangodb_scheduler_threads_stopped`: Total number of scheduler threads
    stopped
  - `arangodb_scheduler_jobs_done`: Total number of scheduler queue jobs done
  - `arangodb_scheduler_jobs_submitted`: Total number of jobs submitted to the
    scheduler queue
  - `arangodb_scheduler_jobs_dequeued`: Total number of jobs dequeued from the
    scheduler queue
  - `arangodb_scheduler_num_working_threads`: Number of currently working
    scheduler threads

* Added startup option `--server.unavailability-queue-fill-grade`. This option
  has a consequence for the `/_admin/server/availability` API only, which is
  often called by load-balancers and other availability probing systems.
  The `/_admin/server/availability` API will now return HTTP 200 if the fill
  grade of the scheduler's queue is below the configured value, or HTTP 503 if
  the fill grade is above it. This can be used to flag a server as unavailable
  in case it is already highly loaded.
  The default value for this option is `1`, which will mean that the
  availability API will start returning HTTP 503 responses in case the scheduler
  queue is completely full. This is mostly compatible with previous versions of
  ArangoDB.
  Previously the availability API still returned HTTP 200 in this situation, but
  this can be considered a bug, because the server was effectively totally
  overloaded.
  To restore 100% compatible behavior with previous version, it is possible to
  set the option to a value of `0`, which is a special value indicating that the
  queue fill grade will not be honored.

  To prevent sending more traffic to an already overloaded server, it can be
  sensible to reduce the default value to `0.75` or even `0.5`.
  This would mean that instances with a queue longer than 75% (or 50%, resp.) of
  their maximum queue capacity would return HTTP 503 instead of HTTP 200 when
  their availability API is probed.

  nb: the default value for the scheduler queue length is 4096.

* Fixed bug with ArangoSearch views on SmartGraph edge collections which could
  contain some documents twice.
  This change removes `_to_*` local auxiliary link creation and existence within
  a view linked with a SmartGraph edge collection.

* Fixed an AQL bug that ignored PRUNE statements in OneShard setups.

* Added arangobench options:
  `--create-database` to create the test database on start
  `--duration` to run test for a duration rather than a defined count

* Fixed a deadlock between AQL write transactions and hotbackup, since in AQL
  write transactions follower transactions did not know they are follower
  transactions.

* Make the DOCUMENT AQL function eligible for running on DB servers in OneShard
  deployment mode. This allows pushing more query parts to DB servers for
  execution.

* Fix REST API endpoint PUT `/_api/collection/<collection>/recalculateCount` on
  coordinators. Coordinators sent a wrong message body to DB servers here, so
  the request could not be handled properly.

* Fixed issue #12778: fails validation if additionalProperties: false.

* Added missing exceptions catch clause for some parts of supervision and
  heartbeat threads.

* Fixed potential deadlock in cluster transactions if a transaction is returned
  that was soft-aborted by transaction garbage collection before.
  This deadlock should rarely ever occur in practice, as it can only be
  triggered once during the server shutdown sequence.

* Fix a memory leak because server internal connections were not cleaned up for
  agency communication.

* Added compile option USE_JEMALLOC_PROF to enable memory profiling.

* Fixed BTS-233 issue: Fixed invalid IndexId comparator.

* Fixed very spurious errors if the `holdReadLockCollection` replication API for
  the getting-in-sync procedure of shards was called during server shutdown.
  In this case that method could ask the transaction manager for a specific
  transaction, but wasn't returning one due to the server shutdown.

* Agency cache clears change history. This keeps the change history, introduced
  in v3.7.4, from growing in size too much.

* Bug-fix: Allow to unlink a view created on a SmartGraphEdge collection.

* If a collection (or database) is dropped during the instantiation of an AQL
  query, the setup code now aborts with an ERROR_QUERY_COLLECTION_LOCK_FAILED
  and earlier.
  Before the setup code could abort with TRI_ERROR_INTERNAL in the same case.

* Bug-fix: Creating an additional index on the edge collection of a disjoint
  SmartGraph could falsely result in an error:
  `Could not find all smart collections ...`
  This is now ruled out and indexes can be created as expected.

* Fixed issue #12248: Web UI - Added missing HTML escaping in the setup script
  section of a foxx app.

* Add parameter so `db.collection.truncate({compact: false})` will stop
  compaction from happening. Compaction may have performance impacts even if the
  truncate was invoked on nearly empty collections.

* Instead of failing to connect to INADDR_ANY refuse it as a parameter, with a
  descriptive error message for novice users (issue #12871).

* Fixed collection count which could be off after a server crash.


v3.7.4 (2020-10-16)
-------------------

* Data definition reconciliation in cluster has been modified
  extensively to greatly accelerate the creation of 1000s of
  databases through following means:
  - AgencyCache offers change sets API based on Raft index.
  - ClusterInfo caches are only updated using change sets.
  - Maintenance uses local as well as agency change sets to limit
    the scope of every runtime to these change sets.


v3.7.3 (2020-10-14)
-------------------

* Added the following metrics for synchronous replication in the cluster:

  - `arangodb_refused_followers_count`: Number of times a shard leader received
    a refusal answer from a follower during synchronous replication.
  - `arangodb_sync_wrong_checksum`: Number of times a mismatching shard
    checksum was detected when syncing shards. In case this happens, a resync
    will be triggered for the shard.

* Fixed handling of failedLeaderJob. In case of a plan modification, that
  removes a server from the plan, e.g. reduce replication factor. Directly
  followed by a failure of the current shard leader, would reinsert the just
  removed server in the plan, which is undesired, we first need to have a full
  "desync" cycle on this server to be reusable in the plan again.

* Make sure the optimizer doesn't pick another index than the TTL index itself
  while fulfilling the expiry of TTL.

* Added optional verbose logging for agency write operations. This logging is
  configurable by using the new log topic "agencystore".

  The following log levels can be used for for the "agencystore" log topic to
  log writes to the agency:
  - DEBUG: will log all writes on the leader
  - TRACE: will log all writes on both leaders and followers
  The default log level for the "agencystore" log topic is WARN, meaning no
  agency writes will be logged.
  Turning on this logging can be used for auditing and debugging, but it is not
  recommended in the general case, as it can lead to large amounts of data being
  logged, which can have a performance impact and will lead to higher disk space
  usage.

* Print image base address and CPU context (if available) in crash handler
  messages.

* Added configuration option `--query.tracking-slow-queries` to decide whether
  slow queries are tracked extra.

* Added configuration option `--query.tracking-with-querystring` to decide
  whether the query string is shown in the slow query log and the list of
  currently running queries. The option is true by default.

  When turned off, querystrings in the slow query log and the list of currently
  running queries are just shown as "<hidden>".

* Added configuration option `--query.tracking-with-datasources` to toggle
  whether the names of data sources used by queries are shown in the slow query
  log and the list of currently running queries. The option is false by default.
  When turned on, the names of data sources used by the query will be shown in
  the slow query log and the list of currently running queries.

* Fixed handling of failoverCandidates. Sometimes, a server can still be a
  failoverCandidate even though it has been taken out of the Plan. With this
  fix, such a server is quickly taken out of failoverCandidates and it can never
  be re-added to the Plan before this has happened.

* Fix #12693: SORT inside a subquery could sometimes swallow part of its input
  when it crossed boundaries of internal row batches.

* Fixed issue BTS-212: Web UI doesn't let to make partial view update and
  partial view update should be audited (also reported as ES-700).
  Fixed link definition comparison logic: equality wasn`t properly detected and
  led to link recreation.

* Added configuration option `--rocksdb.sync-delay-threshold`.
  This option can be used to track if any RocksDB WAL sync operation is
  delayed by more than the configured value (in milliseconds). The intention
  is to get aware of severely delayed WAL sync operations.

* Add database, shard name and error information to several shard-related log
  messages.

* Display shard names of a collection in the web interface when in the details
  view of the collection.

* Added HTTP requests metrics for tracking the number of superuser and normal
  user requests separately:

  - `arangodb_http_request_statistics_superuser_requests`: Total number of HTTP
    requests executed by superuser/JWT
  - `arangodb_http_request_statistics_user_requests`: Total number of HTTP
    requests executed by clients

* Fixed a bug in handling of followers which refuse to replicate operations.
  In the case that the follower has simply been dropped in the meantime, we now
  avoid an error reported by the shard leader.

* Fix a performance regression when a LIMIT is combined with a COLLECT WITH
  COUNT INTO. Reported in ES-692.

* Fix REST handler GET /_admin/status when called with URL parameter value
  `overview=true`. For generating the `hash` attribute in the response, the
  current Plan was retrieved and analyzed. Due to a change in the internal Plan
  format the REST handler code failed to pick up the number of servers, which
  resulted in the REST handler returning HTTP 500 in cluster mode.

* Use rclone built from v1.51.0 source with go1.15.2 instead of prebuilt
  v1.51.0 release.

* Fixed a bug in AQL COLLECT with OPTIONS { "hash" } that led to a quadratic
  runtime in the number of output rows.

* Added startup option `--database.old-system-collections` to toggle automatic
  creation of system collections `_modules` and `_fishbowl`, along with their
  internal usage. These collections are useful only in very few cases, so it
  is normally not worth to create them in all databases.
  The `_modules` collection is only used to register custom JavaScript modules,
  for which there exists no API, and `_fishbowl` is used to store the temporary
  list of Foxx apps retrieved from the Github Foxx store.
  If the option value is `false` (which is the default from v3.8 onwards, but
  for v3.7 the default value is `true` for downwards-compatibility), the two
  collections will not be created for any new database. The `_fishbowl`
  collection will still be created dynamically when needed. If the option value
  is `true` (the default value in v3.7), the collections will be created
  regularly as before.
  The default value for the option is going to change to `false` in v3.8,
  meaning the collections will not be created anymore there by default.

  Any functionality related to the `_modules` system collection is deprecated
  and will be removed in ArangoDB v3.9.

  Two side effects of turning this option off are:
  * there will no be iteration over all databases at server startup just to
    check the contents of all `_modules` collections.
  * less collections/shards will be around for deployments that create a large
    number of databases.
  Already existing `_modules` and `_fishbowl` system collections will not be
  modified by this PR, even though they will likely be empty and unused.

* Don't iterate over all databases at server startup in order to initialize the
  routing information. This is not necessary, as the routing information is
  global and not tied to a specific database.

* Fixed a possible crash during instantiation of an AQL graph traversal.
  Reported in #12597.

* Added safeguards against using V8 internally in environments that have
  JavaScript turned off via the `--javascript.enabled false` option.

* Make scheduler properly count down the number of working threads in case an
  exception happens in a worker thread.

* Turn off upgrade checks in arangod in alpha/beta/preview Enterprise builds,
  too.
  Previously it was already turned off in arangod for Enterprise builds already,
  but only for stable releases and not preview releases.

* Fixed and extended LDAP log messages.

* Added LDAP_OFF if referrals and restart are false.

* If LDAP search fails, also retry (update to given number of retries).

* Fixed infinite reload of the login window after logout of an LDAP user.

* Make the reboot tracker catch failed coordinators, too. Previously the reboot
  tracker was invoked only when a DB server failed or was restarted, and when a
  coordinator was restarted. Now it will also act if a coordinator just fails
  (without restart).

* Added scheduler thread creation/destruction metrics:

  - `arangodb_scheduler_threads_started`: Number of scheduler threads started
  - `arangodb_scheduler_threads_stopped`: Number of scheduler threads stopped

* Added startup option `--query.max-runtime` to limit the maximum runtime of all
  AQL queries to a specified threshold value (in seconds). By default, the
  threshold is 0, meaning that the runtime of AQL queries is not limited.
  Setting it to any positive value will restrict the runtime of all AQL queries
  unless it is overwritten in the per-query "maxRuntime" query option.
  Please note that setting this option will affect *all* queries in all
  databases, and also queries issues for administration and database-internal
  purposes.
  If a query exceeds the configured runtime, it will be killed on the next
  occasion when the query checks its own status. Killing is best effort, so it
  is not guaranteed that a query will no longer than exactly the configured
  amount of time.

* Ensure that the argument to an AQL OPTIONS clause is always an object which
  does not contain any dynamic (run-time) values. Previously, this was only
  enforced for traversal options and options for data-modification queries. This
  change extends the check to all occurrences of OPTIONS.

* Added `details` option to figures command of a collection:
  `collection.figures(details)`

  Setting `details` to `true` will return extended storage engine-specific
  details to the figures. The details are intended for debugging ArangoDB itself
  and their format is subject to change. There is not much use in using the
  details from a client application.
  By default, `details` is set to `false`, so no details are returned and the
  behavior is identical to previous versions of ArangoDB.

* Implement RebootTracker usage for AQL queries in case of coordinator restarts
  or failures. This will clean up the rest of an AQL query on dbservers more
  quickly and in particular release locks faster.

* Serialize maintenance actions for each shard. This addresses lost document
  problems found in chaos testing.

* Enforce a maximum result register usage limit in AQL queries. In an AQL query,
  every user-defined or internal (unnamed) variable will need a register to
  store results in.

  AQL queries that use more result registers than allowed (currently 1000) will
  now abort deterministically during the planning stage with error 32
  (`resource limit exceeded`) and the error message
  "too many registers (1000) needed for AQL query".

  Before this fix, an AQL query that used more than 1000 result registers
  crashed the server when assertions were turned on, and the behavior was
  undefined when assertions were turned off.

* Fixed some cases where subqueries in PRUNE did not result in a parse error,
  but either in an incomprehensible error (in 3.7), or undefined behavior
  during execution (pre 3.7).

* Fixed an issue with audit logging misreporting some document requests as
  internal instead of logging the proper request information.

* Add attributes `database` and `user` when tracking current and slow AQL
  queries.
  `database` contains the name of the database the query is/was running in,
  `user` contains the name of the user that started the query.
  These attributes will be returned in addition when calling the APIs for
  current and slow query inspection:
  * GET `/_api/query/current` and `require("arangodb/aql/queries").current()`
  * GET `/_api/query/slow` and `require("arangodb/aql/queries").slow()`

  The "slow query" log message has also been augmented to contain the database
  name and the user name.

  The `user` attribute is now also displayed in the web interface in the
  "Running queries" and "Slow queries" views.

* Added metrics for V8 contexts usage:
  * `arangodb_v8_context_alive`: number of V8 contexts currently alive.
  * `arangodb_v8_context_busy`: number of V8 contexts currently busy.
  * `arangodb_v8_context_dirty`: number of V8 contexts currently dirty.
  * `arangodb_v8_context_free`: number of V8 contexts currently free.
  * `arangodb_v8_context_max`: maximum number of concurrent V8 contexts.
  * `arangodb_v8_context_min`: minimum number of concurrent V8 contexts.

* Updated arangosync to 0.7.11.

* Make followers in active failover run a compaction after they process a
  truncate operation and the truncate removed more than 4k documents. This can
  help to reclaim disk space on the follower earlier than without running the
  truncate.

* The REST API PUT `/_api/collection/<name>/truncate` will now also run a
  compaction if the truncation affected more than 4k documents. This may add
  extra latency to the truncate operation, but can help to reclaim disk space
  earlier.

* Added REST API PUT `/_admin/compact` for compacting the entire database data.
  This endpoint can be used to reclaim disk space after substantial data
  deletions have taken place. The command is also exposed via the JavaScript API
  as `db._compact();`.

  This command can cause a full rewrite of all data in all databases, which may
  take very long for large databases. It should thus only be used with care and
  only when additional I/O load can be tolerated for a prolonged time.

  This command requires superuser access and is only available for the RocksDB
  storage engine.

* Don't allow creation of smart satellite graphs or collections (i.e. using
  `"isSmart":true` together with `"replicationFactor":"satellite"` when creating
  graphs or collections. This combination of parameters makes no sense, so that
  the server will now respond with "bad parameter" and an HTTP status code of
  HTTP 400 ("Bad request").

* Add exit code for ICU database loading startup errors.

* Fixed issue #12507: SegFault when using an AQL for loop through edges.

* Make the `IS_IPV4` AQL function behave identical on macOS as on other
  platforms. It previously allowed leading zeros in octets on macOS, whereas on
  other platforms they were disallowed.
  Now this is disallowed on macOS as well.

* Added new metric "arangodb_aql_slow_query" for slow AQL queries, so this can
  be monitored more easily.

* Added new metric "arangodb_scheduler_queue_length" for the scheduler's
  internal queue length.

* Added new metric "arangodb_scheduler_queue_full_failures" for tracking cases
  of a full scheduler queue and dropping requests.

* Added new metrics for the number of V8 contexts dynamically created and
  destroyed ("arangodb_v8_context_created" and "arangodb_v8_context_destroyed")
  and for the number of times a V8 context was entered and left
  ("arangodb_v8_context_entered" and "arangodb_v8_context_exited"). There is
  also a new metric for tracking the cases when a V8 context cannot be
  successfully acquired and an operation is not performed
  ("arangodb_v8_context_enter_failures").

* Added extra info to "queue full" and "giving up waiting for unused v8 context"
  log messages.

* Request to the `/_admin/statistics` API now processed via the CLIENT_FAST
  lane.
  Previously they were handled in the CLIENT_SLOW lane, meaning that monitoring
  requests using that API didn't get through when the queue was rather full.

* Introduce an internal high-water mark for the maximum row number that was
  written to in an AqlItemBlock. Using this number several operations on the
  whole block, such as cleaning up or copying can be made more efficient when
  run on only partially filled blocks.

* Fixed issue BTS-169: cost estimation for LIMIT nodes showed wrong number of
  estimated items.


v3.7.2.2 (2020-10-07)
---------------------

* Fixed issue ES-664: SEARCH vs FILTER lookup performance.
  Consolidation functionality for ArangoSearch view links was able to hit non-
  mergable enormous amount of segments due to improper scheduling logic.

* Fix for issue BTS-183: added pending operations purging before ArangoSearch
  index truncation.

* Fixed: More cases in AQL can now react to a query being killed, so reaction
  time to query abortion is now shortened. This was a regression in comparison
  to 3.6 series.

* Fixed internal issue #741: STARTS_WITH fails to accept 'array' as variable.

* Fixed internal issue #738: PHRASE doesn't accept a reference to an array of
  arguments.

* Fixed internal issue #747: fixed possible dangling open files in ArangoSearch
  index after remove operations.


v3.7.2.1 (2020-09-02)
---------------------

* Add option `--rocksdb.max-write-buffer-size-to-maintain` with default of 0.
  This configures how much memory RocksDB is allowed to use for immutable
  flushed memtables/write-buffers. The default of 0 will usually be good for all
  purposes and restores the 3.6 memory usage for write-buffers.


v3.7.2 (2020-08-21)
-------------------

* Fixed internal issue #744: LIMIT with only offset and constrained heap
  optimization will use estimation value for ArangoSearch views.

* Make UPSERT statement with collection bind parameter behave identical to its
  non-bind parameter counterpart.

  For example, the query

      FOR d IN ["key_1", "key_2", "key_3", "key_1"]
        UPSERT d INSERT d UPDATE d IN @@collection

  would fail with "unique constraint violation" when used with a collection bind
  parameter, but the equivalent query

      FOR d IN ["key_1", "key_2", "key_3", "key_1"]
        UPSERT d INSERT d UPDATE d IN collectionName

  with a hard-coded collection name would succeed. This is now fixed so both
  queries have the same behavior (no failure) in single server.

* Updated arangosync to 0.7.10.

* Fixed internal issue #742: Add tick storage in index meta payload for
  ArangoSearch view links after collection truncate operation.

* Fixed issue #12304: insert in transaction causing
  com.arangodb.ArangoDBException: Response: 500, Error: 4 - Builder value not
  yet sealed.

  This happened when too deeply-nested documents (more than 63 levels of
  nesting) were inserted. While indefinite nesting is still not supported, the
  error message has been corrected from the internal HTTP 500 error "Builder
  value not yet sealed" to the correct HTTP 400 "Bad parameter".

* Fixed: During a move-shard job which moves the leader there is a
  situation in which the old owner of a shard can reclaim ownership
  (after having resigned already), with a small race where it allows to
  write documents only locally, but then continue the move-shard to a
  server without those documents. An additional bug in the MoveShard
  Supervision job would then leave the shard in a bad configuration
  with a resigned leader permanently in charge.

* Fixed a problem with potentially lost updates because a failover could happen
  at a wrong time or a restarted leader could come back at an unlucky time.

* Fixed BTS-167: Lingering Queries - Canceled from the UI.
  This fixes queries not vanishing from the list of running queries in the web
  UI in case the query was canceled using the "Cancel" button in web UI's query
  editor.

* Fixed bad behavior in agency supervision in some corner cases involving
  already resigned leaders in Current.

* Fixed issue ES-664: the optimizer rule `inline-subqueries` must not pull out
  subqueries that contain a COLLECT statement if the subquery is itself called
  from within a loop. Otherwise the COLLECT will be applied to the values in the
  outer FOR loop, which can produce a different result.

* Fixed that dropping a vanished follower works again. An exception response
  to the replication request is now handled properly.

* Make AQL user-defined functions (UDFs) work in a cluster in case the UDF runs
  an AQL query inside its own function code (BTS-159).

* Fixed hotbackup S3 credentials validation and error reporting for upload and
  download.

* Fixed a blockage on hotbackup when writes are happening concurrently, since
  followers could no longer replicate leader transactions.

* Fix: The 'sorted' COLLECT variant would return undefined instead of null when
  grouping by a null value.

* Fix: writeConcern is now honored correctly (ES-655).

* Fixed internal issue #739: ArangoSearch filter volatility takes into account
  calculation nodes dependencies.

* Fixed OASIS-278 issue: Added proper sort/calc nodes cleanup for late
  materialzation after OneShard optimization.

* Slightly improve the performance of cluster DDL maintenance operations.

* Added AQL functions `IS_IPV4`, `IPV4_TO_NUMBER`, `IPV4_FROM_NUMBER` for IPv4
  address checks and conversions.

* Added AQL function `PRODUCT` to calculate the product of an array.

* Improve performance of internal cluster Plan and Current reload operations.

* Fixed issue #12349: arangosh compact Arangoerror 404.

* Improve performance of many non-subquery AQL queries, by optimizing away some
  storage overhead for subquery context data.

* Always fetch data for /_api/cluster/agency-dump from leader of the agency.
  Add option "redirectToLeader=true" to internal /_api/agency/state API.

* Fixed regression with view use-after-create in cluster (BTS-137).

* Slightly improved the performance of some k-shortest-path queries.


v3.7.1 (2020-08-07)
-------------------

* Fixed issue #12297: ArangoDB 3.6.5 Swagger Error?
  This issue caused the Swagger UI for displaying the APIs of user-defined Foxx
  services to remain invisible in the web UI, because of a JavaScript exception.
  This PR fixes the JavaScript exception, so the services API is displayed
  again properly.

* Fixed issue #12248: Web UI on 3.6.5: 404 error on adding new index.
  This issue caused "404: not found" errors when creating indexes via the web
  UI. The indexes were created successfully despite the error message popping
  up. This fix removes the misleading unconditional error message.


v3.7.1-rc.1 (2020-07-24)
------------------------

* Added startup option `--rocksdb.encryption-key-rotation` to
  activate/deactivate the encryption key rotation REST API. The API is disabled
  by default.

* Add internal caching for LogicalCollection objects inside
  ClusterInfo::loadPlan.

  This allows avoiding the recreation of LogicalCollection objects that did not
  change from one loadPlan run to the next. It reduces CPU usage considerably on
  both Coordinators and DB-servers.

* Fixed reading analyzers revisions for freshly updated cluster.

* Fixed undefined behavior in AQL COLLECT with multiple group variables (issue
  #12267).
  If you are grouping on "large" values that occur multiple times in different
  groups, and two different groups with the same large value are written to
  different batches in the output then the memory could be invalid.
  e.g. the following query is affected:
  ```
  FOR d IN documents
  FOR batchSizeFiller IN 1..1001
  COLLECT largeVal = d.largeValue, t = batchSizeFiller
  RETURN 1
  ```

* Revive faster out-of-range comparator for secondary index scans that do a full
  collection index scan for index types "hash", "skiplist", "persistent".

* Conflict error codes (1200) will now use the proper error message instead of a
  generic and misleading "precondition failed".

* Improve performance of agency cache by not copying the hierarchical Node
  tree result for serialization, but serializing it directly.

* Turn off maintenance threads on Coordinators, as they are not needed there.

* Fixed internal issue #733: Primary sort compression in ArangoSearch views now
  used properly.

* Change HTTP response code for error 1450 ("too many shards") from HTTP 500 to
  HTTP 400, as this is clearly a client error.

* Fix spurious lock timeout errors when restoring collections.

* Make sure cluster statistics in web UI work in case a coordinator is down.

* Updated arangosync to 0.7.8.

* Fixed a race between a new request and the keepAlive timeout.

* Fixed hotbackup upload and download with encryption at rest key indirection.

* Fixed crash in cleanup of parallel traversal queries.

* Added cluster metrics `arangodb_load_plan_accum_runtime_msec` and
  `arangodb_load_current_accum_runtime_msec` to track the total time spent in
  `loadPlan()` and `loadCurrent()` operations.

* Fixed wrong reporting of failures in all maintenace failure counter metrics
  (`arangodb_maintenance_action_failure_counter`). Previously, each successful
  maintenance operation was reported as a failure, so the failure counters
  actually were counters for the number of successful maintenance actions.

* Adjusted the scale of the `arangodb_maintenance_action_queue_time_msec` to
  cover a more useful range.

* The filter executor will now overfetch data again if followed by a limit, same
  as in 3.6 series.
  The following queries are effected:
  ```
  something
  FILTER a == b
  LIMIT 10
  ```
  something will now be asked for a full batch, instead of only 10 documents.

* In rare cases SmartBFS could use a wrong index for looking up edges. This is
  fixed now.

* The internally used JS-based ClusterComm send request function can now again
  use JSON, and does not require VelocyPack anymore. This fixes an issue with
  Foxx-App management (install, updated, remove) got delayed in a sharded
  environment, all servers do get all apps eventually, now the fast-path will
  work again.

* Fixed a rare race in Agents, if the leader is rebooted quickly there is a
  chance that it is still assumed to be the leader, but delivers a state shortly
  in the past.

* Keep the list of last-acknowledged entires in Agency more consistent.
  During leadership take-over it was possible to get into a situation that the
  new leader does not successfully report the agency config, which was
  eventually fixed by the Agent itself. Now this situation is impossible.

* Fixed a race in the ConnectionPool which could lease out a connection that got
  its idle timeout after the lease was completed. This could lead to sporadic
  network failures in TLS and to inefficiencies with TCP.

* Fixed restoring a SmartGraph into a database that already contains that same
  graph.
  The use case is restoring a SmartGraph from backup, apply some modifications,
  which are undesired, and then resetting it to the restored state, without
  dropping the database. One way to achieve this is to use arangorestore with
  the `overwrite` option on the same dataset, effectively resetting the
  SmartGraph to the original state. Without this fix, the workaround for is to
  either drop the graph (or the database) before the restore call, yielding an
  identical result.

* Fixed potential garbled output in syslog log output for the program name.

* Fixed infinite recursion when printing error messages about invalid logger
  configuration on startup.

* Fixed sporadic use-after-free ASan issue in logger shutdown.

* Added missing state rollback for failed attempt-based write locking of spin
  locks.

* Disable internal network protocol switch for cluster-internal communication,
  and hard-code the internal communication protocol to HTTP.

* Added vertex collection validation in case of a SmartGraph edge definition
  update.

* Fix cluster-internal request forwarding for VST requests that do not have any
  Content-Type header set. Such requests could have been caused by the Java
  driver (ES-635).

* Fixed issue OASIS-252: Hotbackup agency locks without clientId.

* Fixed internal-issue #726: added restore handling for custom analyzers.

* Fixed a potential agency crash if trace logging is on.

* Network layer now reports connection setup issues in more cases this replaces
  some INTERNAL_ERROR reports by more precise errors, those are only reached
  during failover scenarios.

* Fixed a document parsing bug in the Web UI. This issue occured in the
  document list view in case a document had an attribute called `length`.
  The result was an incorrect representation of the document preview.

* Re-enable access to GET `/_admin/cluster/numberOfServers` for all users by
  default. Requests to PUT `/_admin/cluster/numberOfServers` require admin
  user privileges. This restores the pre-3.7 behavior.
  In contrast to pre-3.7 behavior, the `--server.harden` option now can be
  used to restrict access to GET `/_admin/cluster/numberOfServers` to admin
  users, too. This can be used to lock down this API for non-admin users
  entirely.

* Fixed crash in HTTP2 implementation.

* Improve readability of running and slow queries in web UI by properly left-
  aligning the query strings.

* ClusterInfo will wait for syncer threads to shutdown.

* Correct some log entries.

* Allow changing collection properties for smart edge collections as well.
  Previously, collection property changes for smart edge collections were not
  propagated.

* Fixed BTS-110: Fulltext index with minLength <= 0 not allowed.

* Web UI: Removed unnecessary menubar entry in case of database node inspection.

* Adjust arangodump integration test to desired behavior, and make sure
  arangodump behaves as specified when invoking it with non-existing
  collections.

* Disallow using V8 dependent functions in SEARCH statement.

* Remove superflous `%>` output in the UI modal dialog in case the JSON editor
  was embedded.

* The Web UI now displays an error message inside the node information view in
  case the user has no access to retrieve the neccessary information.

* The `_from` and `_to` attributes of an edge document can now be edited from
  within the UI.

* Fixed a misleading error message in AQL.

* Fixed issue ES-609: "Transaction already in use" error when running
  transaction.
  Added option `--transaction.streaming-lock-timeout` to control the timeout in
  seconds in case of parallel access to a streaming transaction.

* Fixed internal issue BTS-107, offset over the main query passed through a
  subquery which has modification access to shards could yield incorrect
  results, if shards are large enough and skipping was large enough, both to
  overflow the batch-size in AQL for each shard individually.

  The following query would be affected:

  FOR x IN 1..100
  LET sub = (
    FOR y IN Collection
    UPDATE y WITH {updated: true} IN Collection
    RETURN new
  )
  LIMIT 5, 10
  REUTRN {x,sub}

  If a shard in Collection, has enough entries to fill a batch, the second shard
  could run into the issue that it actually does not skip the first 5 main
  queries, but reports their results in addition. This has the negative side
  effect that merging the subqueries back together was off.

* Fix undistribute-remove-after-enum-coll which would allow calculations to be
  pushed to a DBServer which are not allowed to run there.

* Removed the edge id's hover styling in case of embedded document editor in the
  Web UI as this functionality is disabled. This was misleading because the
  elements are not clickable.

* Internal issue BTS-71: Fixed error handling regarding communication with the
  agency. This could in a specific case cause collection creation in a cluster
  report success when it failed.

* Internal issue BTS-71: In a cluster, for collections in creation, suspend
  supervision jobs concerning replication factor until creation is completed.
  Previously, it could cause collection creation to fail (e.g. when a server
  failed during creation), even when it didn't have to.

* Internal issue BTS-71: Added a precondition to prevent creating a collection
  with an invalid `distributeShardsLike` property.

* In case a document got requested via the UI of a collection which does not
  exist, the UI now properly displays an error view instead of having a bad
  display state.

* Returned `AQL_WARNING()` to emit warnings from UDFs.

* Fixed internal issue #725: Added analyzers revision for _system database in
  queries.

* The Web UI is not disabling the query import button after file upload takes
  place.

* In case of a graph deletion failure, the Web UI displays now the correct
  error message.

* Privatized load plan / current in cluster info and cleanup following agency
  cache implementation.

* Allow removal of existing schemas by saving a schema of either `null` or `{}`
  (empty object). Using an empty string as schema will produce  an error in the
  web interface and will not remove the schema.

  The change also adjusts the behavior for the SCHEMA_VALIDATE AQL function in
  case the first parameter was no document/object. In this case, the function
  will now return nulland register a warning in the query, so the user can
  handle it.

* The Web UI is now reporting errors properly in case of editing ArangoSearch
  Views with invalid properties.

* Fixed bug in IResearchViewExecutor that lead to only up to 1000 rows being
  produced.

* Fixed an error scenario where a call could miss count skip.
  It was triggered in the case of Gather in Cluster, if we skipped over a full
  shard, and the shard did actually skip, but there are more documents to skip
  on another shard.

* Fixed that the hotbackup agency lock is released under all circumstances
  using scope guards. This addresses a rare case in which the lock was left
  behind.

* Allow restoring collections from v3.3.0 with their all-numeric collection GUID
  values, by creating a new, unambiguous collection GUID for them.
  v3.3.0 had a bug because it created all-numeric GUID values, which can be
  confused with numeric collection ids in lookups. v3.3.1 already changed the
  GUID routine to produce something non-numeric already, but collections created
  with v3.3.0 can still have an ambiguous GUID. This fix adjusts the restore
  routine to drop such GUID values, so it only changes something if v3.3.0
  collections are dumped, dropped on the server and then restored with the
  flawed GUIDs.

* Fixed hotbackup agency lock cleanup procedure.

* The Web UI's replication view is now checking the replication state
  automatically without the need of a manual reload.

* Changing the current users profile icon in the Web UI now renders the new icon
  and data directly without the need of a full UI browser reload.

* Fixed cluster behavior with HotBackup and non-existing backups on DB-Servers.

* Fixed that, when performing a graph AQL query while a (graceful) failover for
  the leader of the system collections is in progress, ArangoDB would report a
  "Graph not found error".
  The real error, namely that an agency transaction failed, was swallowed in
  the graph lookup code due to a wrong error code being used from Fuerte.
  We now generate a more appropriate 503 - Service Unavailable error.

* Fixed bad behavior that led to unnecessary additional revision tree rebuilding
  on server restart.

* Only advance shard version after follower is reported in sync in agency.

* Disabled new, potentially unsafe revision-based storage format.

* Allow for faster cluster shutdown. This should reduce the number of shutdown
  hangers in case the agents are stopped already and then coordinators or
  DB-Servers are shut down.

* Fixed issue ES-598. Web UI now shows correct permissions in case wildcard
  database level and wildcard collection level permissions are both set.

* Fixed non-deterministic test failure in Pregel WCC test.

* Add a recovery test to check that there are no warnings at server start after
  a graceful shutdown.

* Fixed unintentional connection re-use for cluster-internal communications.

* Fixed problem with newer replication protocol and ArangoSearch which could
  lead to server crashes during normal operation.

* Allow AQL queries on DB-Servers again. This is not an official supported
  feature, but is sometimes used for debugging. Previous changes made it
  impossible to run a query on a local shard.

* Fix restoring old arangodumps from ArangoDB 3.3 and before, which had index
  information stored in slightly different places in the dump.

* Fix internal test helper function `removeCost` to really remove costs.

* Fix invalid calls to `AgencyCommResult::slice()` method, which must only be
  made in case of an agency result was retrieved successfully. In case the call
  to the agency was not successful, `slice()` must not be called on it. This
  change makes sure this invariant holds true.

* Fix potential AQL query shutdown hanger.

* Modified the exception handling in the RestHandler. Asynchronous communication
  could lead to less detailed failure information.

* Use smarter default value preset in web UI for replication factor in case
  there are constraints established for the replication factor by the startup
  options `--cluster.min-replication-factor` and
  `--cluster.max-replication-factor`.

* Added permissions check before trying to read data from `_analyzers`
  collection.
  If these permissions are not there, no load is performed (user can not use
  analyzers from database anyway).

* Make the reboot tracker catch a failed server and permanently removed servers.
  This allows other servers in the cluster to move on more quickly, when a
  server fails and does not immediately come back.

* Added WCC pregel algorithm for weakly connected components.

* Display "geoJson" value for geo indexes in index overview in web UI.

* Allow specifiying arbitrary vertex attributes when creating a vertex via the
  graph viewer in the web UI. This is necessary for collections that have a
  schema and that require new vertices to have certain attributes/values.

* Pass through AQL query errors to UI graph viewer without turning them into
  "Internal Server Error"s.

* Augment nodes view in web UI with node id and full datetime of startup.

* Fixed "Create Debug Package" in web UI in cluster mode.

* Pick up selected "write concern" value entered in web UI when creating a
  collection in the cluster. Previously, the "write concern" value was not
  picked up correctly and collections were created with the default write
  concern.

* Fixed the creation of edges in the web UI, which did not work due to a
  JavaScript error.

* Fixed velocypack validator for proper check of keys.

* Handle missing "shardingStrategy" in Plan agency entries for collections.
  "shardingStrategy" entries can be missing for collections created with
  ArangoDB clusters in version 3.3 or earlier. A missing "shardingStrategy"
  entry can cause issues for certain queries that need to check the shard
  distribution on the DB-Servers.
  This change transparently handles missing "shardingStrategy" entries.


v3.7.1-beta.1 (2020-06-07)
--------------------------

* Fixed issue #8941: `frontendConfig.basePath` is duplicated for users API call

* Taken collection dropping from fast track in maintenance. This avoids blocking
  fast track maintenance threads when a shard cannot immediately be dropped
  because of some pending lock.

* Updated ArangoDB Starter to 0.14.15.

* Fixed a spuriously occuring shutdown deadlock between AgencyCallbackRegistry
  and AgencyCache.

* Fixed `"Plan" is not an object in agency` error messages when restarting
  DB-Servers that contained ArangoSearch Views with links.

* Fixed misordering of skipped number of rows report. Only triggered if you do a
  modification on a subquery nesting level 2 or more, e.g.:
  ```
    LET s1 = (
      LET s2 = (
        INSERT {} INTO collection
      )
      COLLECT WITH COUNT INTO c
      RETURN c
    )
  ```
  Here the c would be off, or a count on main query would be off.

* Fixed crash in execution of non-spliced subqueries if remainder of subquery is
  skipped.

* Added missing mutex to ConnectionPool::cancelConnections().

* Foxx API now respects "idiomatic" flag being explicitly set to false.

* Made modification subqueries non-passthrough. The passthrough logic only works
  if exactly as many output rows are produced as input rows are injected.
  If the subquery with modification is skipped however this API is violated, we
  perform the subquery with the modification, but then discard the input if
  skipped).

* Optimized Web UI styling for ArangoSearch Views when in read-only mode.

* Fixed a bug in the internal `recalculateCount()` method for adjusting
  collection counts. The bug could have led to count adjustments being
  applied with the wrong sign.

* Removed external file loading in a used 3rd party library (Web UI).

* More substantial metrics in agency.

* arangodump and arangorestore will now fail when using the `--collection`
  option and none of the specified collections actually exist in the database
  (on dump) or in the dump to restore (on restore). In case some of the
  specified collections exist, arangodump/restore will issue warnings about the
  invalid collections, but will continue to work for the valid collections.

* Improved network send request for more robustness.

* Added multiple RocksDB configuration options to arangod:
  * `--rocksdb.cache-index-and-filter-blocks` to make the RocksDB block cache
    quota also include RocksDB memtable sizes.
  * `--rocksdb.cache-index-and-filter-blocks-with-high-priority` to use cache
    index and filter blocks with high priority making index and filter blocks be
    less likely to be evicted than data blocks.
  * `--rocksdb.pin-l0-filter-and-index-blocks-in-cache` make filter and index
    blocks be pinned and only evicted from cache when the table reader is freed.
  * `--rocksdb.pin-top-level-index-and-filter` make the top-level index of
    partitioned filter and index blocks pinned and only be evicted from cache
    when the table reader is freed.

* Don't move potentially expensive AQL function calls into loops in the
  `remove-unnecessary-calculations-rule`.

  For example, in the query

      LET x = NOOPT(1..100)
      LET values = SORTED(x)
      FOR j IN 1..100
        FILTER j IN values
        RETURN j

  there is only one use of the `values` variable. So the optimizer can remove
  that variable and replace the filter condition with `FILTER j IN SORTED(x)`.
  However, that would move the potentially expensive function call into the
  inner loop, which could be a pessimization.

  Now the optimizer will not move the calculation of values into the loop when
  it merges calculations in the `remove-unnecessary-calculations` optimizer
  rule.

* Fixed modification executors inside of a subquery, where the subquery decided
  to fetch all rows from upstream first and the amount of rows is higher then
  the batch size.

* Fixed reporting of skipped number of documents if we have a LIMIT x, 0 right
  after the modification.

* Added exceptions catching in agency callbacks.

* Fixed issue #11104 case with complex condition on second branch of OR
  operator.

* Fixed bad behavior when dropping followers. A follower should be dropped
  immediately when it is officially FAILED, not only after a longish timeout.

* Fixed a bug in CollectionNameResolver which could lead to an extended busy
  spin on a core when a collection was dropped, but documents of it still
  remained in the WAL.

* Fixed return value of fuerte::H1Connection in case of timeout.

* Usability improvements for arangodump and arangorestore:
  - when a dump is restricted to one or multiple collections using the
    `--collection` option of arangodump, warnings are issued for all specified
    collections that are not present in the database. Previously there were
    no warnings, so missing a collection by misspelling its name could easily
    go unnoticed.
  - when a restore is restricted to one or multiple collections using the
    `--collection` option of arangorestore, warnings are issued for all
    specified collections that are not present in the dump. Previously there
    were no warnings, so missing a collection by misspelling its name could
    easily go unnoticed.
  - when a dump was taken using the `--overwrite` option, there was no check
    that validated whether the encryption mode used in the existing dump
    directory was the same as the requested encryption mode. This could have led
    to dump directories with both encrypted and unencrypted files. This was only
    the case when using `--overwrite true`, which is not the default.
  - when a restore was performed using the `--encryption.keyfile` option, there
    was no check whether the to-be-restored files were actually encrypted.
    Now this check is enforced and arangorestore will bail out with an error if
    the requested encryption mode for restore is not the same as for the stored
    dump files.

* Fixed traversal issue: If you have a traversal with different minDepth and
  maxDepth values and filter on path elements that are larger then minDepth, in
  a way that a shorter path would match the condition, the shorter paths would
  in some cases not be returned, even if they are valid. e.g.

  FOR v, e, p IN 1..3 OUTBOUND @start Graph "myGraph"
    FILTER p.vertices[2].label != "foo"
    RETURN p

  In the above query, a path of length 1 would be valid. There p.vertices[2]
  does not exist => Evaluated to `null`. `null`.label is again evaluated to
  `null` => `null != "foo"` is true, so the path is valid.

* Fixed traversal issue: If you have a filter on the path that is based on a
  variable value, which could not be deduced as constant during runtime, in a
  sharded GeneralGraph the filter was not applied correctly.
  SmartGraphs and SingleServer traversals are not effected by this issue.
  Also OneShard traversals are not effected.

* Added `securityScheme`, `securityScope` and `security` methods to Foxx
  endpoints to allow defining Swagger security schemes.

* Foxx routes now always have a Swagger `operationId`. If the route is unnamed,
  a distinct operationId will be generated based on the HTTP method and URL.

* Fixed, if you have a collection access within a Subquery, where the main query
  is fully skipped and the "splice-subqueries" rule is active. The information
  of the skip was not transported correctly. This could cause incorrect counting
  reports.
  If splice-subqueries are disabled, or the main-query is only partly skipped,
  everything worked as expected.

* Expanded -std=c++17 flag to all compilers.

* Fixed issue ES-600: log out of LDAP automatically.

  Option `--ldap.allow-offline` was added with `false` value by default to let
  server use cached offline credentials in case of LDAP server communication was
  lost and ArangoDB can't prolong cache online.

* Include all requests to the web interface at endpoint `/_admin/aardvark` in
  a node's requests statistics. Previously, requests going to `/_admin/aardvark`
  were excluded from the request statistics if and only if the requested
  database was the `_system` database.


v3.7.1-alpha.2 (2020-05-27)
---------------------------

* Fixed an issue with truncate of a collection after a dbserver was restarted
  very quickly. This could block arangosync from making progress because the
  _jobs collection could no longer be truncated.

* Web UI is now using pre-compiled templates.

* Fixed issue #11652: Error: AQL: missing variable #2 (1) for node #6
  (FilterNode) while planning registers.

* Fixed internal issue #650: Added analyzer cache update before ArangoSearch
  view creation/update.

* Update OpenLDAP to 2.4.50.

* Fixed issue ES-554: failed to build filter while querying ArangoSearch view.

  This change provides global fix for ArangoSearch analyzers distribution and
  management within a cluster.

* Fixed issue ES-545: failing modification operations led to random crashes with
  "Index out of bound" error message.

* Fix edge cases in RocksDB primary index range lookups for operators >= and <.

* Fixed issue #11525: Address security vulnerability by updating Swagger-UI
  dependency (upgraded Swagger UI to 3.25.1).

* Updated arangosync to 0.7.5.


v3.7.1-alpha.1 (2020-05-15)
---------------------------

* Fixed ability to edit graph edge in Graph Viewer of web UI.

* Fixed issue #10371: For k-shortest-paths queries on certain graphs a condition
  to abort search was too eager and hence missed the shortest path between two
  vertices and returned a longer one.

* Added feature: Disjoint SmartGraphs

  SmartGraphs have been extended to a new subtype, called `Disjoint SmartGraphs`.
  A Disjoint SmartGraph prohibits edges between different SmartGraph components.
  In case the graph schema can be represented without the need of connected
  SmartGraph components, a Disjoint SmartGraph should be used as this knowledge
  can be used by the internal optimizer to gain even better query traversal
  execution times.

* Fixed issue ES-605: arangodump broken when using encryption.

* Fixed a lockup in dropCollection due to a mutex being held for too long.

* Add an optimizer rule that enables execution of certain subqueries on a DB
  Server. For this optimization to work, the subquery must contain exactly one
  DISTRIBUTE/GATHER pair and only access at most one collection.

  This proves particularly useful for traversals, shortest path, and k-shortest
  paths queries on disjoint smart graphs where the entire traversal is executed
  on the DB Server without involvement of a coordinator.

* ClusterInfo does its own updating of plan / current caches.

* Properly forward requests based on the transaction header.

* Fix a bug in Agency poll, which would sometimes return a wrong firstIndex.

* Made several improvements to AQL's internal register planning. This will
  improve query performance in several cases; mainly of large queries with many
  variables, and of spliced subqueries. This also fixes the github issues #10853
  and #11358 which reported performance degradation caused by subquery splicing.

* Fixed issue #11590: Querying for document by _key returning only a single
  seemingly random property on entity ("old", in this case).

  This fixes single-key document lookups in the cluster for simple by-key AQL
  queries, such as `FOR doc IN collection FILTER doc._key == @key RETURN doc`
  in case the document has either an "old" or a "new" attribute.

* Restored behavior of Foxx API documentation being expanded to show all routes
  rather than collapsing all sections by default.

* Add optimization for subqueries for which only the number of results matters.
  The optimization will be triggered for read-only subqueries that use a full
  collection scan or an index scan, without any additional filtering (early
  pruning or document post-filtering) and without LIMIT.

  It will help in the following situation:

      FOR doc IN ...
        LET count = COUNT(
          FOR sub IN subCollection
            FILTER sub._from == doc._id
            RETURN sub
        )
        ...

  The restrictions are that the subquery result must only be used with the
  COUNT/LENGTH function and not for anything else. The subquery itself must be
  read-only (no data-modification subquery), not use nested FOR loops nor LIMIT,
  nor a FILTER condition or calculation that requires accessing the document
  data. Accessing index data is supported for filtering, but not for further
  calculations.

  If the optimization is triggered, it will show up in the query execution plan
  under the name `optimize-count`.

* Integrated libiresearch log topic properly into ArangoDB logging system.

* Fix a bad deadlock because transaction cleanup was pushed with too low
  priority.

* Allow specifying graph names as unquoted string in an AQL graph traversal
  query, e.g. `FOR ... IN ... GRAPH abc`. Previously, the graph name had to be a
  bind parameter or a string enclosed in quotes.

* loadPlan and loadCurrent have been fixed to not miss out on increments.

* ClusterInfo and all data definition changes use the local agency cache.

* Coordinators and DB servers cache the agency's read key value store.

* Agency offers the new poll API to subscribe to the Raft log stream.

* Added option `--rocksdb.edge-cache` to toggle in-memory caching for edges. The
  option is turned on by default. This normally helps with performance in
  read-only and read-mostly workloads.

* Fixed a bug in Maintenance which could prevent collection creation from
  working (made CreateCollection action idempotent).

* Fix potential undefined behavior in some operations issued to the REST handler
  at `/_api/collection` in cluster mode.

* `--cluster.agency-prefix` marked as obsolete. Did never work and is not
  supported.

* Removed old AgencyComm.

* Improve continuation behavior of AQL queries. We post the continuation handler
  on lane CLUSTER_AQL instead of CLIENT_AQL. This leads to MEDIUM prio instead
  of LOW.

* Fixed a sleeping barber in fuerte. Added TLA+ model to prove that there is not
  another one hiding somewhere.

* Fix spurious bugs in `resilience_move` tests due to replication context of
  to-be-dropped collections lingering around until timeout.

* Fixed issue 11243: `db._isSystem()` returns true for a non-system DB.

* Fix REST API `/_admin/cluster/removeServer`.
  This also fixes the removal of servers via the web interface.

* Updated OpenSSL to 1.1.1g.

* Lower the default values for `--rocksdb.block-cache-size` and
  `--rocksdb.total-write-buffer-size` for systems with less than 1 GB of RAM.
  For these systems the default value were presumably too high and needed manual
  adjustment.

* Allow to override the detected number of available CPU cores via an environment
  variable ARANGODB_OVERRIDE_DETECTED_NUMBER_OF_CORES.

* Fix a bug in the optimizer which handled combinations of SmartJoins with
  query components that had to run on a coordinator wrong.

* Remove the scatter-arangosearch-view-in-cluster optimizer rule; it was
  integrated into scatter-in-cluster.

* Issue #11447: expose extra RocksDB options `--rocksdb.target-file-size-base`
  and `--rocksdb.target-file-size-multiplier` to make them configurable.

* Allow STARTS_WITH function to process several prefixes with min match count
  parameter.

* Added collection property changes into auditing.

* Fix creation of example graphs when `--cluster.min-replication-factor` is set
  and no replication factor was specified for a graph. In this case, the new
  graph will use the configured minimum replication factor.

* Show some index details in index overview of the web interface, e.g. the
  `minLength` attribute for fulltext indexes or the `expireAfter` attribute for
  TTL indexes.

* Added startup option `--server.validate-utf8-strings`.

  This option controls whether strings in incoming JSON or VPack requests are
  validated for UTF-8 conformity. If set to `true` (which is the default value),
  all strings are validated, and requests containing invalid UTF-8 data will be
  rejected. Setting the option to `false` will turn off the UTF-8 string
  validation for incoming requests. This mode should only be used for
  deployments that are known to already contain invalid UTF-8 data and to keep
  them operational until the wrong string encoding is fixed in the data.

* Improve continuation behavior of AQL queries. We post the continuation handler
  on lane CLUSTER_AQL instead of CLIENT_AQL. This leads to MEDIUM prio instead
  of LOW.

* When relaying requests to other coordinators in a load-balanced setup, don't
  forward the "http/1.1" HTTP header from the remote response. Forwarding that
  header in the same way as any other header is not required, as each response
  will store and set its status separately anyway. Adding it to the headers
  again will only cause confusion and some client applications to choke.

* Fixed issue #11104: `ANALYZER` function doesn't correctly process `!= OR`
  queries.

* Obsoleted startup option `--database.maximal-journal-size`. This option was
  useful for the MMFiles storage engine only, but did not have an effect with
  the RocksDB engine. Using this startup option is not an error, but has no
  effect anymore.

* Added `JACCARD` AQL function.

* storedValues property is removed from ArangoSearch link properties output.

* Added primarySortCompression property to ArangoSearch views.

* Added compression property to ArangoSearch view storedValues.

* Removed deprecated MMFiles storage engine and also the `arango-dfdb`(datafile
  debugger) executable that could be used to validate MMFiles datafiles.

  This change also obsoletes all MMFiles-specific startup options in the
  `--wal.*` section. Using these startup options is not an error, but has no
  effect anymore.

* Fixed a bug in the agency supervision, which ignored the `failoverCandidates`
  field.

* Added `INTERLEAVE` AQL function.

* Upgraded bundled RocksDB library to version 6.8.0.


v3.7.0 (2020-04-11)
-------------------

* Updated OpenSSL to 1.1.1f.

* Fixed a bug which occurred if a DB-Server was shut down exactly when it was
  supposed to resign from its leadership for a shard.

* Fix a bug in the agency supervision, which could declare an already FAILED
  dbserver temporarily as GOOD again after an agency leader change.

* Added overwrite mode "ignore" for document inserts. This mode allows ignoring
  primary key conflicts on insert when the target document already exists.

  The following overwrite modes now exist:

  - "ignore": if a document with the specified `_key` value already exists,
    nothing will be done, and no write operation will be carried out. The
    insert operation will return success in this case. This mode does not
    support returning the old or new document versions using the `returnOld`
    and `returnNew` attributes.
  - "replace": if a document with the specified `_key` value already exists,
    it will be overwritten with the specified document value. This mode will
    also be used when no overwrite mode is specified but the `overwrite`
    flag is set to `true`.
  - "update": if a document with the specified `_key` value already exists,
    it will be patched (partially updated) with the specified document value.

  The overwrite mode "ignore" can also be used from AQL INSERT operations by
  specifying it in the INSERT's `OPTIONS`, e.g.

  INSERT { _key: ..., .... } INTO collection OPTIONS { overwriteMode: "ignore" }

  Again, when the overwrite mode "ignore" is used from AQL, it does not support
  returning the old or new document versions. Using "RETURN OLD" in an INSERT
  operation that uses the "ignore" overwrite mode will trigger a parse error, as
  there will be no old version returned, and "RETURN NEW" will only return the
  document in case it was inserted. In case the document already existed,
  "RETURN NEW" will return "null".

  The main use case of inserting documents with overwrite mode "ignore" is to
  make sure that certain documents exist in the cheapest possible way.
  In case the target document already exists, the "ignore" mode is most
  efficient, as it will not retrieve the existing document from storage and not
  write any updates to it.


v3.7.0-preview.1 (2020-03-27)
-----------------------------

* Added AQL function `IN_RANGE`.

* Added startup option `--ssl.prefer-http1-in-alpn` to optionally let the server
  prefer HTTP/1.1 over HTTP/2 in ALPN protocol negotiations.

* Compilation issues with wrong cv-qualifiers and unnecessary temporary copying.


v3.7.0-alpha.2 (2020-03-20)
---------------------------

* Add DTRACE points to track a request through the infrastructure.

* Fixed issue #11275: indexes backward compatibility broken in 3.5+.

* Updated snowball dependency to the latest version.
  More stemmers are available. Built-in analyzer list is unchanged.

* Reactive REST API endpoint at `/_admin/auth/reload`, as it is called by DC2DC.

* Fix an endless loop in FollowerInfo::persistInAgency which could trigger a
  hanger if a collection was dropped at the wrong time.

* Updated LZ4 dependency to version 1.9.2.

* Fix cluster representation of the collection figures for RocksDB.

* Changed behavior for creating new collections in OneShard databases (i.e.
  databases with "sharding" attribute set to "single"):

  Previously it was allowed to override "distributeShardsLike" and
  "numberOfShards" for each new collection. The default values were "_graphs"
  and 1 and were not modified if the user did not alter them, but it was still
  possible to alter them.
  This is now (silenly) ignored. Any attempt to set any value for
  "distributeShardsLike" or "numberOfShards" for new collections in a OneShard
  database will silently be ignored. The collection will automatically be
  sharded like the sharding prototype and will have a single shard.

  The behavior has been adjusted to match the behavior when creating collections
  in a cluster started with `--cluster.force-one-shard true` option. Here any
  user-supplied values for "distributeShardsLike" or "numberOfShards" were
  always ignored.

  Now the behavior is identical for OneShard databases and databases in cluster
  started with `--cluster.force-one-shard true`.

  The web interface now also hides the "Distribute shards like" settings in this
  case, and makes the "Number of shards" input box read-only.

* Fix premature access to temporary path before a user-specified path was read
  from the config options.

* Rebuild UI and update swagger.

* Added ability to store values in ArangoSearch views.

* Added LIKE operator/function support to SEARCH clause.

* Added NGRAM_SIMILARITY and NGRAM_POSITIONAL_SIMILARITY functions for
  calculating Ngram similarity.

* Added ngram fuzzy search. Supported by NGRAM_MATCH filter function for SEARCH
  clause.

* Allow to override the detected total amount of memory via an environment
  variable ARANGODB_OVERRIDE_DETECTED_TOTAL_MEMORY.

* `splice-subqueries` optimization is not limited by any type of operation
  within the subquery any more. It can now be applied on every subquery and will
  be by default.
  However they may be a performance impact on some queries where
  splice-subqueries are not as performant as non-spliced subqueries. This is due
  to internal memory management right now and will be addressed in future
  versions. Spliced subqueries can be less performant if the query around the
  subquery is complex and requires lots of variables, or variables with large
  content, but the subquery itself does not require a lot of variables and
  produces many intermediate results s.t. good batching within the query does
  not pay off against memory overhead.

* Supervision to clean up zombie servers after 24h, if no responsibility for
  shards.

* Fix SORT RAND() LIMIT 1 optimization for RocksDB when only a projection of the
  attributes was used. When a projection was used and that projection was
  covered by an index (e.g. `_key` via the primary index), then the access
  pattern was transformed from random order collection seek to an index access,
  which always resulted in the same index entry to be returned and not a random
  one.

* Mark server startup options `--foxx.*`, `--frontend.*` and `--javascript.*` as
  single server and Coordinator only for documentation (`--dump-options`).

* Supervision hot backup and supervision maintenance modes ttl fix.

* Fix a bug that leads to graph traversals yielding empty output when none of
  the output variables (vertex, edge, path) are used. This is relevant when a
  query is only interested in a COUNT of the outputs, for example.

* MoveShard to check, if target is in sync follower before promotion to leader.

* Agency ttl bug fix.

* Added the SNI feature for TLS. This means that one can configure multiple
  server keys and certificate chains and the system dynamically uses the right
  one depending on the value of the TLS servername extension.
  This allows to use different TLS setups for the same server which is reachable
  behind different DNS names, for example (Enterprise Edition only).

* Do not create a reboot tracker for empty serverId ubin sync repl.

* Fix the usage of the AQL functions `CALL` and `APPLY` for calling user-defined
  AQL functions when invoking an AQL query from the arangosh or a client
  application. Previously, invoking an AQL query and using the `CALL` or `APPLY`
  AQL functions to call user-defined AQL function caused undefined behavior.

* Improved graph traversal performance via some internal code refactoring:

  - Traversal cursors are reused instead of recreated from scratch, if possible.
    This can save lots of calls to the memory management subsystem.
  - Unnecessary checks have been removed from the cursors, by ensuring some
    invariants.
  - Each vertex lookup needs to perform slightly less work.

  The traversal speedups observed by these changes alone were around 8 to 10%
  for single-server traversals and traversals in OneShard setups. Cluster
  traversals will also benefit from these changes, but to a lesser extent. This
  is because the network roundtrips have a higher share of the total query
  execution times there.

* Traversal performance can also be improved by not fetching the visited
  vertices from the storage engine in case the traversal query does not refer to
  them.
  For example, in the query

      FOR v, e, p IN 1..3 OUTBOUND 'collection/startVertex' edges
        RETURN e

  the vertex variable (`v`) is never accessed, making it unnecessary to fetch
  the vertices from storage. If this optimization is applied, the traversal node
  will be marked with `/* vertex optimized away */` in the query's execution
  plan output.

* The existing optimizer rule "move-calculations-down" is now able to also move
  unrelated subqueries beyond SORT and LIMIT instructions, which can help avoid
  the execution of subqueries for which the results are later discarded.
  For example, in the query

    FOR doc IN collection1
      LET sub1 = FIRST(FOR sub IN coll2 FILTER sub.ref == doc._key RETURN sub)
      LET sub2 = FIRST(FOR sub IN coll3 FILTER sub.ref == doc._key RETURN sub)
      SORT sub1
      LIMIT 10
      RETURN { doc, sub1, sub2 }

  the execution of the `sub2` subquery can be delayed to after the SORT and
  LIMIT, turning it into

    FOR doc IN collection1
      LET sub1 = FIRST(FOR sub IN coll2 FILTER sub.ref == doc._key RETURN sub)
      SORT sub1
      LIMIT 10
      LET sub2 = FIRST(FOR sub IN coll3 FILTER sub.ref == doc._key RETURN sub)
      RETURN { doc, sub1, sub2 }

* Added JSON-Schema (draft-4) document validation. The validation can be
  specified by providing the new `schema` collection property when creating
  a new collection or when updating the properties of an existing collection.

  db.mycollection.properties({
    schema: {
      rule : { a : { type : "array", items : { type : "number", maximum : 6 }}},
      message : "Json-Schema validation failed"
    }
  });

* Fix supervision mode detection when unlocking agency in hot backup.

* Foxx Response#throw will no longer generate incomplete error objects when
  passed invalid arguments.

* Foxx will now always generate a numeric HTTP status code even if the response
  status code has been set to a non-numeric value.

* Make arangobench return a proper error message when its initial attempt to
  create the test collection fails.

* Added traversal options `vertexCollections` and `edgeCollections` to restrict
  traversal to certain vertex or edge collections.

  The use case for `vertexCollections` is to not follow any edges that will
  point to other than the specified vertex collections, e.g.

      FOR v, e, p IN 1..3 OUTBOUND 'products/123' components
        OPTIONS { vertexCollections: [ "bolts", "screws" ] }

  The traversal's start vertex is always considered valid, regardless of whether
  it is present in the `vertexCollections` option.

  The use case for `edgeCollections` is to not take into consideration any edges
  from edge collections other than the specified ones, e.g.

      FOR v, e, p IN 1..3 OUTBOUND 'products/123' GRAPH 'components'
        OPTIONS { edgeCollections: [ "productsToBotls", "productsToScrews" ] }

  This is mostly useful in the context of named graphs, when the named graph
  contains many edge collections. Not restricting the edge collections for the
  traversal will make the traversal search for edges in all edge collections of
  the graph, which can be expensive. In case it is known that only certain edges
  from the named graph are needed, the `edgeCollections` option can be a handy
  performance optimization.

* Make arangobench return a proper error message when its initial attempt to
  create the test collection fails.

* In some cases with a COLLECT LIMIT situation on a small limit the collect does
  more calls to upstream than without a limit to provide the same result. We
  improved this situation and made sure that LIMIT does not cause the COLLECT to
  fetch too few input rows. There is a chance that queries with a very small
  amount of data might suffer from this modification, most queries will benefit
  however.

* Use OpenSSL's EVP interface for SHA256 instead of the deprecated low-level
  message digest APIs.

* Prevent startup error "specifiy either '--server.jwt-secret-keyfile' or
  '--server.jwt-secret-folder' but not both." from occurring when specifying the
  JWT secret keyfile option.

* Honor `cacheEnabled` attribute when updating the properties of a collection
  in case of the cluster. Previously, this attribute was ignored in the cluster,
  but it shouldn't have been.

* Fixed issue #11137: http-proxy can't pass delete method req.body to foxx.

* Disable "collect-in-cluster" AQL optimizer rule in case a LIMIT node is
  between the COLLECT and the source data. In this case it is not safe to apply
  the distributed collect, as it may alter the results.


v3.7.0-alpha.1 (2020-02-19)
---------------------------

* Rebuild UI.

* Fix arangorestore:

  If a smartGraphAttribute value was changed after document creation,
  the _key of the document becomes inconsistent with the smartGraphAttribute
  entry. However, the sharding of documents stays intact.

  When restoring data from a dump that contains a document with inconsistent
  _key and smartGraphAttribute this would cause an error and the document
  would not be inserted. This is fixed by ignoring this inconsistency in
  the case of restore.

* Updated rclone to 1.51.0.

* Fixed a memory leak in ModificationExecutors.

* Updated arangosync to 0.7.3.

* Updated ArangoDB Starter to 0.14.14.

* Fixed handling of `geometry` attributes in query editor of the web UI.

  Previously, document attributes named `geometry` were treated in a special
  way, assuming they were JSON objects. In case a document contained a
  `geometry` attribute that had a non-object value (e.g. `null`) the web UI
  threw a JavaScript exception and would not display AQL query results properly.

* Disable "collect-in-cluster" AQL optimizer rule in case a LIMIT node is
  between the COLLECT and the source data. In this case it is not safe to apply
  the distributed collect, as it may alter the results.

* Fixed internal issue #4932: COLLECT WITH COUNT together with FILTER yields
  zero.

  This bugfix fixes an issue when skipping over documents in an index scan
  using a covering index and also at the same time using an early-pruning
  filter.

  In this case wrong document data may have been injected into the filter
  condition for filtering, with the filter wrongfully deciding which documents
  to filter out.

* Added crash handler for Linux builds that taps into the following signals:
  - SIGSEGV (segmentation violation)
  - SIGBUS (bus error)
  - SIGILL (illegal instruction)
  - SIGFPE (floating point exception)

  In case the arangod process catches one these signals, the crash handler
  tries to log a message and a backtrace into the installation's logfile before
  terminating the process.

* Removed dependency on libcurl library.

* Remove storage engine selection from package installers. The default storage
  engine for new installations is now RocksDB. The MMFiles storage engine cannot
  be selected for new installations.

* Remove misleading error messages about upgrade commands to carry out in
  case of a database upgrade. These commands are highly platform-dependent and
  also depend on whether ArangoDB is started manually, via the ArangoDB starter
  or as a service.
  In order to not confuse end users, remove the potentially misleading
  instructions.

* Clear backups from DB servers and agency, when plan unchanged not met and not
  allowing for inconsistency.

* V8 version upgrade to 7.9.317; ICU version upgrade to 64.2.
  - JSON parsing is roughly 60% faster than in V8 7.1; you should prefer
    JSON.parse(string) over deeply nested javascript variable declarations.
  - several new javascript language features

* Clean out server job checks preconditions plan version unchanged for start.

* Cluster collection creation preconditioned on involved db servers not in
  process of being cleaned and Fixed.

* Fixed a bug, where a single host agency logged too early.

  This issue caused spurious "Could not integrate into cluster" error messages
  when starting coordinators.

* Fixed issue #11051: AQL query wrongfully returning no results on 3.6.1.

  This change prevents the "late-document-materialization" optimizer rule from
  kicking in in cases when an index scan also uses early pruning of index
  entries. In this case, the late document materialization will lead to the
  filter condition used in early pruning not being applied correctly,
  potentially producing wrong results.

* Fixed internal issue #596: Added ability to disable DNF conversion in SEARCH
  condition for ArangoSearch views. As this conversion might result in
  overcomplicated disjunction and heavily slowdown execution.

* Implement a new API to reload TLS server key and certificates as well as
  client certificate CA. This allows to rotate TLS certificates without a
  restart. One can also query the currently loaded state of the TLS data to
  decide if a reload has actually happened. The new certificates will only be
  used for new TLS-based connections.

* Add acquisition of system report to arangod instances.

* JS `query` helper function can now be used with options.

* Fix a bug in collection creation with `distributeShardsLike`.

* Added load-balancing support for listing currently running and slow AQL
  queries, killing running AQL queries and clearing the list of slow AQL
  queries.

  This change also will also modify the values returned in the "id" attribute of
  AQL query objects. While the values of the "id" attribute remain strings with
  numeric content, the ids will now start at arbitrary offsets after server
  start and are supposed to have much higher numeric values than in previous
  ArangoDB versions.
  In previous ArangoDB versions, query ids always started at value 1 after a
  server start/restart and were increased in increments of 1.
  This change may lead to query ids being greater than what a 4 byte integer can
  hold, which may affect client applications that treat the ids as numeric
  values and do not have proper support for integer numbers requiring more than
  4 byte storage.

* remove unused REST API /_admin/aql/reload, and make the JavaScript bindings
  for `internal.reloadAqlFunctions()` do nothing. The reason for this is that
  AQL functions are automatically reloaded when functions are changed, and
  there is no need to call these APIs manually.

* Fixed issue #10949: k shortest paths behavior wrong with zero weights.

* Added `REPLACE_NTH` AQL function to replace a member inside an array.

* Added JWT secret rotation (Enterprise Edition only).

* The `--dump-options` command for arangod now also provides the "components"
  attribute that shows for which components (agent, coordinator, db server,
  single server) an option is relevant. Additionally, each option provides an
  "os" attribute that indicates on which operating systems the option is
  supported.
  This can be used when reading the options descriptions programmatically, e.g.
  for auto-generating the documentation.

* Add update-insert operation similar to existing replace-insert functionality
  of insert.

  Like for the existing variant the `overwrite` flag has to be set to true.
  Then the update version can be selected by setting the `overwriteMode` to
  `"update"`.


v3.6.1 (2020-01-29)
-------------------

* Updated ArangoDB Starter to 0.14.13.

* Added HotBackup events into auditing.

* Internal statistics API now uses integer timestamps instead of doubles. The
  old behavior sometimes leads to failed requests because of parse errors
  which occurred in the internally used JavaScript joi library.

* Fixed issue #10896: Variables defined inside spliced subqueries would leak
  into following COLLECT ... INTO var operations.

* Fixed COLLECT not invalidating variables for following COLLECT ... INTO var
  operations.

* Removed an unnecessary and wrong request from within the Web UI to the
  `/_admin/cluster/health` API. This lead to unauthorized network calls.

* Allowed PHRASE function to process empty arrays without generating error.

* Add acquisition of system report to arangod instances.

* React dev mode now supports hot reload combined with proxy for development.

* Fix issue #10897, using COLLECT in a subquery could lead to unexpected and
  confusing error messages

* Fix potential nullptr dereference in view optimizer rule, when there were was
  a LIMIT outside of a FOR loop.

* Added missing "global" parameter in Swagger REST API documentation for some
  replication endpoints.

* Fixed an edge case in VelocyPack when the padding of a 1-byte offsetSize array
  is removed but the first few entries of the array contain a Slice of type
  None.

* Fix Foxxmaster failover retry loop to not spin forever and give up after a
  while.

* Fix "random" Valgrind "invalid free/delete" errors caused by usage of
  `alignas(64)` for shared_ptrs in the SupervisedScheduler.

* MoveShard to check, if target is in sync follower before promotion to leader.

* Fixed issue #10867: arangod based binaries lack product version info and icon
  on Windows

* Fixed internal traversal edge collection cache being filled up correctly.
  Edges are able to point to other edges, but those we're not applied to the
	cache.

* Fixed issue #10852: Nested spliced subqueries could return wrong results in
  some cases when some of the concerned subqueries did not return any values.

* Fix string comparison bug that lead to traversal queries accepting prefixes of
  "edges" and "vertices" to be used as object accessors for the path object.

* Fix bug affecting spliced subqueries when memory blocks are reused.

* Now clearing an internal map inside the traverser engine correctly.

* Add ability to choose logging to file in Windows installer of ArangoDB server
  (enabled by default).

  ArangoDB-logs folder with arangod.log should be stored in %PROGRAMDATA% and
  %LOCALAPPDATA% for all users and single user installation respectively.

* Fix installation of arangoinspect and libraries in Windows client installer.

* Disable cluster AQL parallelization for queries that contain traversal,
  shortest path or k_shortest_path nodes.
  This avoids potential undefined behavior in case a parallel GatherNode is used
  in graph queries.

* Fixed internal issue #656: While executing large amount of concurrent
  insert/removes iresearch seems to leak open file handles. This results in
  error 32 in cleanup (observed only on Windows as Linux doesn`t have file
  sharing locks).

* Updated arangosync to 0.7.2.

* Fix Windows client package JS installation paths.

* Added option that makes ArangoDB write logfiles in the Windows installer.


v3.6.0 (2020-01-08)
-------------------

* Do not create a reboot tracker for empty serverId ubin sync repl.

* Update swagger.


v3.6.0-rc.2 (2019-12-23)
------------------------

* Fixed issue #10725: Wrong document count shown inside an empty collection in
  the web UI.

* Fixed bug that made AQL queries eligible for parallelization even in case
  they couldn't be parallelized, leading to undefined behavior due to the
  thread races.

* Fixed bug in `--query.parallelize-gather-writes-behavior` which led to
  the option not working correctly.

* Agency relational operators TTL fix.

* Improve latency in scheduler.

* Fixed internal issue #4748: Editing a single edgeDefinition using the graph
  API failed if it was not shared between all available graphs.

* Fixed undefined behavior on node delete.

* Fixed agency invalid operation.

* Added google tests for Node.

* Bugfix: An AQL ternary expression with the condition being true at query
  compile time would not execute its false branch.

* When starting the RocksDB engine, first create small-sized WAL files.

  This is a precaution so that when repeatedly trying to start an arangod
  instance, an instant instance startup failure will not lead to the disk
  filling up so quickly with WAL file data.

  The WAL file size is increased later on in the startup sequence, so
  everything should be fine if the startup works.

  This fixes a problem with the disk filling up quickly when the arangodb
  starter tries to start an instance 100 times in a row but instantaneously
  gets a failure back from it.

* Fixed a permissions bug for /_admin/cluster/rebalanceShards.


v3.6.0-rc.1 (2019-12-10)
------------------------

* Renamed document / index / vertex "handle" to "identifier" / "id" for
  consistency in documentation and error messages.

* Fixed a bug in smart graph bfs traversals that might violate path uniqueness
  requirements in rare cases.

* Fixed permissions for dump/restore.

* Rename `minReplicationFactor` into `writeConcern` to make it consistent with
  `--cluster.write-concern` and avoid confusion with
  `--cluster.min-replication-factor`

* ArangoDB web UI switched build system from grunt to webpack.
  It supports React now, but the previous framework is still in use.

* Enable the `parallelize-gather` AQL optimizer rule for certain write queries.
  The optimization is turned on by default and can be disabled by setting the
  startup option `--query.parallelize-gather-writes` to `false`.

* Shard synchronization readlock aware of rebootId.

* Bugfix: In an AQL cluster query, when gathering unsorted data in combination
  with a LIMIT with non-zero offset, if this offset exactly matches the number
  of documents in the first shards consumed, the rest of the documents was not
  returned.

* REMOTE and GATHER no longer make subqueries unsuitable for the
  `splice-subqueries` optimization.

* New internal counter and histogram support.

* Add a Prometheus endpoint for metrics, expose new metrics, old statistics
  and RocksDB metrics.

* Fixed known issue #509: ArangoSearch index consolidation does not work during
  creation of a link on existing collection which may lead to massive file
  descriptors consumption.

* Added support of array comparison operators to ArangoSearch.

* Added support of arrays to TOKENS function.

* Added support of arrays to PHRASE function.

* Added a new optimization called `late-document-materialization`,
  `late-document-materialization` for indexes and views correspondingly.

  This optimization reduces amount of documents to read from storage engine to
  the limit explicitly stated in LIMIT node.

* Added support of "Edge N-grams" to `text` analyzer.

* Added support of UTF8 to `ngram` analyzer.

* Added ability to mark beginning/end of the sequence to `ngram` analyzer.

* Added a new subquery optimization called `splice-subqueries`.

  This optimization splices the execution of subqueries inside the
  execution of their surrounding query. For subqueries with a fairly
  small number of returned documents this saves a significant amount of
  overhead initializing cursors.

  This optimization is performed by the optimizer rule
  `splice-subqueries`, and currently only works on suitable
  subqueries; A subquery becomes unsuitable if it contains a LIMIT,
  REMOTE, GATHER or a COLLECT node where the operation is WITH COUNT INTO.
  A subquery also becomes unsuitable if it is contained in an
  (sub)query which contains unsuitable operations after the subquery.

  The implementation of this feature required reworking the dataflow
  query execution.

* Added startup option `--query.optimizer-rules` to selectively enable or
  disable optimizer rules by default. The option can be specified multiple
  times, and takes the same input as the query option of the same name.

  For example, to turn off the rule _use-indexes-for-sort_, use

      --query.optimizer-rules "-use-indexes-for-sort"

  The purpose of this option is to be able to enable potential future
  experimental optimizer rules, which may be shipped in a disabled-by-default
  state.

* Allow the AQL query optimizer to remove the DistributeNodes for several
  data-modification queries. So far, only REMOVE queries benefitted. Now the
  optimization can also be applied for REPLACE and UPDATE queries in case the
  query does not use LIMIT and there is no further cluster-internal
  communication after the REMOVE, REPLACE or UPDATE node.

* Include ArangoSearch data in HotBackups.

* Allow to restore 3.5 HotBackups in 3.6.

* Fixed an issue where removeServer left behind current coordinators

* Allow usage of AQL function `RANDOM_TOKEN` with an argument value of `0`. This
  now produces an empty string, whereas in older versions this threw an invalid
  value exception.

* Add startup option `--rocksdb.exclusive-writes` to avoid write-write
  conflicts.

  This options allows for an easier transition from MMFiles to the RocksDB
  storage engine, but comes with a big performance penalty as all collections
  will be locked exclusively for writes.

* Added new maxRuntime option for queries. If a query does not finish execution
  within the given time (in seconds) it will be killed.

* Added limit for AQL range materialization to prevent out-of-memory errors.

  When materializing ranges created by either the AQL `RANGE` function or by
  using the built-in `..` operator (e.g. `1 .. 1000000`), a check is now
  performed if the range is too big to be materialized. The threshold value is
  set to 10 million members. Ranges with at most that many members can be
  materialized, ranges with more members will fail to materialize and abort the
  query with the exception `number out of range` (error code 1504).

  It is still possible to create ranges with more than 10 million members as
  long as they are not materialized. For example, the following is still valid:

      FOR i IN 1 .. 1000000000 INSERT {_key: CONCAT('test', i)} INTO collection

* No longer put system services into `_apps` on single server. On cluster, this
  has never worked. This was unnecessary.

* Added AQL optimizer rule "move-filters-into-enumerate", to allow for early
  pruning of non-matching documents while full-scanning or index-scanning
  documents. This optimization can help to avoid a lot of temporary document
  copies.

* Added "SmartJoins for Views" to the ArangoDB Enterprise Edition that allows
  running cluster joins between two certain sharded collections or views with
  performance close to that of a local join operation.

* Allow collection names to be at most 256 characters long, instead of 64
  characters in previous versions.

* Upgraded bundled Boost library version to 1.71.

* Use `-std=c++17` for ArangoDB compilation.

* Made the mechanism in the Web UI of replacing and upgrading a foxx app more
  clear.

* Show shards of all collections (including system collections) in the web UI's
  shard distribution view.

  This is necessary to access the prototype collections of a collection sharded
  via `distributeShardsLike` in case the prototype is a system collection, and
  the prototype should be moved to another server.

* Rclone URL normalization.

* Disallow using `_id` or `_rev` as shard keys in clustered collections.

  Using these attributes for sharding was not supported before, but didn't
  trigger any errors. Instead, collections were created and silently using
  `_key` as the shard key, without making the caller aware of that an
  unsupported shard key was used.

* Use execvp instead of execv in HotBackup restore.

* Re-enabled the AQL sort-limit optimization rule in conjunction with fullCount
  in the cluster. It now also may speed up fullCount with sorted indexes and a
  limit.

* Make the scheduler enforce the configured queue lengths. The values of the
  options `--server.scheduler-queue-size`, `--server.prio1-size` and
  `--server.maximal-queue-size` will now be honored and not exceeded.

  The default queue sizes in the scheduler for requests buffering have
  also been changed as follows:

      request type        before      now
      -----------------------------------
      high priority          128     4096
      medium priority    1048576     4096
      low priority          4096     4096

  The queue sizes can still be adjusted at server start using the above-
  mentioned startup options.

* Add replicationFactor, minReplicationFactor and sharding strategy to database
  creation dialog in web UI. Preselect database default values for collection
  creation in web UI.

* Add new JavaScript function `db._properties()` that provides information about
  the current database's properties.

* Add new options `sharding` and `replicationFactor` for database creation
  methods. The specified values will provide the defaults for all collections
  created in a database.

  Valid values for `sharding` are `""`, "flexible", "single". The first 2 values
  are treated equally. Values for `replicationFactor` are natural numbers or the
  string `satellite`.

* Add new server option `--cluster.default-replication-factor` that allows to
  set the default replication factor for non-system collections (default: 1).

* Enabled IPO with cmake as an option, default is on for release builds without
  google tests.

* Bugfix: The AQL sort-limit optimization was applied in some cases it
  shouldn't, resulting in undefined behavior.

* Remove operations for documents in the cluster will now use an optimization,
  if all sharding keys are specified. Should the sharding keys not match the
  values in the actual document, a not found error will be returned.

* Retry hot backup list in cluster for 2 minutes before reporting error.

* Allowing inconsistent rather than forcing hot backups.

* Geo functions will now have better error reporting on invalid input.

* Upgraded bundled jemalloc library to version 5.2.1.

* Added TransactionStatistics to ServerStatistics (transactions started /
  aborted / committed and number of intermediate commits).

* Added AQL function DATE_ROUND to bin a date/time into a set of equal-distance
  buckets.

* Enforced the valid date range for working with date/time in AQL. The valid
  date ranges for any AQL date/time function are:

  - for string date/time values: `"0000-01-01T00:00:00.000Z"` (including) up to
    `"9999-12-31T23:59:59.999Z"` (including)
  - for numeric date/time values: -62167219200000 (including) up to
    253402300799999 (including). These values are the numeric equivalents of
    `"0000-01-01T00:00:00.000Z"` and `"9999-12-31T23:59:59.999Z"`.

  Any date/time values outside the given range that are passed into an AQL date
  function will make the function return `null` and trigger a warning in the
  query, which can optionally be escalated to an error and stop the query.

  Any date/time operations that produce date/time outside the valid ranges
  stated above will make the function return `null` and trigger a warning too.
  An example for this is.

      DATE_SUBTRACT("2018-08-22T10:49:00+02:00", 100000, "years")

* Fixed bug in MoveShard::abort which causes a duplicate entry in the follower
  list. (Internal Bug #4378)

* Updated TOKENS function to deal with primitive types and arrays.


v3.5.3 (2019-11-28)
-------------------

* Fixed GET _api/gharial to also include the name property in every returned
  graph. This is a consistency fix within the API as all other APIs include the
  name. As a work around the returned _key can be used, which is identical to
  the name.

* The _users collection is now properly restored when using arangorestore.

* Allow the optimizer to use indexes when a collection attribute is compared to
  an expansion followed by an attribute name, e.g.
  `doc.value IN something[*].name`.

* Updated arangosync to 0.7.0.

* Fixed issue #10470: The WebUI now shows potential errors and details which
  occured using _api/import (e.g. unique constraint violated).

* Fixed issue #10440: Incorrect sorting with sort criteria partially covered
  by index.

* Make the timeouts for replication requests (for active failover and master-slave
  replication configurable via startup options:

      --replication.connect-timeout
      --replication.request-timeout

* Fixed internal issue #4647: dead Coordinators are not removed for agency.

* Fixed UPSERT matching.

  Empty objects in the `UPSERT { ... }` expression will not be omitted anymore:

      db._collection("col").insert({ "find" : "me" });
      db._query(` UPSERT { "find" : "me", "foo" : {} }
                  UPDATE { "foo"  : "not gonna happen" }
                  INSERT { "find" : "me", "foo"  : {} }
                  INTO col
      `)

  This will now correctly insert a document instead of updating the existing,
  that only partially matches the upsert-expression.

* Fixed undefined behaviour with creation of ArangoSearch links with custom
  analyzers in cluster environment.

* Fixed internal issue #651: analyzer duplication in _analyzers collection.

* Fixed internal issue #4597: rebalanceShards API cannot work on any database
  other than the _system database.

* Stop putting system services in _apps on single server, this has never
  worked on cluster and was not needed.

* Fixed issue #10371: K_SHORTEST_PATHS LIMIT 1 can not return the shortest path.
  Now the shortest path is returned as the first one in such queries.

* Improve killability of some types of cluster AQL queries. Previously, several
  cluster queries, especially those containing a `DistributeNode` in their
  execution plans, did not respond to a kill instruction.

  This change also introduces a new query status "killed", which may now be
  returned by the REST APIs at `/_api/query/current` and `/_api/query/slow` in
  the `state` attribute of each query.

* Improve shutdown of some cluster AQL queries on the coordinator in case the
  query has multiple coordinator snippets (true for queries involving more than
  one collection) and the database server(s) cannot be reached on query
  shutdown. In this case the proper shutdown of the coordinator parts of the
  query previously was deferred until the coordinator snippets were removed by
  the automatic garbage collection. Now, the cleanup of the coordinator snippets
  will happen much more quickly, which reduces the chances of the queries
  blocking resources.

* Fixed ArangoSearch index removes being discarded on commiting consolidation
  results with pending removes after some segments under consolidation were
  already committed.

* Assertion fail when no timestamp in agency's persistence.

* Fixed internal issue #647: custom analyzer provokes errors on Active Failover
  deployment.

* Upgraded bundled version of libcurl to 7.66.0.
* When starting a coordinator, wait up to 15 seconds for it to appear
  in the agency under key `Supervision/Health` before reporting as "ready".
  This is necessary because if the coordinator reports ready beforehand
  and is used to create databases etc., the supervision may remove all
  of the jobs started by non-ready coordinators, considering them to be
  from a failed coordinator.
  To avoid huge startup delays, the startup will proceed after waiting
  futilely for 15 seconds and log a message.

* Fixed issue #10270: Query: Expecting type Array or Object (while executing).

* Fix a problem with AQL constrained sort in the cluster, which might abort
  queries. The AQL sort-limit optimization rule may now also speed up fullCount
  with sorted indexes and a limit in the cluster.

* Prevent spurious log message "Scheduler queue is filled more than 50% in last
  x s" from occurring when this is not the case. Due to a data race, the
  message could previously also occur if the queue was empty.

* The General Graph document API is now consistent with the document API in its
  error messages. When attempting to create / modify edges pointing to non
  existing vertex collections HTTP 400 is returned instead of 404.

* Disallow the usage of subqueries inside AQL traversal PRUNE conditions.
  Using subqueries inside PRUNE conditions causes undefined behavior,
  so such queries will now be aborted early on with a parse error
  instead of running into undefined behavior.

* Fixed available flag for hotbackup.

* Fixed list with id for partially available hotbackups.

* Fixed agency TTL bug happening under certain rare conditions.

* Improved performance of some agency helper functions.

* Fixed search not working in document view while in code mode.

* Fixed issue #10090: fix repeatable seek to the same document in
  SEARCH operations for ArangoSearch views.

* Fixed issue #10193: Arangoexport does not handle line feeds when exporting as
  csv.

* Removed debug log messages "found comm task ..." that could be logged
  on server shutdown.

* Fixed issue #10183: Arangoimport imports on _system when you try to
  create a new database.

  This bugfix fixes the output of arangoimport, which could display a
  wrong target database for the import if the option `--create-database`
  was used.

* Fixed issue #10158: Invalid Query Crashes ArangoDB.

  This fixes traversal queries that are run on a static empty start vertex
  string.


v.3.5.2 (2019-11-06)
--------------------

* Fixed ArangoSearch upgrade procedure from previous minor version and
  patches.

* Separately account for superuser and user request traffic.


v3.5.1 (2019-10-07)
-------------------

* Properly report parse errors for extraneous unterminated string literals
  at the end of AQL query strings. For example, in the query `RETURN 1 "abc`,
  the `RETURN 1` part was parsed fully, and the `"abc` part at the end was
  parsed until the EOF and then forgotten. But as the fully parsed tokens
  `RETURN 1` already form a proper query, the unterminated string literal
  at the end was not reported as a parse error.
  This is now fixed for unterminated string literals in double and single
  quotes as well as unterminated multi-line comments at the end of the query
  string.

* Fix config directory handling, so we don't trap into UNC path lookups on Windows.

* Ignore symlinks when copying JavaScript files at startup via the option
  `--javascript.copy-installation`. This potentially fixes the following
  error message at startup:

      Error copying JS installation files to '...':
      failed to open source file ...: No such file or directory

* Added startup option `--cluster.max-number-of-shards` for restricting the
  maximum number of shards when creating new collections. The default
  value for this setting is `1000`, which is also the previously hard-coded
  built-in limit. A value of `0` for this option means "unrestricted".
  When the setting is adjusted, it will not affect already existing
  collections, but only collections that are created or restored
  afterwards.

* Added startup options for managing the replication factor for newly
  created collections:

  - `--cluster.min-replication-factor`: this settings controls the minimum
    replication factor value that is permitted when creating new collections.
    No collections can be created which have a replication factor value
    below this setting's value. The default value is 1.
  - `--cluster.max-replication-factor`: this settings controls the maximum
    replication factor value that is permitted when creating new collections.
    No collections can be created which have a replication factor value
    above this setting's value. The default value is 10.
  - `--cluster.default-replication-factor`: this settings controls the default
    replication factor value that is used when creating new collections and
    no value of replication factor has been specified.
    If no value is set for this option, the value of the option
    `--cluster.min-replication-factor` will be used.

* Fixed unintended multiple unlock commands from coordinator to
  transaction locked db servers.

* DB server locking / unlocking for hot backup revisited and enhanced.

* Rely on reboot ids for declaring end of cluster hot restore on coordinators.

* Obtain new unique IDs via a background thread.

* Fixed issue #10078: FULLTEXT with sort on same field not working.

* Fixed issue #10062: AQL: could not extract custom attribute.

* Fix compilation issue with clang 10.

* Fixed error message for error 1928 ("not in orphan") to "collection is
  not in list of orphan collections".

* Fix strange shutdown hanger which came from the fact that currently
  libgcc/libmusl wrongly detect multi-threadedness in statically linked
  executables.

* Fixed a shutdown bug coming from a read/write lock race.

* Fixed a bug in the edge cache's internal memory accounting, which led
  to the edge cache underreporting its current memory usage.

* Fixed "ArangoDB is not running in cluster mode" errors in active failover setups.
  This affected at least /_admin/cluster/health.

* Made the mechanism in the Web UI of replacing and upgrading a foxx app more clear.

* Fixed the AQL sort-limit optimization which was applied in some cases it should
  not, resulting in undefined behaviour.

* Add --server.statistics-history flag to allow disabling of only the historical
  statistics.  Also added rocksdbengine.write.amplification.x100 statistics
  for measurement of compaction option impact.  Enabled non-historical
  statistics for agents.

* Fixed AQL constrained-heap sort in conjunction with fullCount.

* Added support for AQL expressions such as `a NOT LIKE b`, `a NOT =~ b` and
  `a NOT !~ b`. Previous versions of ArangoDB did not support these expressions,
  and using them in an AQL query resulted in a parse error.

* Disallow creation of TTL indexes on sub-attributes.

  Creation of such indexes was not caught before, but the resulting
  indexes were defunct. From now on the creation of TTL indexes on sub-
  attributes is disallowed.

* Added HotBackup feature.

* Improved database creation within the cluster. In the case of
  coordinator outages during the creation of the database there was a
  chance that not all relevant system collections had been created.
  Although this database was accessible now some features did not work
  as expected (e.g. creation of graphs). We modified creation of a new
  database as an all or nothing operation and only allow access to the
  database after all system collections are properly perpared to avoid
  the above inconsistencies. Also creation of databases are now secured
  against coordinator outages, they will either be fully created or not
  visible and eventually dropped. This does not require any change on the
  client code.

* Added UI support to create documents in a collection using smartGraphAttribute
  and/or smartJoinAttribute.

* Add count of objects to latency reporting in arangoimport.

* Harden database creation against spurious "duplicate name" errors that
  were caused by other parallel operations lazily creating required
  system collections in the same database.

* Fixed internal issue #633: made ArangoSearch functions BOOST, ANALYZER, MIN_MATCH
  callable with constant arguments. This will allow running queries where all arguments
  for these functions are deterministic and do not depend on loop variables.

* Automatically turn values for deprecated startup option parameters
  `--log.use-microtime` and `--log.use-local-time` into equivalent option values
  of the new, preferred option `--log.time-format`.

* Drop collection action to timeout more quickly to stay on fast lane.

* Make arangorestore restore data into the `_users` collection last. This is to
  ensure that arangorestore does not overwrite the credentials of its invoker while
  the restore is running, but only at the very end of the process.
  This change also makes arangorestore restore the `_system` database last if it
  is started with the `--all-databases` option.

* Fixed the removal (including a collection drop) of an orphanCollection from a
  graph definition when using the arango shell. The boolean flag whether to drop
  the collection or not was not transferred properly.

* Check for duplicate server endpoints registered in the agency in sub-keys of
  `/Current/ServersRegistered`.

  Duplicate endpoints will be registered if more than one arangod instance is
  started with the same value for startup option `--cluster.my-address`. This can
  happen unintentionally due to typos in the configuration, copy&paste remainders etc.

  In case a duplicate endpoint is detected on startup, a warning will be written
  to the log, indicating which other server has already "acquired" the same endpoint.

* Make graph operations in general-graph transaction-aware.

* Fixed adding an orphan collections as the first collection in a SmartGraph.

* Fixed non-deterministic occurrences of "document not found" errors in sharded
  collections with custom shard keys (i.e. non-`_key`) and multi-document lookups.

* Fixed issue #9862: ServerException: RestHandler/RestCursorHandler.cpp:279

  This fixes an issue with the RocksDB primary index IN iterator not resetting its
  internal iterator after being rearmed with new lookup values (which only happens
  if the IN iterator is called from an inner FOR loop).

* Geo functions will now have better error reporting on invalid input.

* The graph viewer of the web interface now tries to find a vertex document of
  all available vertex collections before it aborts.

* Fixed issue #9795: fix AQL `NOT IN` clause in SEARCH operations for ArangoSearch
  views.

* Make minimum timeout for synchronous replication configurable via parameter
  (--cluster.synchronous-replication-timeout-minimum) and increase default value
  to prevent dropping followers unnecessarily.

* Added support for TLS 1.3 for the arangod server and the client tools.

  The default TLS protocol for the arangod server is still TLS 1.2 however, in order
  to keep compatibility with previous versions of ArangoDB.

  The arangod server and any of the client tools can be started with option
  `--ssl.protocol 6` to make use of TLS 1.3.

  To configure the TLS version for arangod instances started by the ArangoDB starter,
  one can use the `--all.ssl.protocol=VALUE` startup option for the ArangoDB starter,
  where VALUE is one of the following:

  - 4 = TLSv1
  - 5 = TLSv1.2
  - 6 = TLSv1.3

* Fixed parsing of "NOT IN" in AQL, which previously didn't correctly parse
  "NOT IN_RANGE(...)" because it checked if the "NOT" token was followed by
  whitespace and then the two letters "IN".

* Changed log level for message "keep alive timeout - closing stream!" from INFO to
  DEBUG.

* Don't create temporary directories named "arangosh_XXXXXX" anymore.

* Add TransactionStatistics to ServerStatistics (transactions started /
  aborted / committed and number of intermediate commits).

* Upgraded version of bundled curl library to 7.65.3.

* Don't retry persisting follower information for collections/shards already
  dropped. The previous implementation retried (unsuccessfully in this case)
  for up to 2 hours, occupying one scheduler thread.

* Fixed internal issue #4407: remove storage engine warning.

* Agents to remove callback entries when responded to with code 404.

* Fixed internal issue #622: Analyzer cache is now invalidated for dropped database.

* Show query string length and cacheability information in query explain output.

* The AQL functions `FULLTEXT`, `NEAR`, `WITHIN` and `WITHIN_RECTANGLE` are now
  marked as cacheable, so they can be used in conjunction with the AQL query
  results cache on a single server.

* Fixed issue #9612: fix ArangoSearch views getting out of sync with collection.

* Fix an issue with potential spurious wakeups in the internal scheduler code.

* Changes the _idle_ timeout of stream transactions to 10 seconds and the total
  per DB server size of stream transaction data to 128 MB. The idle timer is
  restarted after every operation in a stream transaction, so it is not the
  total timeout for the transaction.

  These limits were documented in the manual for stream transactions since 3.5.0,
  but are enforced only as of 3.5.1. Enforcing the limits is useful to free up
  resources from abandoned transactions.

* Consistently honor the return value of all attempts to queue tasks in the
  internal scheduler.

  Previously some call sites did not check the return value of internal queueing
  operations, and if the scheduler queue was full, operations that were thought
  to be requeued were silently dropped. Now, there will be reactions on such
  failures. Requeuing an important task with a time offset (Scheduler::queueDelay)
  is now also retried on failure (queue full) up to at most five minutes. If after
  five minutes such a task still cannot be queued, a fatal error will be logged
  and the server process will be aborted.

* Made index selection much more deterministic in case there are
  multiple competing indexes.

* Fixed issue #9654: honor value of `--rocksdb.max-write-buffer-number` if it
  is set to at least 9 (which is the recommended value). Ignore it if it is
  set to a lower value than 9, and warn the end user about it.

  Previous versions of ArangoDB always silently ignored the value of this setting
  and effectively hard-coded the value to 9.

* Fixed internal issue #4378: fix bug in MoveShard::abort which causes a
  duplicate entry in the follower list.

* Fixed cut'n'pasting code from the documentation into arangosh.

* Fixed issue #9652: fix ArangoSearch wrong name collision and raising
  "Name collision detected" error during creation of a custom analyzer with
  stopwords.

* Fixed an agency bug found in Windows tests.

* Added initial support for wgs84 reference ellipsoid in GEO_DISTANCE through
  third optional parameter to AQL function.

* Added support for area calculations with GEO_AREA AQL function.

* Correct RocksDB statistics to report sums from column families instead of
  single value from default column family.

* Fixed agency nodes to not create bogus keys on delete / observe / unobserve.

* Fixed issue #9660: fix multiple plans processing during optimization of AQL
  query which uses ArangoSearch scorers.

* Fixed issue #9679: improved error message for FULLTEXT function invocation
  failures.

* Fixed error message "Invalid argument: assume_tracked is set but it is not
  tracked yet" when trying to write to the same keys in concurrent RocksDB
  transactions. This error will now be reported as a "Lock timeout" error,
  with error code 1200.

* Added resign leadership job to supervision.

* Hide MMFiles-specific information in web UI (in detail view of a collection)
  when the storage engine is not MMFiles or when the information is not
  available.

* Keep followers in sync if the old leader resigned and stopped writes.

* Abort a FailedLeader job when its _to_ server fails.

* Decreased unnecessary wait times for agency callbacks in case they were
  called earlier than expected by main thread.

* Make arangosh not close the connection after the user aborts an operation.
  This restores the same behavior as in previous versions of ArangoDB, which
  also left the connection open.

* Refactor maintenance to use a TakeoverShardLeadership job. This fixes a bug
  that a shard follower could have set the wrong leader for the shard locally.


v3.5.0 (2019-08-20)
-------------------

* Fix web UI link to ArangoSearch views documentation.

* Rebuild swagger for web UI and documentation.


v3.5.0-rc.7 (2019-08-01)
------------------------

* Upgraded arangodb starter version to 0.14.12.


v3.5.0-rc.6 (2019-07-29)
------------------------

* Fixed issue #9459: Optimization rule remove-collect-variables does not KEEP all necessary data.

* Added gzip and encryption options to arangoimport and arangoexport.

* Added missing REST API route GET /_api/transaction for retrieving the list of
  currently ongoing transactions.

* Fixed issue #9558: RTRIM not working as expected.

* Added startup error for bad temporary directory setting.

  If the temporary directory (--temp.path) setting is identical to the database
  directory (`--database.directory`) this can eventually lead to data loss, as
  temporary files may be created inside the temporary directory, causing overwrites of
  existing database files/directories with the same names.
  Additionally the temporary directory may be cleaned at some point, and this would lead
  to an unintended cleanup of the database files/directories as well.
  Now, if the database directory and temporary directory are set to the same path, there
  will be a startup warning about potential data loss (though in ArangoDB 3.4 allowing to
  continue the startup - in 3.5 and higher we will abort the startup).

* Make TTL indexes behave like other indexes on creation.

  If a TTL index is already present on a collection, the previous behavior
  was to make subsequent calls to `ensureIndex` fail unconditionally with
  the error "there can only be one ttl index per collection".

  Now we are comparing the attributes of the to-be-created index with the
  attributes of the existing TTL index and make it only fail when the
  attributes differ. If the attributes are identical, the `ensureIndex`
  call succeeds and returns the existing index.


v3.5.0-rc.5 (2019-07-22)
------------------------

* MinReplicationFactor:
  Collections can now be created with a minimal replication factor (minReplicationFactor), which defaults to 1.
  If minReplicationFactor > 1 a collection will go into "read-only" mode as soon as it has less then minReplicationFactor
  many insync followers. With this mechanism users can avoid to have collections diverge too much in case of failure scenarios.
  minReplicationFactor can have the values: 1 <= minReplicationFactor <= replicationFactor.
  Having minReplicationFactor == 1 ArangoDB behaves the same way as in any previous version.

* Fixed a query abort error with SmartJoins if both collections were restricted to a
  single shard using the "restrict-to-single-shard" optimizer rule.

* Fixed a performance regression of COLLECT WITH COUNT INTO.

* Fixed some races in cluster collection creation, which allowed collections with the
  same name to be created in parallel under some rare conditions.

* arangoimport would not stop, much less report, communications errors.  Add CSV reporting
  of line numbers that are impacted during such errors

* Prevent rare cases of duplicate DDL actions being executed by Maintenance.

* Coordinator code was reporting rocksdb error codes, but not the associated detail message.
  Corrected.

* The keep alive timeout specified via --http.keep-alive-timeout is now being honored

* Replication requests on Document API are now on higher priority then client-triggered requests.
  This should help to keep in sync replication up and running even if the server is overloaded.

* Bugfix: Save distinct WAL ticks for multiple replication clients from the same
  server. Also, when a follower is added for synchronous replication, the WAL
  tick held by the client is freed immediately, rather than waiting for a
  timeout.
  The corresponding APIs get a new parameter `syncerId`, which, if given,
  supersedes `serverId`. This affects the routes /_api/wal/tail,
  /_api/replication/batch, /_api/replication/logger-follow and the internal
  route /_api/replication/addFollower. The new field `syncerId` is also added to
  the response of /_api/replication/logger-state.

* Disallow indexing the `_id` attribute even as a sub-attribute.
  Previous versions of ArangoDB did not allow creating indexes on the `_id` attribute, but didn't
  check if an index was created on the `_id` attribute of a sub-attribute, e.g. `referredTo._id`
  or `data[*]._id`.
  Such indexes could be created with previous versions of ArangoDB, but they were non-functional.
  From now on, such indexes cannot be created anymore, and any attempts to create them will fail.

* Added option `--replication.max-parallel-tailing-invocations` to limit the maximum number
  of concurrent WAL tailing invocations.

  The option can be used to limit the usage of the WAL tailing APIs in order to control
  server load.

* Fixed agency bug with election lock step.

* Fixed some error reporting and logging in Maintenance.

* Fixed an error condition in which an ex-leader for a short still believed
  to be the leader and wrongly reported to Current.


v3.5.0-rc.4 (2019-06-15)
------------------------

* Speed up collection creation process in cluster, if not all agency callbacks are
  delivered successfully.

* Increased performance of document inserts, by reducing the number of checks in unique / primary indexes.

* Fixed a callback function in the web UI where the variable `this` was out of scope.

* Fixed editing a user within the web UI if the user added a gravatar profile picture.

* Allow pregel to select the shard key via `shardKeyAttribute` in pregel start parameters.

* Added --server.jwt-secret-keyfile to arangosh.

* Fixed internal issue #4040: gharial api is now checking existence of `_from` and `_to` vertices
  during edge replacements and edge updates.

* Fixed `Buffer.alloc` method.

* `Buffer` is now iterable and accepts `ArrayBuffer` values as input.

* Fix timeout-response in case of simultaneous index create/drop in cluster.

* Enabled dropping an index by its name.

* Fixed lookup of index from collection by fully qualified name, e.g.
  `db.testCollection.index('testCollection/primary')`.

* Fixed agency bug with TTL object writes discovered in 3.4.6.


v3.5.0-rc.3 (2019-05-31)
------------------------

* Fixed issue #9106: Sparse Skiplist Index on multiple fields not used for FILTER + SORT query.

  Allow AQL query optimizer to use sparse indexes in more cases, specifically when
  indexes could not be used for filtering and there finally was an `EnumerateCollectionNode`
  in the query execution plan followed by a `SortNode`. In this case, sparse indexes were
  not considered for enumeration in sorted order, because it was unclear to the optimizer
  if the result set would contain null values or not.

* Upgraded RocksDB to version 6.2.

* Updated ArangoDB Starter to 0.14.4.

* The system collection '_jobs' will from now on be created with non-unique, non-sparse indexes.

* Bugfix for smart graph traversals with uniqueVertices: path, which could
  sometimes lead to erroneous traversal results.

* Pregel algorithms can be run with the option "useMemoryMaps: true" to be
  able to run algorithms on data that is bigger than the available RAM.

* Fixed a race in TTL thread deactivation/shutdown.

* Fixed internal issue #3919: The web UI is now using precompiled ejs templates.

* Fixed agency issue in abort of cleanOutServer job.

v3.5.0-rc.2 (2019-05-23)
------------------------

* Fixed "collection not found" exception during setup of 3-way SmartJoin queries in the
  cluster.

* Fixed an edge case of handling `null` values in the AQL function `MIN` for input
  sequences that started with a `null` value. In this case, `null` was always returned as the
  minimum value even though other non-null values may have followed, and `MIN` was supposed
  to return `null` only if there are no input values or all input values are `null`.

* Fixed a crash when posting an async request to the server using the "x-arango-async"
  request header and the server's request queue was full.

* Added error code 1240 "incomplete read" for RocksDB-based reads which cannot retrieve
  documents due to the RocksDB block cache being size-restricted (with size limit enforced)
  and uncompressed data blocks not fitting into the block cache.

  The error can only occur for collection or index scans with the RocksDB storage engine
  when the RocksDB block cache is used and set to a very small size, plus its maximum size is
  enforced by setting the `--rocksdb.enforce-block-cache-size-limit` option to `true`.

  Previously these incomplete reads could have been ignored silently, making collection or
  index scans return less documents than there were actually present.

* Fixed internal issue #3918: added optional second parameter "withId" to AQL
  function PREGEL_RESULT.

  This parameter defaults to `false`. When set to `true` the results of the Pregel
  computation run will also contain the `_id` attribute for each vertex and not
  just `_key`. This allows distinguishing vertices from different vertex collections.

* Made Swagger UI work with HTTPS-enabled ArangoDBs too.

* Improved scheduler performance for single-connection cases.

* Internally switched unit tests framework from catch to gtest.

* Disabled selection of index types "hash" and "skiplist" in the web interface when
  using the RocksDB engine. The index types "hash", "skiplist" and "persistent" are
  just aliases of each other with the RocksDB engine, so there is no need to offer all
  of them. After initially only offering "hash" indexes, we decided to only offer
  indexes of type "persistent", as it is technically the most
  appropriate description.

* Fixed client id lookup table in state.


v3.5.0-rc.1 (2019-05-14)
------------------------

* Removed bug during start up with a single agent, that leads to dbserver crash.

* Fixed issue #7011: description when replacing a foxx application was misleading.

* Fixed issue #8841: Graph Viewer dropped ability to edit an edge after
  rerendering.

* Upgraded arangodb starter version to 0.14.3.

* ArangoQueryStreamCursor.prototype.id needs to be a string, v8 32 bit integers
  can't hold the full data.

* Upgraded Swagger UI to 3.22.1.

* Added --compress-output flag to arangodump. Activates gzip compression for
  collection data. Metadata files, such as .structure.json and .view.json,
  do not get compressed. No option is needed for arangorestore to restore
  .data.json.gz files.

* Added options to make server more secure:

  - `--server.harden`: denies access to certain REST APIs that return server internals
  - `--foxx.api`: set to false disables Foxx management API
  - `--foxx.store`: set to false disables Foxx UI
  - `--javascript.allow-port-testing`: enables internal.testPort()
  - `--javascript.allow-external-process-control`: enables external process control
  - `--javascript.harden`: disables getPid() and logLevel()
  - `--javascript.startup-options-whitelist`: control startup options visible in JavaScript
  - `--javascript.environment-variables-whitelist`: control environment variables visible in JavaScript
  - `--javascript.endpoints-whitelist`: control accessible endpoints in JavaScript
  - `--javascript.files-whitelist`: control file access in JavaScript

  Note: There is a [detailed description of all options](https://www.arangodb.com/docs/devel/security-security-options.html).

* Prevented arangod from making a call to www.arangodb.com at startup.

  This call was done to check for available updates, but it could have contributed
  to small startup delays in case outgoing connections were blocked.

* Removed support for undocumented HTTP header `x-arango-v8-context`, which
  allowed controlling in which particular V8 context number a JavaScript-based
  action was executed. This header was only used internally for testing.

* `db._query` now handles additional arguments correctly when passing an AQL
  query object instead of a query string and separate bindVars.

* Added req.auth property to Foxx.

* Added collection.documentId method to derive document id from key.

* Indexes created with the 'inBackground', will not hold an
  exclusive collection lock for the entire index creation period (rocksdb only).

* Fixed internal issue #536: ArangoSearch may crash server during term lookup.

* Fixed internal issue #2946: Create graph autocomplete was not working under
  certain circumstances.

* Added `filter` option to Foxx HTTP API for running tests.

* Added function `db.<collection>.getResponsibleShard()` to find out which is the
  responsible shard for a given document. Note that this function is only available
  in a cluster coordinator.

* Updated bundled version of jemalloc memory allocator to 5.2.0.

* Don't create per-database system collection `_frontend` automatically.
  This collection is only needed by the web UI, and it can be created lazily
  when needed.

* Added logging option `--log.time-format` to configure the time format used
  in log output. The possible values for this option are:

  - timestamp: unix timestamps, in seconds
  - timestamp-millis: unix timestamps, in seconds, with millisecond precision
  - timestamp-micros: unix timestamps, in seconds, with microsecond precision
  - uptime: seconds since server start
  - uptime-millis: seconds since server start, with millisecond precision
  - uptime-micros: seconds since server start, with microsecond precision
  - utc-datestring: UTC-based date and time in format YYYY-MM-DDTHH:MM:SSZ
  - utc-datestring-millis: UTC-based date and time in format YYYY-MM-DDTHH:MM:SS.FFFZ
  - local-datestring: local date and time in format YYYY-MM-DDTHH:MM:SS

  This change deprecates the existing options `--log.use-microtime` and
  `--log.use-localtime`, because the functionality provided by these options
  is covered by `--log.time-format` too.

* Added "SmartJoins" to the ArangoDB Enterprise Edition that allows running cluster
  joins between two certain sharded collections with performance close to that of a
  local join operation.

* Fixed internal issue #3815: fixed the removal of connected edges when
  removing a vertex graph node in a smart graph environment.

* Show startup warning in case kernel setting `vm.overcommit_memory` is set
  to a value of 2 and the jemalloc memory allocator is in use. This combination
  does not play well together.

* Added AQL function DECODE_REV for decomposing `_rev` values into their
  individual parts.

* Added AQL functions CRC32 and FNV64 for hashing data.

* Renamed attribute key `openssl-version` in server/client tool version
  details output to `openssl-version-compile-time`.

  This change affects the output produced when starting one of the ArangoDB
  executables with the `--version` command. It also changes the attribute
  name in the detailed response of the `/_api/version` REST API.

* Fixed the sorting of the databases in the database
  selection dropdown in the web ui. The sort order differed based on whether
  authentication was enabled or disabled. (Internal issue #2276)

* Improved the shards view in the web ui if there
  is only one shard to display. (Internal issue #3546)

* Restricted the allowed query names for user
  defined custom queries within the web ui. (Internal issue #3789)

* Upgraded bundled RocksDB version to 6.0.

* Added "--log.ids" option to arangod.

  The default value for this option is `true`. Setting the option to `false`
  will prevent embedding unique log ids into all log messages generated by
  ArangoDB C++ code. The unique ids allow for easy access to the location in
  the source code from which a message originates. This should help customers
  to configure custom monitoring/alerting based on specific log id occurrences
  and will also be helpful for support and development in identifying problems.

* Fixed wrong equals behavior on arrays with ArangoSearch. (Internal issue #8294)

* Fixed ArangoSearch range query sometimes not working
  correctly with numeric values. (Internal issue #528)

* Changed type of the startup option `--rocksdb.recycle-log-file-num` from
  numeric to boolean, as this is also the type the options has in the RocksDB
  library.

* Renamed hidden startup option `--rocksdb.delayed_write_rate` to the more
  consistent `--rocksdb.delayed-write-rate`. When the old option name is
  used, the arangod startup will be aborted with a descriptive error message.

* If not explicitly configured, make agency nodes start removing their unused
  WAL files a few seconds after the completed server startup already. This is
  because on agency nodes, unused WAL files do not need to be retained for
  potential replication clients to read them.

* Added option `--all-databases` to arangodump and arangorestore.

  When set to true, this makes arangodump dump all available databases
  the current user has access to. The option `--all-databases` cannot be
  used in combination with the option `--server.database`.

  When `--all-databases` is used, arangodump will create a subdirectory
  with the data of each dumped database. Databases will be dumped one
  after the other. However, inside each database, the collections of the
  database can be dumped in parallel using multiple threads.

  For arangorestore, this makes it restore all databases from inside the
  subdirectories of the specified dump directory. Using the option for
  arangorestore only makes sense for dumps created with arangodump and
  the `--all-databases` option. As for arangodump, arangorestore cannot
  be invoked with the options `--all-databases` and `--server.database`
  at the same time. Additionally, the option `--force-same-database` cannot
  be used together with `--all-databases`.

  If the to-be-restored databases do not exist on the target server, then
  restoring data into them will fail unless the option `--create-database`
  is also specified. Please note that in this case a database user must
  be used that has access to the `_system` database, in order to create
  the databases on restore.

* Added index hints feature to AQL.

* Added "name" property for indices.

  If a name is not specified on index creation, one will be auto-generated.

* Under normal circumstances there should be no need to connect to a
  database server in a cluster with one of the client tools, and it is
  likely that any user operations carried out there with one of the client
  tools may cause trouble.

  The client tools arangosh, arangodump and arangorestore will now emit
  a warning when connecting with them to a database server node in a cluster.

* Fixed compaction behavior of followers.

* Added "random" masking to mask any data type, added wildcard masking.

* Added option `--console.history` to arangosh for controlling whether
  the command-line history should be loaded from and persisted in a file.

  The default value for this option is `true`. Setting it to `false`
  will make arangosh not load any command-line history from the history
  file, and not store the current session's history when the shell is
  exited. The command-line history will then only be available in the
  current shell session.

* Display the server role when connecting arangosh against a server (e.g.
  SINGLE, COORDINATOR).

* Fixed overflow in Windows NowNanos in RocksDB.

* Allowed MoveShard from leader to a follower, thus swapping the two.

* Supervision fix: Satellite collections, various fixes.

* Added coordinator route for agency dump.

* Supervision fix: abort MoveShard job does not leave a lock behind.

* Supervision fix: abort MoveShard (leader) job moves forwards when point
  of no return has been reached.

* Supervision fix: abort CleanOutServer job does not leave server in
  ToBeCleanedServers.

* Supervision fix: move shard with data stopped to early due to wrong usage
  of compare function.

* Supervision fix: AddFollower only counts good followers, fixing a
  situation after a FailedLeader job could not find a new working
  follower.

* Supervision fix: FailedLeader now also considers temporarily BAD
  servers as replacement followers and does not block servers which
  currently receive a new shard.

* Supervision fix: Servers in ToBeCleanedServers are no longer considered
  as replacement servers.

* Maintenance fix: added precondition of unchanged Plan in phase2.

* Added "PRUNE <condition>" to AQL Traversals. This allows to early abort searching of
  unnecessary branches within a traversal.
  PRUNE is only allowed in the Traversal statement and only between the graph
  definition and the options of the traversal.
  e.g.:
  `FOR v, e, p IN 1..3 OUTBOUND @source GRAPH "myGraph"
    PRUNE v.value == "bar"
    OPTIONS {} /* These options remain optional */
    RETURN v`
  for more details refer to the documentation chapter.

* Fixed a display issue when editing a graph within the web UI.

* Fixed some escaping issues within the web UI.

* Follow up to fix JWT authentication in arangosh (Internal issue #7530):
  also fixed reconnect.

* Now also syncing _jobs and _queues collections in active failover mode.

* Upgraded lodash to 4.17.11 because of CVE-2018-16487.

* `--query.registry-ttl` is now honored in single-server mode, and cursor TTLs
  are now honored on DBServers in cluster mode.

* Added "TTL" index type, for optional auto-expiration of documents.

* Disabled selection of index types "persistent" and "skiplist" in the web
  interface when using the RocksDB engine. The index types "hash", "skiplist"
  and "persistent" are just aliases of each other with the RocksDB engine,
  so there is no need to offer all of them.

* Fixed JS AQL query objects with empty query strings not being recognized
  as AQL queries.

* Updated JavaScript dependencies, including semver major updates to joi, mocha
  and eslint. The eslint config modules were incompatible with the existing
  coding style, so the old rules were inlined and the config dependencies removed.

  Note that joi removed Joi.date().format() in v10.0.0. You can bundle your own
  version of joi if you need to rely on version-dependent features.

  - accepts: 1.3.4 -> 1.3.5
  - ansi_up: 2.0.2 -> 4.0.3
  - content-disposition: 0.5.2 -> 0.5.3
  - dedent: 0.6.0 -> 0.7.0
  - error-stack-parser: 1.3.6 -> 2.0.2
  - eslint: 2.13.1 -> 5.16.0
  - eslint-config-semistandard: 6.0.2 -> removed
  - eslint-config-standard: 5.3.1 -> removed
  - eslint-plugin-promise: 1.3.2 -> removed
  - eslint-plugin-standard: 1.3.2 -> removed
  - highlight.js: 9.12.0 -> 9.15.6
  - http-errors: 1.6.2 -> 1.7.2
  - iconv-lite: 0.4.19 -> 0.4.24
  - joi: 9.2.0 -> 14.3.1
  - joi-to-json-schema: 2.3.0 -> 4.0.1
  - js-yaml: 3.10.0 -> 3.13.1
  - marked: 0.3.9 -> 0.6.2
  - mime-types: 2.1.12 -> 2.1.22
  - mocha: 2.5.3 -> 6.1.3
  - qs: 6.5.1 -> 6.7.0
  - semver: 5.4.1 -> 6.0.0
  - statuses: 1.4.0 -> 1.5.0
  - timezone: 1.0.13 -> 1.0.22
  - type-is: 1.6.15 -> 1.6.16
  - underscore: 1.8.3 -> 1.9.1

* Updated V8 to 7.1.302.28.

  New V8 behavior introduced herein:

  - ES2016 changed the default timezone of date strings to be conditional on
    whether a time part is included. The semantics were a compromise approach
    based on web compatibility feedback from V8, but until now, we have been
    shipping ES5.1 default timezone semantics. This patch implements the
    new semantics, following ChakraCore and SpiderMonkey (though JSC
    implements V8's previous semantics).

* Fixed JS AQL query objects with empty query strings not being recognized as AQL queries.

* Report run-time openssl version (for dynamically linked executables).

* Added greeting warning about maintainer mode.

* Improved insertion time into non-unique secondary indexes with the RocksDB
  engine.

* Fixed possible segfault when using COLLECT with a LIMIT and an offset.

* Fixed COLLECT forgetting top-level variables after 1000 rows.

* Added sort-limit optimization in AQL; Improved memory usage and execution
  time for some queries.

* Upgraded to OpenSSL 1.1.0j.

* Added configurable masking of dumped data via `arangodump` tool to obfuscate
  exported sensible data.

* Fixed arangoimp script for MacOSX CLI Bundle.

* Added "peakMemoryUsage" in query results figures, showing the peak memory
  usage of the executed query. In a cluster, the value the peak memory usage
  of all shards, but it is not summed up across shards.

* Fixed an issue where a crashed coordinator can lead to some Foxx queue jobs
  erroneously either left hanging or being restarted.

* Fixed bind values of `null` are not replaced by
  empty string anymore, when toggling between json and table
  view in the web-ui. (Interal issue #7900)

* Fixed regression on ISO8601 string compatibility in AQL. (Internal issue #2786)

  millisecond parts of AQL date values were limited to up to 3 digits.
  Now the length of the millisecond part is unrestricted, but the
  millisecond precision is still limited to up to 3 digits.

* The RocksDB primary index can now be used by the optimizer to optimize queries
  that use `_key` or `_id` for sorting or for range queries.

* The web UI will now by default show the documents of a collection lexicographically
  sorted when sorting documents by their `_key` values.

  Previous versions of ArangoDB tried to interpret `_key` values as numeric values if
  possible and sorted by these. That previous sort strategy never used an index and
  could have caused unnecessary overhead. The new version will now use an index for
  sorting for the RocksDB engine, but may change the order in which documents are
  shown in the web UI (e.g. now a `_key` value of "10" will be shown before a `_key`
  value of "9").

* Fixed known issue #445: ArangoSearch ignores `_id` attribute even if `includeAllFields`
  is set to `true`.

* Upgraded bundled boost library to version 1.69.0.

* Upgraded bundled curl library to version 7.63.

* Used base64url to encode and decode JWT parts.

* Added --server.jwt-secret-keyfile option.

* Sped up data-modification operations in exclusive transactions in the RocksDB
  storage engine.

* An AQL query that uses the edge index only and returns the opposite side of
  the edge can now be executed in a more optimized way, e.g.

    FOR edge IN edgeCollection FILTER edge._from == "v/1" RETURN edge._to

  is fully covered by RocksDB edge index. For MMFiles this rule does not apply.

* Reverted accidental change to error handling in geo index.

  In previous versions, if non-valid geo coordinates were contained in the
  indexed field of a document, the document was simply ignored an not indexed.
  In 3.4.0, this was accidentally changed to generate an error, which caused
  the upgrade procedure to break in some cases.

* Fixed TypeError being thrown instead of validation errors when Foxx manifest
  validation fails.

* Improved confirmation dialog when clicking the
  Truncate button in the Web UI. (Internal issue #2786)

* Made `--help-all` now also show all hidden program options.

  Previously hidden program options were only returned when invoking arangod or
  a client tool with the cryptic `--help-.` option. Now `--help-all` simply
  retuns them as well.

  The program options JSON description returned by `--dump-options` was also
  improved as follows:

  - the new boolean attribute "dynamic" indicates whether the option has a dynamic
    default value, i.e. a value that depends on the target host capabilities or
    configuration

  - the new boolean attribute "requiresValue" indicates whether a boolean option
    requires a value of "true" or "false" when specified. If "requiresValue" is
    false, then the option can be specified without a boolean value following it,
    and the option will still be set to true, e.g. `--server.authentication` is
    identical to `--server.authentication true`.

  - the new "category" attribute will contain a value of "command" for command-like
    options, such as `--version`, `--dump-options`, `--dump-dependencies` etc.,
    and "option" for all others.

* Fixed issue #7586: a running query within the user interface was not shown
  if the active view was `Running Queries` or `Slow Query History`.

* Fixed issue #7743: Query processing discrepancy between Rocks and MMFiles databases.

  This change enforces the invalidation of variables in AQL queries after usage of
  a COLLECT statement as documented. The documentation for variable invalidation claims
  that

      The COLLECT statement will eliminate all local variables in the current scope.
      After COLLECT only the variables introduced by COLLECT itself are available.

  However, the described behavior was not enforced when a COLLECT was preceded by a
  FOR loop that was itself preceded by a COLLECT. In the following query the final
  RETURN statement accesses variable `key1` though the variable should have been
  invalidated by the COLLECT directly before it:

      `FOR x1 IN 1..2
        COLLECT key1 = x1
        FOR x2 IN 1..2
          COLLECT key2 = x2
          RETURN [key2, key1]`

  In previous releases, this query was
  parsed ok, but the contents of variable `key1` in the final RETURN statement were
  undefined.

  This change is about making queries as the above fail with a parse error, as an
  unknown variable `key1` is accessed here, avoiding the undefined behavior. This is
  also in line with what the documentation states about variable invalidation.

* Fixed issue #7763: Collect after update does not execute updates.

* Fixed issue #7749: AQL query result changed for COLLECT used on empty data/array.

* Fixed issue #7757: Using multiple filters on nested objects produces wrong results.

* Fixed a rare thread local dead lock situation in replication:
  If a follower tries to get in sync in the last steps it requires
  a lock on the leader. If the follower cancels the lock before the leader
  has succeeded with locking we can end up with one thread being deadlocked.

* Allowed usage of floating point values in AQL without leading zeros, e.g.
  `.1234`. Previous versions of ArangoDB required a leading zero in front of
  the decimal separator, i.e `0.1234`.

* Foxx `req.makeAbsolute` now will return meaningful values when ArangoDB is using
  a unix socket endpoint. URLs generated when using a unix socket follow the format
  http://unix:<socket-path>:<url-path> used by other JS tooling.

* Updated joi library (Web UI), improved foxx mount path validation.

* Do not create `_routing` collection for new installations/new databases,
  as it is not needed anymore. Redirects to the web interface's login screen, which
  were previously handled by entries in the `_routing` collection are now handled
  from the responsible REST action handler directly.

  Existing `_routing` collections will not be touched as they may contain other
  entries as well, and will continue to work.

* Do not create `_modules` collection for new databases/installations.

  `_modules` is only needed for custom modules, and in case a custom
  module is defined via `defineModule`, the _modules collection will
  be created lazily automatically.

  Existing modules in existing `_modules` collections will remain
  functional even after this change

* Disabled in-memory cache for edge and traversal data on agency nodes, as it
  is not needed there.

* Removed bundled Valgrind headers, removed JavaScript variable `valgrind`
  from the `internal` module.

* Upgraded JEMalloc version to 5.1.0.

* Use `-std=c++14` for ArangoDB compilation.

* In case of resigned leader, set `isReady=false` in clusterInventory.

* Abort RemoveFollower job if not enough in-sync followers or leader failure.

* Fixed shrinkCluster for satelliteCollections.

* Fixed crash in agency supervision when leadership is lost.

* Leader's timestamps in append entries protocol.

* Sped up agency supervision in case of many jobs.

* Fixed log spam in agency supervision when leader resigned.

* Made AddFollower less aggressive.

* Several agency performance improvements, mostly avoiding copying.

* Priority queue for maintenance jobs.

* Do not wait for replication after each job execution in Supervision.

* Added support for configuring custom Analyzers via JavaScript and REST.


v3.4.7 (2019-07-02)
-------------------

* updated arangosync to 0.6.4

* bug-fix for a very rare race condition on cluster collection creation process. It can only occur in the following
  situation:
  1) DBServer sucessfully creates all shards for collections, and reports it back
  2) DBServer dies
  3) Agency recognized death and reorganizes shards
  4) Coordinator receives (1) but not (3) and decided everything was good
  5) => Coordinator reverts (3) only on this collection
  6) Coordinator receives (3) now
  The bugfix disallows (5) if (3) is run in Agency.

* fixed internal issue #4040: gharial api is now checking existence of `_from` and `_to` vertices
  during edge replacements and edge updates

* fix timeout-response in case of simultaneous index create/drop in cluster

* Fixed editing a user within the web UI if the user added a gravatar profile picture

* fixed internal issue #4172: on agent servers the RocksDB WAL files in the archive directory
  were retained due to an error. Now they are removed, eventually

* fix error reporting in arangorestore. previously, when the server returned an HTTP 404 error,
  arangorestore unconditionally translated this into a "collection not found" error, which may
  have masked an actually different server-side error. Instead, make arangorestore return the
  server's actual error message as is.

* allow pregel to select the shard key via `shardKeyAttribute` in pregel start parameters

* Speed up collection creation process in cluster, if not all agency callbacks are
  delivered successfully.

* Fixed parsing of ArangoDB config files with inlined comments. Previous versions didn't handle
  line comments properly if they were appended to an otherwise valid option value.

  For example, the comment in the line

      max-total-wal-size = 1024000 # 1M

  was not ignored and made part of the value. In the end, the value was interpreted as if

      max-total-wal-size = 10240001000000

  was specified.
  This version fixes the handling of comments in the config files so that they behave as intended.

* Fixed AQL query tracing for traversals, this lead to wrong "calls" count in
  query profiles

* Pregel algorithms can be run with the option "useMemoryMaps: true" to be able to run algorithms
  on data that is bigger than the available RAM.

* Bugfix for smart graph traversals with uniqueVertices: path, which could
  sometimes lead to erroneous traversal results

* The system-collection '_jobs' will from now on use non-unique, non-sparse indexes.

* fixed internal issue #3919: The web UI is now using precompiled ejs templates.

* properly create VERSION files for any newly created databases again, as it has been the
  case in 3.3 and before. If no VERSION files are created, this may cause problems on upgrade
  to 3.5. The 3.5 branch was also modified to handle this problem gracefully.

* fixed "collection not found" exception during setup of 3-way SmartJoin queries in the
  cluster

* fixed an edge case of handling `null` values in the AQL function `MIN` for input
  sequences that started with a `null` value. In this case, `null` was always returned as the
  minimum value even though other non-null values may have followed, and `MIN` was supposed
  to return `null` only if there are no input values or all input
  values are `null`.

* fix client id lookup table in agency

* Fix agency's handling of object assignments with TTL

* reduce resident memory usage when creating non-unique indexes with the RocksDB storage engine

* Fix agency election bug.


v3.4.6.1 (2019-06-05)
---------------------

* upgraded ArangoDB Starter version to 0.14.5

* Speed up collection creation process in cluster, if not all agency callbacks are
  delivered successfully.

* Fix agency's handling of object assignments with TTL


v3.4.6 (2019-05-21)
-------------------

* updated ArangoDB Starter to 0.14.4

* fixed internal issue #3912: improved the performance of graph creation with multiple
  relations. They do now create multiple collections within a single round-trip in the agency.

* fixed a crash when posting an async request to the server using the "x-arango-async"
  request header and the server's request queue was full

* added error code 1240 "incomplete read" for RocksDB-based reads which cannot retrieve
  documents due to the RocksDB block cache being size-restricted (with size limit enforced)
  and uncompressed data blocks not fitting into the block cache

  The error can only occur for collection or index scans with the RocksDB storage engine
  when the RocksDB block cache is used and set to a very small size, plus its maximum size is
  enforced by setting the `--rocksdb.enforce-block-cache-size-limit` option to `true`.

  Previously these incomplete reads could have been ignored silently, making collection or
  index scans return less documents than there were actually present.

* fixed internal issue #3918: added optional second parameter "withId" to AQL
  function PREGEL_RESULT

  this parameter defaults to `false`. When set to `true` the results of the Pregel
  computation run will also contain the `_id` attribute for each vertex and not
  just `_key`. This allows distinguishing vertices from different vertex collections.

* fixed issue #7011: description when replacing a foxx application was misleading

* fixed issue #8841: Graph Viewer dropped ability to edit an edge after
  rerendering.

* upgraded arangodb starter version to 0.14.3

* fix removal of sort order for sorted results that were passed into a RETURN DISTINCT
  in an AQL query running on a collection with more than a single shard. In this case,
  the results from the individual shards were sorted, but the overall result was not
  sorted according to the sort criterion.

* removed bug during start up with a single agent, that leads to dbserver crash.

* fix the creation of debug packages (in the web interface) for queries that
  involve smart graphs and/or multiple edge collections from a traversal

* add --compress-output flag to arangodump. Activates gzip compression for
  collection data. Metadata files, such as .structure.json and .view.json,
  do not get compressed. No option is needed for arangorestore to restore
  .data.json.gz files.

* fixed internal issue #3893: correct wrong option name `--database.ignore-logfile-errors`
  in error message and correct it to `--database.ignore-datafile-errors`

* fixed internal issue #2946: create graph autocomplete was not working under
  certain circumstances.

* fixed a memory leak in PRUNE operation for traversals.
  The leak occurred only when PRUNE was actively used in queries.

* fixed a crash (SIGSEGV) when opening a RocksDB database directory that
  contained an empty (0 bytes filesize) journal file and encryption was in use.

* improve test performance by doing the 0.1 delay between phase 1 and 2 only
  if phase 1 is slow

* port back timestamp replication in agency from devel

* better logging when scheduler queue is half-full and full

* fixed internal issue #536: ArangoSearch query may crash during internal lookup in some
  cases due to invalid index structure for exact input data


v3.4.5 (2019-03-27)
-------------------

* fixed a shutdown issue when the server was shut down while there were
  active Pregel jobs executing

* fixed internal issue #3815: fixed the removal of connected edges when
  removing a vertex graph node in a smart graph environment.

* added AQL functions CRC32 and FNV64 for hashing data

* internal issue #2276: fixed the sorting of the databases in the database
  selection dropdown in the web ui. The sort order differed based on whether
  authentication was enabled or disabled.

* fixed internal issue #3789: restricted the allowed query names for user-
  defined custom queries within the web ui.

* fixed internal issue #3546: improved the shards view in the web ui if there
  is only one shard to display.

* fixed a display issues when editing a graph within the web ui

* fixed internal-issue #3787: Let user know about conflicting attribute in
  AQL queries if information is available.

* fixed issue #8294: wrong equals behavior on arrays with ArangoSearch

* fixed internal issue #528: ArangoSearch range query sometimes doesn't work
  correctly with numeric values

* fixed internal issue #3757: when restarting a follower in active failover mode,
  try an incremental sync instead of a full resync. Fixed also the case where a
  double resync was made

* don't check for the presence of ArangoDB upgrades available when firing up an
  arangosh enterprise edition build

* added startup option `--rocksdb.allow-fallocate`

  When set to true, allows RocksDB to use the fallocate call. If false, fallocate
  calls are bypassed and no preallocation is done. Preallocation is turned on by
  default, but can be turned off for operating system versions that are known to
  have issues with it.
  This option only has an effect on operating systems that support fallocate.

* added startup option `--rocksdb.limit-open-files-at-startup`

  If set to true, this will limit the amount of .sst files RocksDB will inspect at
  startup, which can reduce the number of IO operations performed at start.

* don't run compact() on a collection after a truncate() was done in the same
  transaction

  running compact() in the same transaction will only increase the data size on
  disk due to RocksDB not being able to remove any documents physically due to the
  snapshot that is taken at transaction start.

  This change also exposes db.<collection>.compact() in the arangosh, in order to
  manually run a compaction on the data range of a collection should it be needed
  for maintenance.

* don't attempt to remove non-existing WAL files, because such attempt will
  trigger unnecessary error log messages in the RocksDB library

* updated arangosync to 0.6.3

* added --log.file-mode to specify a file mode of newly created log files

* added --log.file-group to specify the group of newly created log files

* fixed some escaping issues within the web ui.

* fixed issue #8359: How to update a document with unique index (constraints)?

* when restarting a follower in active failover mode, try an incremental sync instead
  of a full resync

* add "PRUNE <condition>" to AQL Traversals (internal issue #3068). This allows to early
  abort searching of unnecessary branches within a traversal.
  PRUNE is only allowed in the Traversal statement and only between the graphdefinition
  and the options of the traversal.
  e.g.:

      FOR v, e, p IN 1..3 OUTBOUND @source GRAPH "myGraph"
        PRUNE v.value == "bar"
        OPTIONS {} /* These options remain optional */
        RETURN v

  for more details refer to the documentation chapter.

* added option `--console.history` to arangosh for controlling whether
  the command-line history should be loaded from and persisted in a file.

  The default value for this option is `true`. Setting it to `false`
  will make arangosh not load any command-line history from the history
  file, and not store the current session's history when the shell is
  exited. The command-line history will then only be available in the
  current shell session.

* display the server role when connecting arangosh against a server (e.g.
  SINGLE, COORDINATOR)

* added replication applier state figures `totalDocuments` and `totalRemovals` to
  access the number of document insert/replace operations and the number of document
  removals operations separately. Also added figures `totalApplyTime` and
  `totalFetchTime` for determining the total time the replication spent for
  applying changes or fetching new data from the master. Also added are the figures
  `averageApplyTime` and `averageFetchTime`, which show the average time spent
  for applying a batch or for fetching data from the master, resp.

* fixed race condition in which the value of the informational replication applier
  figure `ticksBehind` could underflow and thus show a very huge number of ticks.

* always clear all ongoing replication transactions on the slave if the slave
  discovers the data it has asked for is not present anymore on the master and the
  `requireFromPresent` value for the applier is set to `false`.

  In this case aborting the ongoing transactions on the slave is necessary because
  they may have held exclusive locks on collections, which may otherwise not be
  released.

* added option `--rocksdb.wal-archive-size-limit` for controlling the
  maximum total size (in bytes) of archived WAL files. The default is 0
  (meaning: unlimited).

  When setting the value to a size bigger than 0, the RocksDB storage engine
  will force a removal of archived WAL files if the total size of the archive
  exceeds the configured size. The option can be used to get rid of archived
  WAL files in a disk size-constrained environment.
  Note that archived WAL files are normally deleted automatically after a
  short while when there is no follower attached that may read from the archive.
  However, in case when there are followers attached that may read from the
  archive, WAL files normally remain in the archive until their contents have
  been streamed to the followers. In case there are slow followers that cannot
  catch up this will cause a growth of the WAL files archive over time.
  The option `--rocksdb.wal-archive-size-limit` can now be used to force a
  deletion of WAL files from the archive even if there are followers attached
  that may want to read the archive. In case the option is set and a leader
  deletes files from the archive that followers want to read, this will abort
  the replication on the followers. Followers can however restart the replication
  doing a resync.

* agents need to be able to overwrite a compacted state with same _key

* in case of resigned leader, set isReady=false in clusterInventory

* abort RemoveFollower job if not enough in-sync followers or leader failure

* fix shrinkCluster for satelliteCollections

* fix crash in agency supervision when leadership is lost

* speed up supervision in agency for large numbers of jobs

* fix log spamming after leader resignation in agency

* make AddFollower less aggressive

* fix cases, where invalid json could be generated in agents' store dumps

* coordinator route for full agency dumps contains compactions and
  time stamps

* lots of agency performance improvements, mostly avoiding copying

* priority queue for maintenance jobs

* do not wait for replication after each job execution in Supervision

* fix a blockage in MoveShard if a failover happens during the operation

* check health of servers in Current before scheduling removeFollower jobs

* wait for statistics collections to be created before running resilience
  tests

* fix ttl values in agency when key overwritten with no ttl


v3.4.4 (2019-03-12)
-------------------

* added missing test for success in failed leader: this could lead
  to a crash

* follow up to fix JWT authentication in arangosh (#7530):
  also fix reconnect

* now also syncing _jobs and _queues collections in active failover mode

* fixed overflow in Windows NowNanos in RocksDB

* fixed issue #8165: AQL optimizer does not pick up multiple geo index


* when creating a new database with an initial user, set the database permission
  for this user as specified in the documentation

* Supervision fix: abort MoveShard job does not leave a lock behind,

* Supervision fix: abort MoveShard (leader) job moves forwards when point
  of no return has been reached,

* Supervision fix: abort CleanOutServer job does not leave server in
  ToBeCleanedServers,

* Supervision fix: move shard with data stopped to early due to wrong usage
  of compare function

* Supervision fix: AddFollower only counts good followers, fixing a
  situation after a FailedLeader job could not find a new working
  follower

* Supervision fix: FailedLeader now also considers temporarily BAD
  servers as replacement followers and does not block servers which
  currently receive a new shard

* Supervision fix: Servers in ToBeCleanedServers are no longer considered
  as replacement servers

* Maintenance fix: added precondition of unchanged Plan in phase2

* Allow MoveShard from leader to a follower, thus swapping the two

* Supervision fix: Satellite collections, various fixes

* Add coordinator route for agency dump

* speed up replication of transactions containing updates of existing documents.

  The replication protocol does not provide any information on whether a document
  was inserted on the master or updated/replaced. Therefore the slave will always
  try an insert first, and move to a replace if the insert fails with "unique
  constraint violation". This case is however very costly in a bigger transaction,
  as the rollback of the insert will force the underlying RocksDB write batch to be
  entirely rewritten. To circumvent rewriting entire write batches, we now do a
  quick check if the target document already exists, and then branch to either
  insert or replace internally.


v3.4.3 (2019-02-19)
-------------------

* fixed JS AQL query objects with empty query strings not being recognized as AQL queries

* fixed issue #8137: NULL input field generates U_ILLEGAL_ARGUMENT_ERROR

* fixed issue #8108: AQL variable - not working query since upgrade to 3.4 release

* fixed possible segfault when using COLLECT with a LIMIT and an offset

* fixed COLLECT forgetting top-level variables after 1000 rows

* fix undefined behavior when calling user-defined AQL functions from an AQL
  query via a streaming cursor

* fix broken validation of tick range in arangodump

* updated bundled curl library to version 7.63.0

* added "peakMemoryUsage" in query results figures, showing the peak memory
  usage of the executed query. In a cluster, the value contains the peak memory
  usage across all shards, but it is not summed up across shards.

* data masking: better documentation, fixed default phone number,
  changed default range to -100 and 100 for integer masking function

* fix supervision's failed server handling to transactionally create
  all failed leader/followers along


v3.4.2.1 (2019-02-01)
---------------------

* upgrade to new velocypack version


v3.4.2 (2019-01-24)
-------------------

* added configurable masking of dumped data via `arangodump` tool to obfuscate exported sensible data

* upgraded to OpenSSL 1.1.0j

* fixed an issue with AQL query IN index lookup conditions being converted into
  empty arrays when they were shared between multiple nodes of a lookup condition
  that used an IN array lookup in an OR that was multiplied due to DNF transformations

  This issue affected queries such as the following

      FILTER (... && ...) || doc.indexAttribute IN non-empty-array

* upgraded arangodb starter version to 0.14.0

* upgraded arangosync version to 0.6.2

* fixed an issue where a crashed coordinator can lead to some Foxx queue jobs
  erroneously either left hanging or being restarted

* fix issue #7903: Regression on ISO8601 string compatibility in AQL

  millisecond parts of AQL date values were limited to up to 3 digits.
  Now the length of the millisecond part is unrestricted, but the
  millisecond precision is still limited to up to 3 digits.

* fix issue #7900: Bind values of `null` are not replaced by
  empty string anymore, when toggling between json and table
  view in the web-ui.

* Use base64url to encode and decode JWT parts.

* added AQL function `CHECK_DOCUMENT` for document validity checks

* when detecting parse errors in the JSON input sent to the restore API, now
  abort with a proper error containing the problem description instead of aborting
  but hiding there was a problem.

* do not respond with an internal error in case of JSON parse errors detected
  in incoming HTTP requests

* added arangorestore option `--cleanup-duplicate-attributes` to clean up input documents
  with redundant attribute names

  Importing such documents without the option set will make arangorestore fail with an
  error, and setting the option will make the restore process clean up the input by using
  just the first specified value for each redundant attribute.

* the arangorestore options `--default-number-of-shards` and `--default-replication-factor`
  are now deprecated in favor of the much more powerful options `--number-of-shards`
  and `--replication-factor`

  The new options `--number-of-shards` and `--replication-factor` allow specifying
  default values for the number of shards and the replication factor, resp. for all
  restored collections. If specified, these default values will be used regardless
  of whether the number of shards or the replication factor values are already present
  in the metadata of the dumped collections.

  It is also possible to override the values on a per-collection level by specifying
  the options multiple times, e.g.

      --number-of-shards 2 --number-of-shards mycollection=3 --number-of-shards test=4

  The above will create all collections with 2 shards, except the collection "mycollection"
  (3 shards) and "test" (4 shards).

  By omitting the default value, it is also possible to use the number of shards/replication
  factor values from the dump for all collections but the explicitly specified ones, e.g.

      --number-of-shards mycollection=3 --number-of-shards test=4

  This will use the number of shards as specified in the dump, except for the collections
  "mycollection" and "test".

  The `--replication-factor` option works similarly.

* validate uniqueness of attribute names in AQL in cases in which it was not
  done before. When constructing AQL objects via object literals, there was
  no validation about object attribute names being unique. For example, it was
  possible to create objects with duplicate attribute names as follows:

      INSERT { a: 1, a: 2 } INTO collection

  This resulted in a document having two "a" attributes, which is obviously
  undesired. Now, when an attribute value is used multiple times, only the first
  assigned value will be used for that attribute in AQL. It is not possible to
  specify the same attribute multiple times and overwrite the attribute's value
  with by that. That means in the above example, the value of "a" will be 1,
  and not 2.
  This changes the behavior for overriding attribute values in AQL compared to
  previous versions of ArangoDB, as previous versions in some cases allowed
  duplicate attribute names in objects/documents (which is undesired) and in
  other cases used the _last_ value assigned to an attribute instead of the _first_
  value. In order to explicitly override a value in an existing object, use the
  AQL MERGE function.

  To avoid all these issues, users are encouraged to use unambiguous attribute
  names in objects/documents in AQL. Outside of AQL, specifying the same attribute
  multiple times may even result in a parse error, e.g. when sending such data
  to ArangoDB's HTTP REST API.

* fixed issue #7834: AQL Query crashes instance

* Added --server.jwt-secret-keyfile option.

* Improve single threaded performance by scheduler optimization.

* Releveling logging in maintenance


v3.4.1 (2018-12-19)
-------------------

* fixed issue #7757: Using multiple filters on nested objects produces wrong results

* fixed issue #7763: Collect after update does not execute updates

* fixed issue #7586: a running query within the user interface was not shown
  if the active view was `Running Queries` or `Slow Query History`.

* fixed issue #7749: AQL Query result changed for COLLECT used on empty data/array

* fixed a rare thread local dead lock situation in replication:
  If a follower tries to get in sync in the last steps it requires
  a lock on the leader. If the follower cancels the lock before the leader
  has succeeded with locking we can end up with one thread being deadlocked.

* fix thread shutdown in _WIN32 builds

  Previous versions used a wrong comparison logic to determine the current
  thread id when shutting down a thread, leading to threads hanging in their
  destructors on thread shutdown

* reverted accidental change to error handling in geo index

  In previous versions, if non-valid geo coordinates were contained in the
  indexed field of a document, the document was simply ignored an not indexed.
  In 3.4.0, this was accidentally changed to generate an error, which caused
  the upgrade procedure to break in some cases.

* fixed TypeError being thrown instead of validation errors when Foxx manifest
  validation fails

* make AQL REMOVE operations use less memory with the RocksDB storage engine

  the previous implementation of batch removals read everything to remove into
  memory first before carrying out the first remove operation. The new version
  will only read in about 1000 documents each time and then remove these. Queries
  such as

      FOR doc IN collection FILTER ... REMOVE doc IN collection

  will benefit from this change in terms of memory usage.

* make `--help-all` now also show all hidden program options

  Previously hidden program options were only returned when invoking arangod or
  a client tool with the cryptic `--help-.` option. Now `--help-all` simply
  retuns them as well.

  The program options JSON description returned by `--dump-options` was also
  improved as follows:

  - the new boolean attribute "dynamic" indicates whether the option has a dynamic
    default value, i.e. a value that depends on the target host capabilities or
    configuration

  - the new boolean attribute "requiresValue" indicates whether a boolean option
    requires a value of "true" or "false" when specified. If "requiresValue" is
    false, then the option can be specified without a boolean value following it,
    and the option will still be set to true, e.g. `--server.authentication` is
    identical to `--server.authentication true`.

  - the new "category" attribute will contain a value of "command" for command-like
    options, such as `--version`, `--dump-options`, `--dump-dependencies` etc.,
    and "option" for all others.

* Fixed a bug in synchroneous replication intialization for, where a
  shard's db server is rebooted during that period


v3.4.0 (2018-12-06)
-------------------

* Add license key checking to enterprise version in Docker containers.


v3.4.0-rc.5 (2018-11-29)
------------------------

* Persist and check default language (locale) selection.
  Previously we would not check if the language (`--default-language`) had changed
  when the server was restarted. This could cause issues with indexes over text fields,
  as it will resulted in undefined behavior within RocksDB (potentially missing entries,
  corruption, etc.). Now if the language is changed, ArangoDB will print out an error
  message on startup and abort.

* fixed issue #7522: FILTER logic totally broke for my query in 3.4-rc4

* export version and storage engine in `_admin/cluster/health` for Coordinators
  and DBServers.

* restrict the total amount of data to build up in all in-memory RocksDB write buffers
  by default to a certain fraction of the available physical RAM. This helps restricting
  memory usage for the arangod process, but may have an effect on the RocksDB storage
  engine's write performance.

  In ArangoDB 3.3 the governing configuration option `--rocksdb.total-write-buffer-size`
  had a default value of `0`, which meant that the memory usage was not limited. ArangoDB
  3.4 now changes the default value to about 50% of available physical RAM, and 512MiB
  for setups with less than 4GiB of RAM.

* lower default value for `--cache.size` startup option from about 30% of physical RAM to
  about 25% percent of physical RAM.

* fix internal issue #2786: improved confirmation dialog when clicking the truncate
  button in the web UI

* Updated joi library (web UI), improved Foxx mount path validation

* disable startup warning for Linux kernel variable `vm.overcommit_memory` settings
  values of 0 or 1.
  Effectively `overcommit_memory` settings value of 0 or 1 fix two memory-allocation
  related issues with the default memory allocator used in ArangoDB release builds on
  64bit Linux.
  The issues will remain when running with an `overcommit_memory` settings value of 2,
  so this is now discouraged.
  Setting `overcommit_memory` to 0 or 1 (0 is the Linux kernel's default) fixes issues
  with increasing numbers of memory mappings for the arangod process (which may lead
  to an out-of-memory situation if the kernel's maximum number of mappings threshold
  is hit) and an increasing amount of memory that the kernel counts as "committed".
  With an `overcommit_memory` setting of 0 or 1, an arangod process may either be
  killed by the kernel's OOM killer or will die with a segfault when accessing memory
  it has allocated before but the kernel could not provide later on. This is still
  more acceptable than the kernel not providing any more memory to the process when
  there is still physical memory left, which may have occurred with an `overcommit_memory`
  setting of 2 after the arangod process had done lots of allocations.

  In summary, the recommendation for the `overcommit_memory` setting is now to set it
  to 0 or 1 (0 is kernel default) and not use 2.

* fixed Foxx complaining about valid `$schema` value in manifest.json

* fix for supervision, which started failing servers using old transient store

* fixed a bug where indexes are used in the cluster while still being
  built on the db servers

* fix move leader shard: wait until all but the old leader are in sync.
  This fixes some unstable tests.

* cluster health features more elaborate agent records

* agency's supervision edited for advertised endpoints


v3.4.0-rc.4 (2018-11-04)
------------------------

* fixed Foxx queues not retrying jobs with infinite `maxFailures`

* increase AQL query string parsing performance for queries with many (100K+) string
  values contained in the query string

* increase timeouts for inter-node communication in the cluster

* fixed undefined behavior in `/_api/import` when importing a single document went
  wrong

* replication bugfixes

* stop printing `connection class corrupted` in arangosh

 when just starting the arangosh without a connection to a server and running
 code such as `require("internal")`, the shell always printed "connection class
 corrupted", which was somewhat misleading.

* add separate option `--query.slow-streaming-threshold` for tracking slow
  streaming queries with a different timeout value

* increase maximum number of collections/shards in an AQL query from 256 to 2048

* don't rely on `_modules` collection being present and usable for arangod startup

* force connection timeout to be 7 seconds to allow libcurl time to retry lost DNS
  queries.

* fixes a routing issue within the web ui after the use of views

* fixes some graph data parsing issues in the ui, e.g. cleaning up duplicate
  edges inside the graph viewer.

* in a cluster environment, the arangod process now exits if wrong credentials
  are used during the startup process.

* added option `--rocksdb.total-write-buffer-size` to limit total memory usage
  across all RocksDB in-memory write buffers

* suppress warnings from statistics background threads such as
  `WARNING caught exception during statistics processing: Expecting Object`
  during version upgrade


v3.4.0-rc.3 (2018-10-23)
------------------------

* fixed handling of broken Foxx services

  Installation now also fails when the service encounters an error when
  executed. Upgrading or replacing with a broken service will still result
  in the broken services being installed.

* restored error pages for broken Foxx services

  Services that could not be executed will now show an error page (with helpful
  information if development mode is enabled) instead of a generic 404 response.
  Requests to the service that do not prefer HTML (i.e. not a browser window)
  will receive a JSON formatted 503 error response instead.

* added support for `force` flag when upgrading Foxx services

  Using the `force` flag when upgrading or replacing a service falls back to
  installing the service if it does not already exist.

* The order of JSON object attribute keys in JSON return values will now be
  "random" in more cases. In JSON, there is no defined order for object attribute
  keys anyway, so ArangoDB is taking the freedom to return the attribute keys in
  a non-deterministic, seemingly unordered way.

* Fixed an AQL bug where the `optimize-traversals` rule was falsely applied to
  extensions with inline expressions and thereby ignoring them

* fix side-effects of sorting larger arrays (>= 16 members) of constant literal
  values in AQL, when the array was used not only for IN-value filtering but also
  later in the query.
  The array values were sorted so the IN-value lookup could use a binary search
  instead of a linear search, but this did not take into account that the array
  could have been used elsewhere in the query, e.g. as a return value. The fix
  will create a copy of the array and sort the copy, leaving the original array
  untouched.

* disallow empty LDAP password

* fixes validation of allowed or not allowed foxx service mount paths within
  the Web UI

* The single database or single coordinator statistics in a cluster
  environment within the Web UI sometimes got called way too often.
  This caused artifacts in the graphs, which is now fixed.

* An aardvark statistics route could not collect and sum up the statistics of
  all coordinators if one of them was ahead and had more results than the others

* Web UI now checks if server statistics are enabled before it sends its first
  request to the statistics API

* fix internal issue #486: immediate deletion (right after creation) of
  a view with a link to one collection and indexed data reports failure
  but removes the link

* fix internal issue #480: link to a collection is not added to a view
  if it was already added to other view

* fix internal issues #407, #445: limit ArangoSearch memory consumption
  so that it won't cause OOM while indexing large collections

* upgraded arangodb starter version to 0.13.5

* removed undocumented `db.<view>.toArray()` function from ArangoShell

* prevent creation of collections and views with the same in cluster setups

* fixed issue #6770: document update: ignoreRevs parameter ignored

* added AQL query optimizer rules `simplify-conditions` and `fuse-filters`

* improve inter-server communication performance:
  - move all response processing off Communicator's socket management thread
  - create multiple Communicator objects with ClusterComm, route via round robin
  - adjust Scheduler threads to always be active, and have designated priorities.

* fix internal issue #2770: the Query Profiling modal dialog in the Web UI
  was slightly malformed.

* fix internal issue #2035: the Web UI now updates its indices view to check
  whether new indices exist or not.

* fix internal issue #6808: newly created databases within the Web UI did not
  appear when used Internet Explorer 11 as a browser.

* fix internal issue #2957: the Web UI was not able to display more than 1000
  documents, even when it was set to a higher amount.

* fix internal issue #2688: the Web UI's graph viewer created malformed node
  labels if a node was expanded multiple times.

* fix internal issue #2785: web ui's sort dialog sometimes got rendered, even
  if it should not.

* fix internal issue #2764: the waitForSync property of a satellite collection
  could not be changed via the Web UI

* dynamically manage libcurl's number of open connections to increase performance
  by reducing the number of socket close and then reopen cycles

* recover short server id from agency after a restart of a cluster node

  this fixes problems with short server ids being set to 0 after a node restart,
  which then prevented cursor result load-forwarding between multiple coordinators
  to work properly

  this should fix arangojs#573

* increased default timeouts in replication

  this decreases the chances of followers not getting in sync with leaders because
  of replication operations timing out

* include forward-ported diagnostic options for debugging LDAP connections

* fixed internal issue #3065: fix variable replacements by the AQL query
  optimizer in arangosearch view search conditions

  The consequence of the missing replacements was that some queries using view
  search conditions could have failed with error messages such as

  "missing variable #3 (a) for node #7 (EnumerateViewNode) while planning registers"

* fixed internal issue #1983: the Web UI was showing a deletion confirmation
  multiple times.

* Restricted usage of views in AQL, they will throw an error now
  (e.g. "FOR v, e, p IN 1 OUTBOUND @start edgeCollection, view")
  instead of failing the server.

* Allow VIEWs within the AQL "WITH" statement in cluster environment.
  This will now prepare the query for all collections linked within a view.
  (e.g. "WITH view FOR v, e, p IN OUTBOUND 'collectionInView/123' edgeCollection"
  will now be executed properly and not fail with unregistered collection any more)

* Properly check permissions for all collections linked to a view when
  instantiating an AQL query in cluster environment

* support installation of ArangoDB on Windows into directories with multibyte
  character filenames on Windows platforms that used a non-UTF8-codepage

  This was supported on other platforms before, but never worked for ArangoDB's
  Windows version

* display shard synchronization progress for collections outside of the
  `_system` database

* change memory protection settings for memory given back to by the bundled
  JEMalloc memory allocator. This avoids splitting of existing memory mappings
  due to changes of the protection settings

* added missing implementation for `DeleteRangeCF` in RocksDB WAL tailing handler

* fixed agents busy looping gossip

* handle missing `_frontend` collections gracefully

  the `_frontend` system collection is not required for normal ArangoDB operations,
  so if it is missing for whatever reason, ensure that normal operations can go
  on.


v3.4.0-rc.2 (2018-09-30)
------------------------

* upgraded arangosync version to 0.6.0

* upgraded arangodb starter version to 0.13.3

* fixed issue #6611: Properly display JSON properties of user defined foxx services
  configuration within the web UI

* improved shards display in web UI: included arrows to better visualize that
  collection name sections can be expanded and collapsed

* added nesting support for `aql` template strings

* added support for `undefined` and AQL literals to `aql.literal`

* added `aql.join` function

* fixed issue #6583: Agency node segfaults if sent an authenticated HTTP
  request is sent to its port

* fixed issue #6601: Context cancelled (never ending query)

* added more AQL query results cache inspection and control functionality

* fixed undefined behavior in AQL query result cache

* the query editor within the web UI is now catching HTTP 501 responses
  properly

* added AQL VERSION function to return the server version as a string

* added startup parameter `--cluster.advertised-endpoints`

* AQL query optimizer now makes better choices regarding indexes to use in a
  query when there are multiple competing indexes and some of them are prefixes
  of others

  In this case, the optimizer could have preferred indexes that covered less
  attributes, but it should rather pick the indexes that covered more attributes.

  For example, if there was an index on ["a"] and another index on ["a", "b"], then
  previously the optimizer may have picked the index on just ["a"] instead the
  index on ["a", "b"] for queries that used all index attributes but did range
  queries on them (e.g. `FILTER doc.a == @val1 && doc.b >= @val2`).

* Added compression for the AQL intermediate results transfer in the cluster,
  leading to less data being transferred between coordinator and database servers
  in many cases

* forward-ported a bugfix from RocksDB (https://github.com/facebook/rocksdb/pull/4386)
  that fixes range deletions (used internally in ArangoDB when dropping or truncating
  collections)

  The non-working range deletes could have triggered errors such as
  `deletion check in index drop failed - not all documents in the index have been deleted.`
  when dropping or truncating collections

* improve error messages in Windows installer

* allow retrying installation in Windows installer in case an existing database is still
  running and needs to be manually shut down before continuing with the installation

* fix database backup functionality in Windows installer

* fixed memory leak in `/_api/batch` REST handler

* `db._profileQuery()` now also tracks operations triggered when using `LIMIT`
  clauses in a query

* added proper error messages when using views as an argument to AQL functions
  (doing so triggered an `internal error` before)

* fixed return value encoding for collection ids ("cid" attribute") in REST API
  `/_api/replication/logger-follow`

* fixed dumping and restoring of views with arangodump and arangorestore

* fix replication from 3.3 to 3.4

* fixed some TLS errors that occurred when combining HTTPS/TLS transport with the
  VelocyStream protocol (VST)

  That combination could have led to spurious errors such as "TLS padding error"
  or "Tag mismatch" and connections being closed

* make synchronous replication detect more error cases when followers cannot
  apply the changes from the leader

* fixed issue #6379: RocksDB arangorestore time degeneration on dead documents

* fixed issue #6495: Document not found when removing records

* fixed undefined behavior in cluster plan-loading procedure that may have
  unintentionally modified a shared structure

* reduce overhead of function initialization in AQL COLLECT aggregate functions,
  for functions COUNT/LENGTH, SUM and AVG

  this optimization will only be noticable when the COLLECT produces many groups
  and the "hash" COLLECT variant is used

* fixed potential out-of-bounds access in admin log REST handler `/_admin/log`,
  which could have led to the server returning an HTTP 500 error

* catch more exceptions in replication and handle them appropriately

* agency endpoint updates now go through RAFT

* fixed a cleanup issue in Current when a follower was removed from Plan

* catch exceptions in MaintenanceWorker thread

* fixed a bug in cleanOutServer which could lead to a cleaned out server
  still being a follower for some shard

v3.4.0-rc.1 (2018-09-06)
------------------------

* Release Candidate for 3.4.0, please check the `ReleaseNotes/KnownIssues34.md`
  file for a list of known issues.

* upgraded bundled RocksDB version to 5.16.0

* upgraded bundled Snappy compression library to 1.1.7

* fixed issue #5941: if using breadth first search in traversals uniqueness checks
  on path (vertices and edges) have not been applied. In SmartGraphs the checks
  have been executed properly.

* added more detailed progress output to arangorestore, showing the percentage of
  how much data is restored for bigger collections plus a set of overview statistics
  after each processed collection

* added option `--rocksdb.use-file-logging` to enable writing of RocksDB's own
  informational LOG files into RocksDB's database directory.

  This option is turned off by default, but can be enabled for debugging RocksDB
  internals and performance.

* improved error messages when managing Foxx services

  Install/replace/upgrade will now provide additional information when an error
  is encountered during setup. Errors encountered during a `require` call will
  also include information about the underlying cause in the error message.

* fixed some Foxx script names being displayed incorrectly in web UI and Foxx CLI

* major revision of the maintenance feature

* added `uuidv4` and `genRandomBytes` methods to crypto module

* added `hexSlice` methods `hexWrite` to JS Buffer type

* added `Buffer.from`, `Buffer.of`, `Buffer.alloc` and `Buffer.allocUnsafe`
  for improved compatibility with Node.js

* Foxx HTTP API errors now log stacktraces

* fixed issue #5831: custom queries in the ui could not be loaded if the user
  only has read access to the _system database.

* fixed issue #6128: ArangoDb Cluster: Task moved from DBS to Coordinator

* fixed some web ui action events related to Running Queries view and Slow
  Queries History view

* fixed internal issue #2566: corrected web UI alignment of the nodes table

* fixed issue #5736: Foxx HTTP API responds with 500 error when request body
  is too short

* fixed issue #6106: Arithmetic operator type casting documentation incorrect

* The arangosh now supports the velocystream transport protocol via the schemas
  "vst+tcp://", "vst+ssl://", "vst+unix://" schemes.

* The server will no longer lowercase the input in --server.endpoint. This means
  Unix domain socket paths will now  be treated as specified, previously they were lowercased

* fixed logging of requests. A wrong log level was used

* fixed issue #5943: misplaced database ui icon and wrong cursor type were used

* fixed issue #5354: updated the web UI JSON editor, improved usability

* fixed issue #5648: fixed error message when saving unsupported document types

* fixed internal issue #2812: Cluster fails to create many indexes in parallel

* Added C++ implementation, load balancer support, and user restriction to Pregel API.

  If an execution is accessed on a different coordinator than where it was
  created, the request(s) will be forwarded to the correct coordinator. If an
  execution is accessed by a different user than the one who created it, the
  request will be denied.

* the AQL editor in the web UI now supports detailed AQL query profiling

* fixed issue #5884: Subquery nodes are no longer created on DBServers

* intermediate commits in the RocksDB engine are now only enabled in standalone AQL queries

  (not within a JS transaction), standalone truncate as well as for the "import" API

* the AQL editor in the web UI now supports GeoJSON types and is able to render them.

* fixed issue #5035: fixed a vulnerability issue within the web ui's index view

* PR #5552: add "--latency true" option to arangoimport.  Lists microsecond latency

* added `"pbkdf2"` method to `@arangodb/foxx/auth` module

* the `@arangodb/foxx/auth` module now uses a different method to generate salts,
  so salts are no longer guaranteed to be alphanumeric

* fixed internal issue #2567: the Web UI was showing the possibility to move a shard
  from a follower to the current leader

* Renamed RocksDB engine-specific statistics figure `rocksdb.block-cache-used`
  to `rocksdb.block-cache-usage` in output of `db._engineStats()`

  The new figure name is in line with the statistics that the RocksDB library
  provides in its new versions.

* Added RocksDB engine-specific statistics figures `rocksdb.block-cache-capacity`,
  `rocksdb.block-cache-pinned-usage` as well as level-specific figures
  `rocksdb.num-files-at-level` and `rocksdb.compression-ratio-at-level` in
  output of `db._engineStats()`

* Added RocksDB-engine configuration option `--rocksdb.block-align-data-blocks`

  If set to true, data blocks are aligned on lesser of page size and block size,
  which may waste some memory but may reduce the number of cross-page I/Os operations.

* Usage RocksDB format version 3 for new block-based tables

* Bugfix: The AQL syntax variants `UPDATE/REPLACE k WITH d` now correctly take
  _rev from k instead of d (when ignoreRevs is false) and ignore d._rev.

* Added C++ implementation, load balancer support, and user restriction to tasks API

  If a task is accessed on a different coordinator than where it was created,
  the request(s) will be forwarded to the correct coordinator. If a
  task is accessed by a different user than the one who created it, the request
  will be denied.

* Added load balancer support and user-restriction to async jobs API.

  If an async job is accessed on a different coordinator than where it was
  created, the request(s) will be forwarded to the correct coordinator. If a
  job is accessed by a different user than the one who created it, the request
  will be denied.

* switch default storage engine from MMFiles to RocksDB

  In ArangoDB 3.4, the default storage engine for new installations is the RocksDB
  engine. This differs to previous versions (3.2 and 3.3), in which the default
  storage engine was the MMFiles engine.

  The MMFiles engine can still be explicitly selected as the storage engine for
  all new installations. It's only that the "auto" setting for selecting the storage
  engine will now use the RocksDB engine instead of MMFiles engine.

  In the following scenarios, the effectively selected storage engine for new
  installations will be RocksDB:

  * `--server.storage-engine rocksdb`
  * `--server.storage-engine auto`
  * `--server.storage-engine` option not specified

  The MMFiles storage engine will be selected for new installations only when
  explicitly selected:

  * `--server.storage-engine mmfiles`

  On upgrade, any existing ArangoDB installation will keep its previously selected
  storage engine. The change of the default storage engine is thus only relevant
  for new ArangoDB installations and/or existing cluster setups for which new server
  nodes get added later. All server nodes in a cluster setup should use the same
  storage engine to work reliably. Using different storage engines in a cluster is
  unsupported.

* added collection.indexes() as an alias for collection.getIndexes()

* disable V8 engine and JavaScript APIs for agency nodes

* renamed MMFiles engine compactor thread from "Compactor" to "MMFilesCompactor".

  This change will be visible only on systems which allow assigning names to
  threads.

* added configuration option `--rocksdb.sync-interval`

  This option specifies interval (in milliseconds) that ArangoDB will use to
  automatically synchronize data in RocksDB's write-ahead log (WAL) files to
  disk. Automatic syncs will only be performed for not-yet synchronized data,
  and only for operations that have been executed without the *waitForSync*
  attribute.

  Automatic synchronization is performed by a background thread. The default
  sync interval is 100 milliseconds.

  Note: this option is not supported on Windows platforms. Setting the sync
  interval to a value greater 0 will produce a startup warning.

* added AQL functions `TO_BASE64`, `TO_HEX`, `ENCODE_URI_COMPONENT` and `SOUNDEX`

* PR #5857: RocksDB engine would frequently request a new DelayToken.  This caused
  excessive write delay on the next Put() call.  Alternate approach taken.

* changed the thread handling in the scheduler. `--server.maximal-threads` will be
  the maximum number of threads for the scheduler.

* The option `--server.threads` is now obsolete.

* use sparse indexes in more cases now, when it is clear that the index attribute
  value cannot be null

* introduce SingleRemoteOperationNode via "optimize-cluster-single-document-operations"
  optimizer rule, which triggers single document operations directly from the coordinator
  instead of using a full-featured AQL setup. This saves cluster roundtrips.

  Queries directly referencing the document key benefit from this:

      UPDATE {_key: '1'} WITH {foo: 'bar'} IN collection RETURN OLD

* Added load balancer support and user-restriction to cursor API.

  If a cursor is accessed on a different coordinator than where it was created,
  the requests will be forwarded to the correct coordinator. If a cursor is
  accessed by a different user than the one who created it, the request will
  be denied.

* if authentication is turned on requests to databases by users with insufficient rights
 will be answered with the HTTP forbidden (401) response.

* upgraded bundled RocksDB library version to 5.15

* added key generators `uuid` and `padded`

  The `uuid` key generator generates universally unique 128 bit keys, which are
  stored in hexadecimal human-readable format.
  The `padded` key generator generates keys of a fixed length (16 bytes) in
  ascending lexicographical sort order.

* The REST API of `/_admin/status` added: "operationMode" filed with same meaning as
  the "mode" field and field "readOnly" that has the inverted meaning of the field
  "writeOpsEnabled". The old field names will be deprecated in upcoming versions.

* added `COUNT_DISTINCT` AQL function

* make AQL optimizer rule `collect-in-cluster` optimize aggregation functions
  `AVERAGE`, `VARIANCE`, `STDDEV`, `UNIQUE`, `SORTED_UNIQUE` and `COUNT_DISTINCT`
  in a cluster by pushing parts of the aggregation onto the DB servers and only
  doing the total aggregation on the coordinator

* replace JavaScript functions FULLTEXT, NEAR, WITHIN and WITHIN_RECTANGLE with
  regular AQL subqueries via a new optimizer rule "replace-function-with-index".

* the existing "fulltext-index-optimizer" optimizer rule has been removed because its
  duty is now handled by the "replace-function-with-index" rule.

* added option "--latency true" option to arangoimport. Lists microsecond latency
  statistics on 10 second intervals.

* fixed internal issue #2256: ui, document id not showing up when deleting a document

* fixed internal issue #2163: wrong labels within foxx validation of service
  input parameters

* fixed internal issue #2160: fixed misplaced tooltips in indices view

* Added exclusive option for rocksdb collections. Modifying AQL queries can
  now set the exclusive option as well as it can be set on JavaScript transactions.

* added optimizer rule "optimize-subqueries", which makes qualifying subqueries
  return less data

  The rule fires in the following situations:
  * in case only a few results are used from a non-modifying subquery, the rule
    will add a LIMIT statement into the subquery. For example

        LET docs = (
          FOR doc IN collection
            FILTER ...
            RETURN doc
        )
        RETURN docs[0]

    will be turned into

        LET docs = (
          FOR doc IN collection
            FILTER ...
            LIMIT 1
            RETURN doc
        )
        RETURN docs[0]

    Another optimization performed by this rule is to modify the result value
    of subqueries in case only the number of results is checked later. For example

        RETURN LENGTH(
          FOR doc IN collection
            FILTER ...
            RETURN doc
        )

    will be turned into

        RETURN LENGTH(
          FOR doc IN collection
            FILTER ...
            RETURN true
        )

  This saves copying the document data from the subquery to the outer scope and may
  enable follow-up optimizations.

* fixed Foxx queues bug when queues are created in a request handler with an
  ArangoDB authentication header

* abort startup when using SSLv2 for a server endpoint, or when connecting with
  a client tool via an SSLv2 connection.

  SSLv2 has been disabled in the OpenSSL library by default in recent versions
  because of security vulnerabilities inherent in this protocol.

  As it is not safe at all to use this protocol, the support for it has also
  been stopped in ArangoDB. End users that use SSLv2 for connecting to ArangoDB
  should change the protocol from SSLv2 to TLSv12 if possible, by adjusting
  the value of the `--ssl.protocol` startup option.

* added `overwrite` option to document insert operations to allow for easier syncing.

  This implements almost the much inquired UPSERT. In reality it is a REPSERT
  (replace/insert) because only replacement and not modification of documents
  is possible. The option does not work in cluster collections with custom
  sharding.

* added startup option `--log.escape`

  This option toggles the escaping of log output.

  If set to `true` (which is the default value), then the logging will work
  as before, and the following characters in the log output are escaped:

  * the carriage return character (hex 0d)
  * the newline character (hex 0a)
  * the tabstop character (hex 09)
  * any other characters with an ordinal value less than hex 20

  If the option is set to `false`, no characters are escaped. Characters with
  an ordinal value less than hex 20 will not be printed in this mode but will
  be replaced with a space character (hex 20).

  A side effect of turning off the escaping is that it will reduce the CPU
  overhead for the logging. However, this will only be noticable when logging
  is set to a very verbose level (e.g. debug or trace).

* increased the default values for the startup options `--javascript.gc-interval`
  from every 1000 to every 2000 requests, and for `--javascript.gc-frequency` from
  30 to 60 seconds

  This will make the V8 garbage collection run less often by default than in previous
  versions, reducing CPU load a bit and leaving more contexts available on average.

* added `/_admin/repair/distributeShardsLike` that repairs collections with
  distributeShardsLike where the shards aren't actually distributed like in the
  prototype collection, as could happen due to internal issue #1770

* Fixed issue #4271: Change the behavior of the `fullCount` option for AQL query
  cursors so that it will only take into account `LIMIT` statements on the top level
  of the query.

  `LIMIT` statements in subqueries will not have any effect on the `fullCount` results
  any more.

* We added a new geo-spatial index implementation. On the RocksDB storage engine all
  installations will need to be upgraded with `--database.auto-upgrade true`. New geo
  indexes will now only report with the type `geo` instead of `geo1` or `geo2`.
  The index types `geo1` and `geo2` are now deprecated.
  Additionally we removed the deprecated flags `constraint` and `ignoreNull` from geo
  index definitions, these fields were initially deprecated in ArangoDB 2.5

* Add revision id to RocksDB values in primary indexes to speed up replication (~10x).

* PR #5238: Create a default pacing algorithm for arangoimport to avoid TimeoutErrors
  on VMs with limited disk throughput

* Starting a cluster with coordinators and DB servers using different storage engines
  is unsupported. Doing it anyway will now produce a warning on startup

* fixed issue #4919: C++ implementation of LIKE function now matches the old and correct
  behavior of the javascript implementation.

* added `--json` option to arangovpack, allowing to treat its input as plain JSON data
  make arangovpack work without any configuration file

* added experimental arangodb startup option `--javascript.enabled` to enable/disable the
  initialization of the V8 JavaScript engine. Only expected to work on single-servers and
  agency deployments

* pull request #5201: eliminate race scenario where handlePlanChange could run infinite times
  after an execution exceeded 7.4 second time span

* UI: fixed an unreasonable event bug within the modal view engine

* pull request #5114: detect shutdown more quickly on heartbeat thread of coordinator and
  DB servers

* fixed issue #3811: gharial api is now checking existence of `_from` and `_to` vertices
  during edge creation

* There is a new method `_profileQuery` on the database object to execute a query and
  print an explain with annotated runtime information.

* Query cursors can now be created with option `profile`, with a value of 0, 1 or 2.
  This will cause queries to include more statistics in their results and will allow tracing
  of queries.

* fixed internal issue #2147: fixed database filter in UI

* fixed internal issue #2149: number of documents in the UI is not adjusted after moving them

* fixed internal issue #2150: UI - loading a saved query does not update the list of bind
  parameters

* removed option `--cluster.my-local-info` in favor of persisted server UUIDs

  The option `--cluster.my-local-info` was deprecated since ArangoDB 3.3.

* added new collection property `cacheEnabled` which enables in-memory caching for
  documents and primary index entries. Available only when using RocksDB

* arangodump now supports `--threads` option to dump collections in parallel

* arangorestore now supports `--threads` option to restore collections in parallel

* Improvement: The AQL query planner in cluster is now a bit more clever and
  can prepare AQL queries with less network overhead.

  This should speed up simple queries in cluster mode, on complex queries it
  will most likely not show any performance effect.
  It will especially show effects on collections with a very high amount of Shards.

* removed remainders of dysfunctional `/_admin/cluster-test` and `/_admin/clusterCheckPort`
  API endpoints and removed them from documentation

* added new query option `stream` to enable streaming query execution via the
  `POST /_api/cursor` rest interface.

* fixed issue #4698: databases within the UI are now displayed in a sorted order.

* Behavior of permissions for databases and collections changed:
  The new fallback rule for databases for which an access level is not explicitly specified:
  Choose the higher access level of:
    * A wildcard database grant
    * A database grant on the `_system` database
  The new fallback rule for collections for which an access level is not explicitly specified:
  Choose the higher access level of:
    * Any wildcard access grant in the same database, or on "*/*"
    * The access level for the current database
    * The access level for the `_system` database

* fixed issue #4583: add AQL ASSERT and AQL WARN

* renamed startup option `--replication.automatic-failover` to
  `--replication.active-failover`
  using the old option name will still work in ArangoDB 3.4, but the old option
  will be removed afterwards

* index selectivity estimates for RocksDB engine are now eventually consistent

  This change addresses a previous issue where some index updates could be
  "lost" from the view of the internal selectivity estimate, leading to
  inaccurate estimates. The issue is solved now, but there can be up to a second
  or so delay before updates are reflected in the estimates.

* support `returnOld` and `returnNew` attributes for in the following HTTP REST
  APIs:

  * /_api/gharial/<graph>/vertex/<collection>
  * /_api/gharial/<graph>/edge/<collection>

  The exception from this is that the HTTP DELETE verb for these APIs does not
  support `returnOld` because that would make the existing API incompatible

* fixed internal issue #478: remove unused and undocumented REST API endpoints
  _admin/statistics/short and _admin/statistics/long

  These APIs were available in ArangoDB's REST API, but have not been called by
  ArangoDB itself nor have they been part of the documented API. They have been
  superseded by other REST APIs and were partially dysfunctional. Therefore
  these two endpoints have been removed entirely.

* fixed issue #1532: reload users on restore

* fixed internal issue #1475: when restoring a cluster dump to a single server
  ignore indexes of type primary and edge since we mustn't create them here.

* fixed internal issue #1439: improve performance of any-iterator for RocksDB

* issue #1190: added option `--create-database` for arangoimport

* UI: updated dygraph js library to version 2.1.0

* renamed arangoimp to arangoimport for consistency
  Release packages will still install arangoimp as a symlink so user scripts
  invoking arangoimp do not need to be changed

* UI: Shard distribution view now has an accordion view instead of displaying
  all shards of all collections at once.

* fixed issue #4393: broken handling of unix domain sockets in JS_Download

* added AQL function `IS_KEY`
  this function checks if the value passed to it can be used as a document key,
  i.e. as the value of the `_key` attribute

* added AQL functions `SORTED` and `SORTED_UNIQUE`

  `SORTED` will return a sorted version of the input array using AQL's internal
  comparison order
  `SORTED_UNIQUE` will do the same, but additionally removes duplicates.

* added C++ implementation for AQL functions `DATE_NOW`, `DATE_ISO8601`,
  `DATE_TIMESTAMP`, `IS_DATESTRING`, `DATE_DAYOFWEEK`, `DATE_YEAR`,
  `DATE_MONTH`, `DATE_DAY`, `DATE_HOUR`, `DATE_MINUTE`, `DATE_SECOND`,
  `DATE_MILLISECOND`, `DATE_DAYOFYEAR`, `DATE_ISOWEEK`, `DATE_LEAPYEAR`,
  `DATE_QUARTER`, `DATE_DAYS_IN_MONTH`, `DATE_ADD`, `DATE_SUBTRACT`,
  `DATE_DIFF`, `DATE_COMPARE`, `TRANSLATE` and `SHA512`

* fixed a bug where clusterinfo missed changes to plan after agency
  callback is registred for create collection

* Foxx manifest.json files can now contain a $schema key with the value
  of "http://json.schemastore.org/foxx-manifest" to improve tooling support.

* fixed agency restart from compaction without data

* fixed agency's log compaction for internal issue #2249

* only load Plan and Current from agency when actually needed


v3.3.21 (XXXX-XX-XX)
--------------------

* fixed TypeError being thrown instead of validation errors when Foxx manifest
  validation fails

* fixed issue #7586: a running query within the user interface was not shown
  if the active view was `Running Queries` or `Slow Query History`.

* improve Windows installer error messages, fix Windows installer backup routine
  and exit code handling

* make AQL REMOVE operations use less memory with the RocksDB storage engine

  the previous implementation of batch removals read everything to remove into
  memory first before carrying out the first remove operation. The new version
  will only read in about 1000 documents each time and then remove these. Queries
  such as

      FOR doc IN collection FILTER ... REMOVE doc IN collection

  will benefit from this change in terms of memory usage.


v3.3.20 (2018-11-28)
--------------------

* upgraded arangodb starter version to 0.13.9

* Added RocksDB option `--rocksdb.total-write-buffer-size` to limit total memory
  usage across all RocksDB in-memory write buffers

  The total amount of data to build up in all in-memory buffers (backed by log
  files). This option, together with the block cache size configuration option,
  can be used to limit memory usage. If set to 0, the memory usage is not limited.
  This is the default setting in 3.3. The default setting may be adjusted in
  future versions of ArangoDB.

  If set to a value greater than 0, this will cap the memory usage for write buffers,
  but may have an effect on write performance.

* Added RocksDB configuration option `--rocksdb.enforce-block-cache-size-limit`

  Whether or not the maximum size of the RocksDB block cache is strictly enforced.
  This option can be set to limit the memory usage of the block cache to at most the
  specified size. If then inserting a data block into the cache would exceed the
  cache's capacity, the data block will not be inserted. If the flag is not set,
  a data block may still get inserted into the cache. It is evicted later, but the
  cache may temporarily grow beyond its capacity limit.

* Export version and storage engine in cluster health

* Potential fix for issue #7407: arangorestore very slow converting from
  mmfiles to rocksdb

* Updated joi library (Web UI), improved foxx mount path validation

* fix internal issue #2786: improved confirmation dialog when clicking the
  Truncate button in the Web UI

* fix for supervision, which started failing servers using old transient store

* fixed Foxx queues not retrying jobs with infinite `maxFailures`

* Fixed a race condition in a coordinator, it could happen in rare cases and
  only with the maintainer mode enabled if the creation of a collection is in
  progress and at the same time a deletion is forced.

* disable startup warning for Linux kernel variable `vm.overcommit_memory` settings
  values of 0 or 1.
  Effectively `overcommit_memory` settings value of 0 or 1 fix two memory-allocation
  related issues with the default memory allocator used in ArangoDB release builds on
  64bit Linux.
  The issues will remain when running with an `overcommit_memory` settings value of 2,
  so this is now discouraged.
  Setting `overcommit_memory` to 0 or 1 (0 is the Linux kernel's default) fixes issues
  with increasing numbers of memory mappings for the arangod process (which may lead
  to an out-of-memory situation if the kernel's maximum number of mappings threshold
  is hit) and an increasing amount of memory that the kernel counts as "committed".
  With an `overcommit_memory` setting of 0 or 1, an arangod process may either be
  killed by the kernel's OOM killer or will die with a segfault when accessing memory
  it has allocated before but the kernel could not provide later on. This is still
  more acceptable than the kernel not providing any more memory to the process when
  there is still physical memory left, which may have occurred with an `overcommit_memory`
  setting of 2 after the arangod process had done lots of allocations.

  In summary, the recommendation for the `overcommit_memory` setting is now to set it
  to 0 or 1 (0 is kernel default) and not use 2.

* force connection timeout to be 7 seconds to allow libcurl time to retry lost DNS
  queries.

* increase maximum number of collections/shards in an AQL query from 256 to 2048

* don't rely on `_modules` collection being present and usable for arangod startup

* optimizes the web ui's routing which could possibly led to unwanted events.

* fixes some graph data parsing issues in the ui, e.g. cleaning up duplicate
  edges inside the graph viewer.

* in a cluster environment, the arangod process now exits if wrong credentials
  are used during the startup process.

* Fixed an AQL bug where the optimize-traversals rule was falsely applied to
  extensions with inline expressions and thereby ignoring them

* fix side-effects of sorting larger arrays (>= 16 members) of constant literal
  values in AQL, when the array was not used only for IN-value filtering but also
  later in the query.
  The array values were sorted so the IN-value lookup could use a binary search
  instead of a linear search, but this did not take into account that the array
  could have been used elsewhere in the query, e.g. as a return value. The fix
  will create a copy of the array and sort the copy, leaving the original array
  untouched.

* fixed a bug when cluster indexes were usable for queries, while
  still being built on db servers

* fix move leader shard: wait until all but the old leader are in sync.
  This fixes some unstable tests.

* cluster health features more elaborate agent records


v3.3.19 (2018-10-20)
--------------------

* fixes validation of allowed or not allowed foxx service mount paths within
  the Web UI

* The single database or single coordinator statistics in a cluster
  environment within the Web UI sometimes got called way too often.
  This caused artifacts in the graphs, which is now fixed.

* An aardvark statistics route could not collect and sum up the statistics of
  all coordinators if one of them was ahead and had more results than the others

* upgraded arangodb starter version to 0.13.6

* turn on intermediate commits in replication applier in order to decrease
  the size of transactional operations on replication (issue #6821)

* fixed issue #6770: document update: ignoreRevs parameter ignored

* when returning memory to the OS, use the same memory protection flags as
  when initializing the memory

  this prevents "hole punching" and keeps the OS from splitting one memory
  mapping into multiple mappings with different memory protection settings

* fix internal issue #2770: the Query Profiling modal dialog in the Web UI
  was slightly malformed.

* fix internal issue #2035: the Web UI now updates its indices view to check
  whether new indices exist or not.

* fix internal issue #6808: newly created databases within the Web UI did not
  appear when used Internet Explorer 11 as a browser.

* fix internal issue #2688: the Web UI's graph viewer created malformed node
  labels if a node was expanded multiple times.

* fix internal issue #2957: the Web UI was not able to display more than 1000
  documents, even when it was set to a higher amount.

* fix internal issue #2785: web ui's sort dialog sometimes got rendered, even
  if it should not.

* fix internal issue #2764: the waitForSync property of a satellite collection
  could not be changed via the Web UI

* improved logging in case of replication errors

* recover short server id from agency after a restart of a cluster node

  this fixes problems with short server ids being set to 0 after a node restart,
  which then prevented cursor result load-forwarding between multiple coordinators
  to work properly

  this should fix arangojs#573

* increased default timeouts in replication

  this decreases the chances of followers not getting in sync with leaders because
  of replication operations timing out

* fixed internal issue #1983: the Web UI was showing a deletion confirmation
  multiple times.

* fixed agents busy looping gossip

* handle missing `_frontend` collections gracefully

  the `_frontend` system collection is not required for normal ArangoDB operations,
  so if it is missing for whatever reason, ensure that normal operations can go
  on.


v3.3.18
-------

* not released


v3.3.17 (2018-10-04)
--------------------

* upgraded arangosync version to 0.6.0

* added several advanced options for configuring and debugging LDAP connections.
  Please note that some of the following options are platform-specific and may not
  work on all platforms or with all LDAP servers reliably:

  - `--ldap.serialized`: whether or not calls into the underlying LDAP library
    should be serialized.
    This option can be used to work around thread-unsafe LDAP library functionality.
  - `--ldap.serialize-timeout`: sets the timeout value that is used when waiting to
    enter the LDAP library call serialization lock. This is only meaningful when
    `--ldap.serialized` has been set to `true`.
  - `--ldap.retries`: number of tries to attempt a connection. Setting this to values
    greater than one will make ArangoDB retry to contact the LDAP server in case no
    connection can be made initially.
  - `--ldap.restart`: whether or not the LDAP library should implicitly restart
    connections
  - `--ldap.referrals`: whether or not the LDAP library should implicitly chase
    referrals
  - `--ldap.debug`: turn on internal OpenLDAP library output (warning: will print
    to stdout).
  - `--ldap.timeout`: timeout value (in seconds) for synchronous LDAP API calls
    (a value of 0 means default timeout).
  - `--ldap.network-timeout`: timeout value (in seconds) after which network operations
    following the initial connection return in case of no activity (a value of 0 means
    default timeout).
  - `--ldap.async-connect`: whether or not the connection to the LDAP library will
    be done asynchronously.

* fixed a shutdown race in ArangoDB's logger, which could have led to some buffered
  log messages being discarded on shutdown

* display shard synchronization progress for collections outside of the
  `_system` database

* fixed issue #6611: Properly display JSON properties of user defined foxx services
  configuration within the web UI

* fixed issue #6583: Agency node segfaults if sent an authenticated HTTP request is sent to its port

* when cleaning out a leader it could happen that it became follower instead of
  being removed completely

* make synchronous replication detect more error cases when followers cannot
  apply the changes from the leader

* fix some TLS errors that occurred when combining HTTPS/TLS transport with the
  VelocyStream protocol (VST)

  That combination could have led to spurious errors such as "TLS padding error"
  or "Tag mismatch" and connections being closed

* agency endpoint updates now go through RAFT


v3.3.16 (2018-09-19)
--------------------

* fix undefined behavior in AQL query result cache

* the query editor within the web ui is now catching http 501 responses
  properly

* fixed issue #6495 (Document not found when removing records)

* fixed undefined behavior in cluster plan-loading procedure that may have
  unintentionally modified a shared structure

* reduce overhead of function initialization in AQL COLLECT aggregate functions,
  for functions COUNT/LENGTH, SUM and AVG

  this optimization will only be noticable when the COLLECT produces many groups
  and the "hash" COLLECT variant is used

* fixed potential out-of-bounds access in admin log REST handler /_admin/log,
  which could have led to the server returning an HTTP 500 error

* catch more exceptions in replication and handle them appropriately


v3.3.15 (2018-09-10)
--------------------

* fixed an issue in the "sorted" AQL COLLECT variant, that may have led to producing
  an incorrect number of results

* upgraded arangodb starter version to 0.13.3

* fixed issue #5941 if using breadth-first search in traversals uniqueness checks
  on path (vertices and edges) have not been applied. In SmartGraphs the checks
  have been executed properly.

* added more detailed progress output to arangorestore, showing the percentage of
  how much data is restored for bigger collections plus a set of overview statistics
  after each processed collection

* added option `--rocksdb.use-file-logging` to enable writing of RocksDB's own
  informational LOG files into RocksDB's database directory.

  This option is turned off by default, but can be enabled for debugging RocksDB
  internals and performance.

* improved error messages when managing Foxx services

  Install/replace/upgrade will now provide additional information when an error
  is encountered during setup. Errors encountered during a `require` call will
  also include information about the underlying cause in the error message.

* fixed some Foxx script names being displayed incorrectly in web UI and Foxx CLI

* added startup option `--query.optimizer-max-plans value`

  This option allows limiting the number of query execution plans created by the
  AQL optimizer for any incoming queries. The default value is `128`.

  By adjusting this value it can be controlled how many different query execution
  plans the AQL query optimizer will generate at most for any given AQL query.
  Normally the AQL query optimizer will generate a single execution plan per AQL query,
  but there are some cases in which it creates multiple competing plans. More plans
  can lead to better optimized queries, however, plan creation has its costs. The
  more plans are created and shipped through the optimization pipeline, the more time
  will be spent in the optimizer.

  Lowering this option's value will make the optimizer stop creating additional plans
  when it has already created enough plans.

  Note that this setting controls the default maximum number of plans to create. The
  value can still be adjusted on a per-query basis by setting the *maxNumberOfPlans*
  attribute when running a query.

  This change also lowers the default maximum number of query plans from 192 to 128.

* bug fix: facilitate faster shutdown of coordinators and db servers

* cluster nodes should retry registering in agency until successful

* fixed some web ui action events related to Running Queries view and Slow
  Queries History view

* Create a default pacing algorithm for arangoimport to avoid TimeoutErrors
  on VMs with limited disk throughput

* backport PR 6150: establish unique function to indicate when
  application is terminating and therefore network retries should not occur

* backport PR #5201: eliminate race scenario where handlePlanChange
  could run infinite times after an execution exceeded 7.4 second time span


v3.3.14 (2018-08-15)
--------------------

* upgraded arangodb starter version to 0.13.1

* Foxx HTTP API errors now log stacktraces

* fixed issue #5736: Foxx HTTP API responds with 500 error when request body
  is too short

* fixed issue #5831: custom queries in the ui could not be loaded if the user
  only has read access to the _system database.

* fixed internal issue #2566: corrected web UI alignment of the nodes table

* fixed internal issue #2869: when attaching a follower with global applier to an
  authenticated leader already existing users have not been replicated, all users
  created/modified later are replicated.

* fixed internal issue #2865: dumping from an authenticated arangodb the users have
  not been included

* fixed issue #5943: misplaced database ui icon and wrong cursor type were used

* fixed issue #5354: updated the web UI JSON editor, improved usability

* fixed issue #5648: fixed error message when saving unsupported document types

* fixed issue #6076: Segmentation fault after AQL query

  This also fixes issues #6131 and #6174

* fixed issue #5884: Subquery nodes are no longer created on DBServers

* fixed issue #6031: Broken LIMIT in nested list iterations

* fixed internal issue #2812: Cluster fails to create many indexes in parallel

* intermediate commits in the RocksDB engine are now only enabled in standalone AQL
  queries (not within a JS transaction), standalone truncate as well as for the
  "import" API

* Bug fix: race condition could request data from Agency registry that did not
  exist yet.  This caused a throw that would end the Supervision thread.
  All registry query APIs no longer throw exceptions.


v3.3.13 (2018-07-26)
--------------------

* fixed internal issue #2567: the Web UI was showing the possibility to move a
  shard from a follower to the current leader

* fixed issue #5977: Unexpected execution plan when subquery contains COLLECT

* Bugfix: The AQL syntax variants `UPDATE/REPLACE k WITH d` now correctly take
  _rev from k instead of d (when ignoreRevs is false) and ignore d._rev.

* put an upper bound on the number of documents to be scanned when using
  `db.<collection>.any()` in the RocksDB storage engine

  previous versions of ArangoDB did a scan of a random amount of documents in
  the collection, up to the total number of documents available. this produced
  a random selection with a good quality, but needed to scan half the number
  of documents in the collection on average.

  The new version will only scan up to 500 documents, so it produces a less
  random result, but will be a lot faster especially for large collections.

  The implementation of `any()` for the MMFiles engine remains unchanged. The
  MMFiles engine will pick a random document from the entire range of the
  in-memory primary index without performing scans.

* return an empty result set instead of an "out of memory" exception when
  querying the geo index with invalid (out of range) coordinates

* added load balancer support and user-restriction to cursor API.

  If a cursor is accessed on a different coordinator than where it was created,
  the requests will be forwarded to the correct coordinator. If a cursor is
  accessed by a different user than the one who created it, the request will
  be denied.

* keep failed follower in followers list in Plan.

  This increases the changes of a failed follower getting back into sync if the
  follower comes back after a short time. In this case the follower can try to
  get in sync again, which normally takes less time than seeding a completely
  new follower.

* fix assertion failure and undefined behavior in Unix domain socket connections,
  introduced by 3.3.12

* added configuration option `--rocksdb.sync-interval`

  This option specifies interval (in milliseconds) that ArangoDB will use to
  automatically synchronize data in RocksDB's write-ahead log (WAL) files to
  disk. Automatic syncs will only be performed for not-yet synchronized data,
  and only for operations that have been executed without the *waitForSync*
  attribute.

  Automatic synchronization is performed by a background thread. The default
  sync interval is 0, meaning the automatic background syncing is turned off.
  Background syncing in 3.3 is opt-in, whereas in ArangoDB 3.4 the default sync
  interval will be 100 milliseconds.

  Note: this option is not supported on Windows platforms. Setting the sync
  interval to a value greater 0 will produce a startup warning.

* fixed graph creation sometimes failing with 'edge collection
  already used in edge def' when the edge definition contained multiple vertex
  collections, despite the edge definitions being identical

* inception could get caught in a trap, where agent configuration
  version and timeout multiplier lead to incapacitated agency

* fixed issue #5827: Batch request handling incompatible with .NET's default
  ContentType format

* fixed agency's log compaction for internal issue #2249

* inspector collects additionally disk data size and storage engine statistics


v3.3.12 (2018-07-12)
--------------------

* issue #5854: RocksDB engine would frequently request a new DelayToken.  This caused
  excessive write delay on the next Put() call.  Alternate approach taken.

* fixed graph creation under some circumstances failing with 'edge collection
  already used in edge def' despite the edge definitions being identical

* fixed issue #5727: Edge document with user provided key is inserted as many
  times as the number of shards, violating the primary index

* fixed internal issue #2658: AQL modification queries did not allow `_rev`
  checking. There is now a new option `ignoreRevs` which can be set to `false`
  in order to force AQL modification queries to match revision ids before
  doing any modifications

* fixed issue #5679: Replication applier restrictions will crash synchronisation
  after initial sync

* fixed potential issue in RETURN DISTINCT CollectBlock implementation
  that led to the block producing an empty result

* changed communication tasks to use boost strands instead of locks,
  this fixes a race condition with parallel VST communication over
  SSL

* fixed agency restart from compaction without data

* fixed for agent coming back to agency with changed endpoint and
  total data loss

* more patient agency tests to allow for ASAN tests to successfully finish


v3.3.11 (2018-06-26)
--------------------

* upgraded arangosync version to 0.5.3

* upgraded arangodb starter version to 0.12.0

* fixed internal issue #2559: "unexpected document key" error when custom
  shard keys are used and the "allowUserKeys" key generator option is set
  to false

* fixed AQL DOCUMENT lookup function for documents for sharded collections with
  more than a single shard and using a custom shard key (i.e. some shard
  key attribute other than `_key`).
  The previous implementation of DOCUMENT restricted to lookup to a single
  shard in all cases, though this restriction was invalid. That lead to
  `DOCUMENT` not finding documents in cases the wrong shard was contacted. The
  fixed implementation in 3.3.11 will reach out to all shards to find the
  document, meaning it will produce the correct result, but will cause more
  cluster-internal traffic. This increase in traffic may be high if the number
  of shards is also high, because each invocation of `DOCUMENT` will have to
  contact all shards.
  There will be no performance difference for non-sharded collections or
  collections that are sharded by `_key` or that only have a single shard.

* reimplemented replication view in web UI

* fixed internal issue #2256: ui, document id not showing up when deleting a document

* fixed internal issue #2163: wrong labels within foxx validation of service
  input parameters

* fixed internal issue #2160: fixed misplaced tooltips in indices view

* added new arangoinspect client tool, to help users and customers easily collect
  information of any ArangoDB server setup, and facilitate troubleshooting for the
  ArangoDB Support Team


v3.3.10 (2018-06-04)
--------------------

* make optimizer rule "remove-filter-covered-by-index" not stop after removing
  a sub-condition from a FILTER statement, but pass the optimized FILTER
  statement again into the optimizer rule for further optimizations.
  This allows optimizing away some more FILTER conditions than before.

* allow accessing /_admin/status URL on followers too in active failover setup

* fix cluster COLLECT optimization for attributes that were in "sorted" variant of
  COLLECT and that were provided by a sorted index on the collected attribute

* apply fulltext index optimization rule for multiple fulltext searches in
  the same query

  this fixes https://stackoverflow.com/questions/50496274/two-fulltext-searches-on-arangodb-cluster-v8-is-involved

* validate `_from` and `_to` values of edges on updates consistently

* fixed issue #5400: Unexpected AQL Result

* fixed issue #5429: Frequent 'updated local foxx repository' messages

* fixed issue #5252: Empty result if FULLTEXT() is used together with LIMIT offset

* fixed issue #5035: fixed a vulnerability issue within the web ui's index view

* inception was ignoring leader's configuration


v3.3.9 (2018-05-17)
-------------------

* added `/_admin/repair/distributeShardsLike` that repairs collections with
  distributeShardsLike where the shards aren't actually distributed like in the
  prototype collection, as could happen due to internal issue #1770

* fixed Foxx queues bug when queues are created in a request handler with an
  ArangoDB authentication header

* upgraded arangosync version to 0.5.1

* upgraded arangodb starter version to 0.11.3

* fix cluster upgrading issue introduced in 3.3.8

  the issue made arangod crash when starting a DB server with option
  `--database.auto-upgrade true`

* fix C++ implementation of AQL ZIP function to return each distinct attribute
  name only once. The previous implementation added non-unique attribute names
  multiple times, which led to follow-up issues.
  Now if an attribute name occurs multiple times in the input list of attribute
  names, it will only be incorporated once into the result object, with the
  value that corresponds to the first occurrence.
  This fix also changes the V8 implementation of the ZIP function, which now
  will always return the first value for non-unique attribute names and not the
  last occurring value.

* self heal during a Foxx service install, upgrade or replace no longer breaks
  the respective operation

* make /_api/index, /_api/database and /_api/user REST handlers use the scheduler's
  internal queue, so they do not run in an I/O handling thread

* fixed issue #4919: C++ implementation of LIKE function now matches the old and
  correct behavior of the JavaScript implementation.

* added REST API endpoint /_admin/server/availability for monitoring purposes

* UI: fixed an unreasonable event bug within the modal view engine

* fixed issue #3811: gharial api is now checking existence of _from and _to vertices
  during edge creation

* fixed internal issue #2149: number of documents in the UI is not adjusted after
  moving them

* fixed internal issue #2150: UI - loading a saved query does not update the list
  of bind parameters

* fixed internal issue #2147 - fixed database filter in UI

* fixed issue #4934: Wrong used GeoIndex depending on FILTER order

* added `query` and `aql.literal` helpers to `@arangodb` module.

* remove post-sort from GatherNode in cluster AQL queries that do use indexes
  for filtering but that do not require a sorted result

  This optimization can speed up gathering data from multiple shards, because
  it allows to remove a merge sort of the individual shards' results.

* extend the already existing "reduce-extraction-to-projection" AQL optimizer
  rule for RocksDB to provide projections of up to 5 document attributes. The
  previous implementation only supported a projection for a single document
  attribute. The new implementation will extract up to 5 document attributes from
  a document while scanning a collection via an EnumerateCollectionNode.
  Additionally the new version of the optimizer rule can also produce projections
  when scanning an index via an IndexNode.
  The optimization is benefial especially for huge documents because it will copy
  out only the projected attributes from the document instead of copying the entire
  document data from the storage engine.

  When applied, the explainer will show the projected attributes in a `projections`
  remark for an EnumerateCollectionNode or IndexNode. The optimization is limited
  to the RocksDB storage engine.

* added index-only optimization for AQL queries that can satisfy the retrieval of
  all required document attributes directly from an index.

  This optimization will be triggered for the RocksDB engine if an index is used
  that covers all required attributes of the document used later on in the query.
  If applied, it will save retrieving the actual document data (which would require
  an extra lookup in RocksDB), but will instead build the document data solely
  from the index values found. It will only be applied when using up to 5 attributes
  from the document, and only if the rest of the document data is not used later
  on in the query.

  The optimization is currently available for the RocksDB engine for the index types
  primary, edge, hash, skiplist and persistent.

  If the optimization is applied, it will show up as "index only" in an AQL
  query's execution plan for an IndexNode.

* added scan-only optimization for AQL queries that iterate over collections or
  indexes and that do not need to return the actual document values.

  Not fetching the document values from the storage engine will provide a
  considerable speedup when using the RocksDB engine, but may also help a bit
  in case of the MMFiles engine. The optimization will only be applied when
  full-scanning or index-scanning a collection without refering to any of its
  documents later on, and, for an IndexNode, if all filter conditions for the
  documents of the collection are covered by the index.

  If the optimization is applied, it will show up as "scan only" in an AQL
  query's execution plan for an EnumerateCollectionNode or an IndexNode.

* extend existing "collect-in-cluster" optimizer rule to run grouping, counting
  and deduplication on the DB servers in several cases, so that the coordinator
  will only need to sum up the potentially smaller results from the individual shards.

  The following types of COLLECT queries are covered now:
  - RETURN DISTINCT expr
  - COLLECT WITH COUNT INTO ...
  - COLLECT var1 = expr1, ..., varn = exprn (WITH COUNT INTO ...), without INTO or KEEP
  - COLLECT var1 = expr1, ..., varn = exprn AGGREGATE ..., without INTO or KEEP, for
    aggregate functions COUNT/LENGTH, SUM, MIN and MAX.

* honor specified COLLECT method in AQL COLLECT options

  for example, when the user explicitly asks for the COLLECT method
  to be `sorted`, the optimizer will now not produce an alternative
  version of the plan using the hash method.

  additionally, if the user explcitly asks for the COLLECT method to
  be `hash`, the optimizer will now change the existing plan to use
  the hash method if possible instead of just creating an alternative
  plan.

  `COLLECT ... OPTIONS { method: 'sorted' }` => always use sorted method
  `COLLECT ... OPTIONS { method: 'hash' }`   => use hash if this is technically possible
  `COLLECT ...` (no options)                 => create a plan using sorted, and another plan using hash method

* added bulk document lookups for MMFiles engine, which will improve the performance
  of document lookups from an inside an index in case the index lookup produces many
  documents


v3.3.8 (2018-04-24)
-------------------

* included version of ArangoDB Starter (`arangodb` binary) updated to v0.10.11,
  see [Starter changelog](https://github.com/arangodb-helper/arangodb/blob/master/CHANGELOG.md)

* added arangod startup option `--dump-options` to print all configuration parameters
  as a JSON object

* fixed: (Enterprise Edition only) If you restore a SmartGraph where the collections
  are still existing and are supposed to be dropped on restore we ended up in
  duplicate name error. This is now gone and the SmartGraph is correctly restored.

* fix lookups by `_id` in smart graph edge collections

* improve startup resilience in case there are datafile errors (MMFiles)

  also allow repairing broken VERSION files automatically on startup by
  specifying the option `--database.ignore-datafile-errors true`

* fix issue #4582: UI query editor now supports usage of empty string as bind parameter value

* fixed internal issue #2148: Number of documents found by filter is misleading in web UI

* added startup option `--database.required-directory-state`

  using this option it is possible to require the database directory to be
  in a specific state on startup. the options for this value are:

  - non-existing: database directory must not exist
  - existing: database directory must exist
  - empty: database directory must exist but be empty
  - populated: database directory must exist and contain specific files already
  - any: any state allowed

* field "$schema" in Foxx manifest.json files no longer produce warnings

* added `@arangodb/locals` module to expose the Foxx service context as an
  alternative to using `module.context` directly.

* `db._executeTransaction` now accepts collection objects as collections.

* supervision can be put into maintenance mode


v3.3.7 (2018-04-11)
-------------------

* added hidden option `--query.registry-ttl` to control the lifetime of cluster AQL
  query parts

* fixed internal issue #2237: AQL queries on collections with replicationFactor:
  "satellite" crashed arangod in single server mode

* fixed restore of satellite collections: replicationFactor was set to 1 during
  restore

* fixed dump and restore of smart graphs:
  a) The dump will not include the hidden shadow collections anymore, they were dumped
     accidentially and only contain duplicated data.
  b) Restore will now ignore hidden shadow collections as all data is contained
     in the smart-edge collection. You can manually include these collections from an
     old dump (3.3.5 or earlier) by using `--force`.
  c) Restore of a smart-graph will now create smart collections properly instead
     of getting into `TIMEOUT_IN_CLUSTER_OPERATION`

* fixed issue in AQL query optimizer rule "restrict-to-single-shard", which
  may have sent documents to a wrong shard in AQL INSERT queries that specified
  the value for `_key` using an expression (and not a constant value)
  Important: if you were affected by this bug in v3.3.5 it is required that you
  recreate your dataset in v3.3.6 (i.e. dumping and restoring) instead of doing
  a simple binary upgrade

* added /_admin/status HTTP API for debugging purposes

* added ArangoShell helper function for packaging all information about an
  AQL query so it can be run and analyzed elsewhere:

  query = "FOR doc IN mycollection FILTER doc.value > 42 RETURN doc";
  require("@arangodb/aql/explainer").debugDump("/tmp/query-debug-info", query);

  Entitled users can send the generated file to the ArangoDB support to facilitate
  reproduction and debugging.

* added hidden option `--server.ask-jwt-secret`. This is an internal option
  for debugging and should not be exposed to end-users.

* fix for internal issue #2215. supervision will now wait for agent to
  fully prepare before adding 10 second grace period after leadership change

* fixed internal issue #2215's FailedLeader timeout bug

v3.3.5 (2018-03-28)
-------------------

* fixed issue #4934: Wrong used GeoIndex depending on FILTER order

* make build id appear in startup log message alongside with other version info

* make AQL data modification operations that are sent to all shards and that are
  supposed to return values (i.e. `RETURN OLD` or `RETURN NEW`) not return fake
  empty result rows if the document to be updated/replaced/removed was not present
  on the target shard

* added AQL optimizer rule `restrict-to-single-shard`

  This rule will kick in if a collection operation (index lookup or data
  modification operation) will only affect a single shard, and the operation can be
  restricted to the single shard and is not applied for all shards. This optimization
  can be applied for queries that access a collection only once in the query, and that
  do not use traversals, shortest path queries and that do not access collection data
  dynamically using the `DOCUMENT`, `FULLTEXT`, `NEAR` or `WITHIN` AQL functions.
  Additionally, the optimizer will only pull off this optimization if can safely
  determine the values of all the collection's shard keys from the query, and when the
  shard keys are covered by a single index (this is always true if the shard key is
  the default `_key`)

* display missing attributes of GatherNodes in AQL explain output

* make AQL optimizer rule `undistribute-remove-after-enum-coll` fire in a few
  more cases in which it is possible

* slightly improve index selection for the RocksDB engine when there are multiple
  competing indexes with the same attribute prefixes, but different amount of
  attributes covered. In this case, the more specialized index will be preferred
  now

* fix issue #4924: removeFollower now prefers to remove the last follower(s)

* added "collect-in-cluster" optimizer rule to have COLLECT WITH COUNT queries
  without grouping being executed on the DB servers and the coordinator only summing
  up the counts from the individual shards

* fixed issue #4900: Nested FOR query uses index but ignores other filters

* properly exit v8::Context in one place where it was missing before

* added hidden option `--cluster.index-create-timeout` for controlling the
  default value of the index creation timeout in cluster
  under normal circumstances, this option does not need to be adjusted

* increase default timeout for index creation in cluster to 3600s

* fixed issue #4843: Query-Result has more Docs than the Collection itself

* fixed the behavior of ClusterInfo when waiting for current to catch
  up with plan in create collection.

* fixed issue #4827: COLLECT on edge _to field doesn't group distinct values as expected (MMFiles)


v3.3.4 (2018-03-01)
-------------------

* fix AQL `fullCount` result value in some cluster cases when it was off a bit

* fix issue #4651: Simple query taking forever until a request timeout error

* fix issue #4657: fixed incomplete content type header

* Vastly improved the Foxx Store UI

* fix issue #4677: AQL WITH with bind parameters results in "access after data-modification"
  for two independent UPSERTs

* remove unused startup option `--ldap.permissions-attribute-name`

* fix issue #4457: create /var/tmp/arangod with correct user in supervisor mode

* remove long disfunctional admin/long_echo handler

* fixed Foxx API:

  * PUT /_api/foxx/service: Respect force flag
  * PATCH /_api/foxx/service: Check whether a service under given mount exists

* internal issue #1726: supervision failed to remove multiple servers
  from health monitoring at once.

* more information from inception, why agent is activated

* fixed a bug where supervision tried to deal with shards of virtual collections

* fix internal issue #1770: collection creation using distributeShardsLike yields
  errors and did not distribute shards correctly in the following cases:
  1. If numberOfShards * replicationFactor % nrDBServers != 0
     (shards * replication is not divisible by DBServers).
  2. If there was failover / move shard case on the leading collection
     and creating the follower collection afterwards.

* fix timeout issues in replication client expiration

* added missing edge filter to neighbors-only traversals
  in case a filter condition was moved into the traverser and the traversal was
  executed in breadth-first mode and was returning each visited vertex exactly
  once, and there was a filter on the edges of the path and the resulting vertices
  and edges were not used later, the edge filter was not applied

* fixed issue #4160: Run arangod with "--database.auto-upgrade" option always crash silently without error log

* fix internal issue #1848: AQL optimizer was trying to resolve attribute accesses
  to attributes of constant object values at query compile time, but only did so far
  the very first attribute in each object

  this fixes https://stackoverflow.com/questions/48648737/beginner-bug-in-for-loops-from-objects

* fix inconvenience: If we want to start server with a non-existing
  --javascript.app-path it will now be created (if possible)

* fixed: REST API `POST _api/foxx` now returns HTTP code 201 on success, as documented.
         returned 200 before.

* fixed: REST API `PATCH _api/foxx/dependencies` now updates the existing dependencies
         instead of replacing them.

* fixed: Foxx upload of single javascript file. You now can upload via http-url pointing
         to a javascript file.

* fixed issue #4395: If your foxx app includes an `APP` folder it got
         accidently removed by selfhealing this is not the case anymore.

* fixed internal issue #1969 - command apt-get purge/remove arangodb3e was failing


v3.3.3 (2018-01-16)
-------------------

* fix issue #4272: VERSION file keeps disappearing

* fix internal issue #81: quotation marks disappeared when switching table/json
  editor in the query editor ui

* added option `--rocksdb.throttle` to control whether write-throttling is enabled
  Write-throttling is turned on by default, to reduce chances of compactions getting
  too far behind and blocking incoming writes.

* fixed issue #4308: Crash when getter for error.name throws an error (on Windows)

* UI: fixed a query editor caching and parsing issue

* Fixed internal issue #1683: fixes an UI issue where a collection name gets wrongly cached
  within the documents overview of a collection.

* Fixed an issue with the index estimates in RocksDB in the case a transaction is aborted.
  Former the index estimates were modified if the transaction commited or not.
  Now they will only be modified if the transaction commited successfully.

* UI: optimized login view for very small screen sizes

* Truncate in RocksDB will now do intermediate commits every 10.000 documents
  if truncate fails or the server crashes during this operation all deletes
  that have been commited so far are persisted.

* make the default value of `--rocksdb.block-cache-shard-bits` use the RocksDB
  default value. This will mostly mean the default number block cache shard
  bits is lower than before, allowing each shard to store more data and cause
  less evictions from block cache

* issue #4222: Permission error preventing AQL query import / export on webui

* UI: optimized error messages for invalid query bind parameter

* UI: upgraded swagger ui to version 3.9.0

* issue #3504: added option `--force-same-database` for arangorestore

  with this option set to true, it is possible to make any arangorestore attempt
  fail if the specified target database does not match the database name
  specified in the source dump's "dump.json" file. it can thus be used to
  prevent restoring data into the "wrong" database

  The option is set to `false` by default to ensure backwards-compatibility

* make the default value of `--rocksdb.block-cache-shard-bits` use the RocksDB
  default value. This will mostly mean the default number block cache shard
  bits is lower than before, allowing each shard to store more data and cause
  less evictions from block cache

* fixed issue #4255: AQL SORT consuming too much memory

* fixed incorrect persistence of RAFT vote and term


v3.3.2 (2018-01-04)
-------------------

* fixed issue #4199: Internal failure: JavaScript exception in file 'arangosh.js'
  at 98,7: ArangoError 4: Expecting type String

* fixed issue in agency supervision with a good server being left in
  failedServers

* distinguish isReady and allInSync in clusterInventory

* fixed issue #4197: AQL statement not working in 3.3.1 when upgraded from 3.2.10

* do not reuse collection ids when restoring collections from a dump, but assign new collection ids, this should prevent collection id conflicts


v3.3.1 (2017-12-28)
-------------------

* UI: displayed wrong wfs property for a collection when using RocksDB as
  storage engine

* added `--ignore-missing` option to arangoimp
  this option allows importing lines with less fields than specified in the CSV
  header line

* changed misleading error message from "no leader" to "not a leader"

* optimize usage of AQL FULLTEXT index function to a FOR loop with index
  usage in some cases
  When the optimization is applied, this especially speeds up fulltext index
  queries in the cluster

* UI: improved the behavior during collection creation in a cluster environment

* Agency lockup fixes for very small machines.

* Agency performance improvement by finer grained locking.

* Use steady_clock in agency whereever possible.

* Agency prevent Supervision thread crash.

* Fix agency integer overflow in timeout calculation.


v3.3.0 (2017-12-14)
-------------------

* release version

* added a missing try/catch block in the supervision thread


v3.3.rc8 (2017-12-12)
---------------------

* UI: fixed broken Foxx configuration keys. Some valid configuration values
  could not be edited via the ui.

* UI: pressing the return key inside a select2 box no longer triggers the modal's
  success function

* UI: coordinators and db servers are now in sorted order (ascending)


v3.3.rc7 (2017-12-07)
---------------------

* fixed issue #3741: fix terminal color output in Windows

* UI: fixed issue #3822: disabled name input field for system collections

* fixed issue #3640: limit in subquery

* fixed issue #3745: Invalid result when using OLD object with array attribute in UPSERT statement

* UI: edge collections were wrongly added to from and to vertices select box during graph creation

* UI: added not found views for documents and collections

* UI: using default user database api during database creation now

* UI: the graph viewer backend now picks one random start vertex of the
  first 1000 documents instead of calling any(). The implementation of
  "any" is known to scale bad on huge collections with RocksDB.

* UI: fixed disappearing of the navigation label in some case special case

* UI: the graph viewer now displays updated label values correctly.
  Additionally the included node/edge editor now closes automatically
  after a successful node/edge update.

* fixed issue #3917: traversals with high maximal depth take extremely long
  in planning phase.


v3.3.rc4 (2017-11-28)
---------------------

* minor bug-fixes


v3.3.rc3 (2017-11-24)
---------------------

* bug-fixes


v3.3.rc2 (2017-11-22)
---------------------

* UI: document/edge editor now remembering their modes (e.g. code or tree)

* UI: optimized error messages for invalid graph definitions. Also fixed a
  graph renderer cleanup error.

* UI: added a delay within the graph viewer while changing the colors of the
  graph. Necessary due different browser behavior.

* added options `--encryption.keyfile` and `--encryption.key-generator` to arangodump
  and arangorestore

* UI: the graph viewer now displays updated label values correctly.
  Additionally the included node/edge editor now closes automatically
  after a successful node/edge update.

* removed `--recycle-ids` option for arangorestore

  using that option could have led to problems on the restore, with potential
  id conflicts between the originating server (the source dump server) and the
  target server (the restore server)


v3.3.rc1 (2017-11-17)
---------------------

* add readonly mode REST API

* allow compilation of ArangoDB source code with g++ 7

* upgrade minimum required g++ compiler version to g++ 5.4
  That means ArangoDB source code will not compile with g++ 4.x or g++ < 5.4 anymore.

* AQL: during a traversal if a vertex is not found. It will not print an ERROR to the log and continue
  with a NULL value, but will register a warning at the query and continue with a NULL value.
  The situation is not desired as an ERROR as ArangoDB can store edges pointing to non-existing
  vertex which is perfectly valid, but it may be a n issue on the data model, so users
  can directly see it on the query now and do not "by accident" have to check the LOG output.

* introduce `enforceReplicationFactor` attribute for creating collections:
  this optional parameter controls if the coordinator should bail out during collection
  creation if there are not enough DBServers available for the desired `replicationFactor`.

* fixed issue #3516: Show execution time in arangosh

  this change adds more dynamic prompt components for arangosh
  The following components are now available for dynamic prompts,
  settable via the `--console.prompt` option in arangosh:

  - '%t': current time as timestamp
  - '%a': elpased time since ArangoShell start in seconds
  - '%p': duration of last command in seconds
  - '%d': name of current database
  - '%e': current endpoint
  - '%E': current endpoint without protocol
  - '%u': current user

  The time a command takes can be displayed easily by starting arangosh with `--console.prompt "%p> "`.

* make the ArangoShell refill its collection cache when a yet-unknown collection
  is first accessed. This fixes the following problem:

      arangosh1> db._collections();  // shell1 lists all collections
      arangosh2> db._create("test"); // shell2 now creates a new collection 'test'
      arangosh1> db.test.insert({}); // shell1 is not aware of the collection created
                                     // in shell2, so the insert will fail

* make AQL `DISTINCT` not change the order of the results it is applied on

* incremental transfer of initial collection data now can handle partial
  responses for a chunk, allowing the leader/master to send smaller chunks
  (in terms of HTTP response size) and limit memory usage

  this optimization is only active if client applications send the "offset" parameter
  in their requests to PUT `/_api/replication/keys/<id>?type=docs`

* initial creation of shards for cluster collections is now faster with
  `replicationFactor` values bigger than 1. this is achieved by an optimization
  for the case when the collection on the leader is still empty

* potential fix for issue #3517: several "filesystem full" errors in logs
  while there's a lot of disk space

* added C++ implementations for AQL function `SUBSTRING()`, `LEFT()`, `RIGHT()` and `TRIM()`

* show C++ function name of call site in ArangoDB log output

  this requires option `--log.line-number` to be set to *true*

* UI: added word wrapping to query editor

* UI: fixed wrong user attribute name validation, issue #3228

* make AQL return a proper error message in case of a unique key constraint
  violation. previously it only returned the generic "unique constraint violated"
  error message but omitted the details about which index caused the problem.

  This addresses https://stackoverflow.com/questions/46427126/arangodb-3-2-unique-constraint-violation-id-or-key

* added option `--server.local-authentication`

* UI: added user roles

* added config option `--log.color` to toggle colorful logging to terminal

* added config option `--log.thread-name` to additionally log thread names

* usernames must not start with `:role:`, added new options:
    --server.authentication-timeout
    --ldap.roles-attribute-name
    --ldap.roles-transformation
    --ldap.roles-search
    --ldap.superuser-role
    --ldap.roles-include
    --ldap.roles-exclude

* performance improvements for full collection scans and a few other operations
  in MMFiles engine

* added `--rocksdb.encryption-key-generator` for enterprise

* removed `--compat28` parameter from arangodump and replication API

  older ArangoDB versions will no longer be supported by these tools.

* increase the recommended value for `/proc/sys/vm/max_map_count` to a value
  eight times as high as the previous recommended value. Increasing the
  values helps to prevent an ArangoDB server from running out of memory mappings.

  The raised minimum recommended value may lead to ArangoDB showing some startup
  warnings as follows:

      WARNING {memory} maximum number of memory mappings per process is 65530, which seems too low. it is recommended to set it to at least 512000
      WARNING {memory} execute 'sudo sysctl -w "vm.max_map_count=512000"'

* Foxx now warns about malformed configuration/dependency names and aliases in the manifest.


v3.2.17 (XXXX-XX-XX)
--------------------

* added missing virtual destructor for MMFiles transaction data context object

* make synchronous replication detect more error cases when followers cannot
  apply the changes from the leader

* fixed undefined behavior in cluster plan-loading procedure that may have
  unintentionally modified a shared structure

* cluster nodes should retry registering in agency until successful

* fixed issue #5354: updated the ui json editor, improved usability

* fixed issue #5648: fixed error message when saving unsupported document
  types

* fixed issue #5943: misplaced database ui icon and wrong cursor type were used


v3.2.16 (2018-07-12)
--------------------

* upgraded arangodb starter version to 0.12.0

* make edge cache initialization and invalidation more portable by avoiding memset
  on non-POD types

* fixed internal issue #2256: ui, document id not showing up when deleting a document

* fixed issue #5400: Unexpected AQL Result

* Fixed issue #5035: fixed a vulnerability issue within the web ui's index view

* issue one HTTP call less per cluster AQL query

* self heal during a Foxx service install, upgrade or replace no longer breaks
  the respective operation

* inception was ignoring leader's configuration

* inception could get caught in a trap, where agent configuration
  version and timeout multiplier lead to incapacitated agency

* more patient agency tests to allow for ASAN tests to successfully finish

* fixed for agent coming back to agency with changed endpoint and
  total data loss

* fixed agency restart from compaction without data


v3.2.15 (2018-05-13)
--------------------

* upgraded arangodb starter version to 0.11.2

* make /_api/index and /_api/database REST handlers use the scheduler's internal
  queue, so they do not run in an I/O handling thread

* fixed issue #3811: gharial api is now checking existence of _from and _to vertices
  during edge creation


v3.2.14 (2018-04-20)
--------------------

* field "$schema" in Foxx manifest.json files no longer produce warnings

* added `@arangodb/locals` module to expose the Foxx service context as an
  alternative to using `module.context` directly.

* the internal implementation of REST API `/_api/simple/by-example` now uses
  C++ instead of JavaScript

* supervision can be switched to maintenance mode f.e. for rolling upgrades


v3.2.13 (2018-04-13)
--------------------

* improve startup resilience in case there are datafile errors (MMFiles)

  also allow repairing broken VERSION files automatically on startup by
  specifying the option `--database.ignore-datafile-errors true`

* fix issue #4582: UI query editor now supports usage of empty string as bind parameter value

* fix issue #4924: removeFollower now prefers to remove the last follower(s)

* fixed issue #4934: Wrong used GeoIndex depending on FILTER order

* fixed the behavior of clusterinfo when waiting for current to catch
  up with plan in create collection.

* fix for internal issue #2215. supervision will now wait for agent to
  fully prepare before adding 10 second grace period after leadership change

* fixed interal issue #2215 FailedLeader timeout bug


v3.2.12 (2018-02-27)
--------------------

* remove long disfunctional admin/long_echo handler

* fixed Foxx API:

  * PUT /_api/foxx/service: Respect force flag
  * PATCH /_api/foxx/service: Check whether a service under given mount exists

* fix issue #4457: create /var/tmp/arangod with correct user in supervisor mode

* fix internal issue #1848

  AQL optimizer was trying to resolve attribute accesses
  to attributes of constant object values at query compile time, but only did so far
  the very first attribute in each object

  this fixes https://stackoverflow.com/questions/48648737/beginner-bug-in-for-loops-from-objects

* fix inconvenience: If we want to start server with a non-existing
  --javascript.app-path it will now be created (if possible)

* fixed: REST API `POST _api/foxx` now returns HTTP code 201 on success, as documented.
         returned 200 before.

* fixed: REST API `PATCH _api/foxx/dependencies` now updates the existing dependencies
         instead of replacing them.

* fixed: Foxx upload of single javascript file. You now can upload via http-url pointing
         to a javascript file.

* fixed issue #4395: If your foxx app includes an `APP` folder it got accidently removed by selfhealing
         this is not the case anymore.

* fix internal issue 1770: collection creation using distributeShardsLike yields
  errors and did not distribute shards correctly in the following cases:
  1. If numberOfShards * replicationFactor % nrDBServers != 0
     (shards * replication is not divisible by DBServers).
  2. If there was failover / move shard case on the leading collection
     and creating the follower collection afterwards.

* fix timeout issues in replication client expiration

+ fix some inconsistencies in replication for RocksDB engine that could have led
  to some operations not being shipped from master to slave servers

* fix issue #4272: VERSION file keeps disappearing

* fix internal issue #81: quotation marks disappeared when switching table/json
  editor in the query editor ui

* make the default value of `--rocksdb.block-cache-shard-bits` use the RocksDB
  default value. This will mostly mean the default number block cache shard
  bits is lower than before, allowing each shard to store more data and cause
  less evictions from block cache

* fix issue #4393: broken handling of unix domain sockets in
  JS_Download

* fix internal bug #1726: supervision failed to remove multiple
  removed servers from health UI

* fixed internal issue #1969 - command apt-get purge/remove arangodb3e was failing

* fixed a bug where supervision tried to deal with shards of virtual collections


v3.2.11 (2018-01-17)
--------------------

* Fixed an issue with the index estimates in RocksDB in the case a transaction is aborted.
  Former the index estimates were modified if the transaction commited or not.
  Now they will only be modified if the transaction commited successfully.

* Truncate in RocksDB will now do intermediate commits every 10.000 documents
  if truncate fails or the server crashes during this operation all deletes
  that have been commited so far are persisted.

* fixed issue #4308: Crash when getter for error.name throws an error (on Windows)

* UI: fixed a query editor caching and parsing issue for arrays and objects

* Fixed internal issue #1684: Web UI: saving arrays/objects as bind parameters faulty

* Fixed internal issue #1683: fixes an UI issue where a collection name gets wrongly cached
  within the documents overview of a collection.

* issue #4222: Permission error preventing AQL query import / export on webui

* UI: optimized login view for very small screen sizes

* UI: Shard distribution view now has an accordion view instead of displaying
  all shards of all collections at once.

* UI: optimized error messages for invalid query bind parameter

* fixed missing transaction events in RocksDB asynchronous replication

* fixed issue #4255: AQL SORT consuming too much memory

* fixed issue #4199: Internal failure: JavaScript exception in file 'arangosh.js'
  at 98,7: ArangoError 4: Expecting type String

* fixed issue #3818: Foxx configuration keys cannot contain spaces (will not save)

* UI: displayed wrong "waitForSync" property for a collection when
  using RocksDB as storage engine

* prevent binding to the same combination of IP and port on Windows

* fixed incorrect persistence of RAFT vote and term


v3.2.10 (2017-12-22)
--------------------

* replication: more robust initial sync

* fixed a bug in the RocksDB engine that would prevent recalculated
  collection counts to be actually stored

* fixed issue #4095: Inconsistent query execution plan

* fixed issue #4056: Executing empty query causes crash

* fixed issue #4045: Out of memory in `arangorestore` when no access
  rights to dump files

* fixed issue #3031: New Graph: Edge definitions with edges in
  fromCollections and toCollections

* fixed issue #2668: UI: when following wrong link from edge to vertex in
  nonexisting collection misleading error is printed

* UI: improved the behavior during collection creation in a cluster environment

* UI: the graph viewer backend now picks one random start vertex of the
  first 1000 documents instead of calling any(). The implementation of
  any is known to scale bad on huge collections with rocksdb.

* fixed snapshots becoming potentially invalid after intermediate commits in
  the RocksDB engine

* backport agency inquire API changes

* fixed issue #3822: Field validation error in ArangoDB UI - Minor

* UI: fixed disappearing of the navigation label in some cases

* UI: fixed broken foxx configuration keys. Some valid configuration values
  could not be edited via the ui.

* fixed issue #3640: limit in subquery

* UI: edge collections were wrongly added to from and to vertices select
  box during graph creation

* fixed issue #3741: fix terminal color output in Windows

* fixed issue #3917: traversals with high maximal depth take extremely long
  in planning phase.

* fix equality comparison for MMFiles documents in AQL functions UNIQUE
  and UNION_DISTINCT


v3.2.9 (2017-12-04)
-------------------

* under certain conditions, replication could stop. Now fixed by adding an
  equality check for requireFromPresent tick value

* fixed locking for replication context info in RocksDB engine
  this fixes undefined behavior when parallel requests are made to the
  same replication context

* UI: added not found views for documents and collections

* fixed issue #3858: Foxx queues stuck in 'progress' status

* allow compilation of ArangoDB source code with g++ 7

* fixed issue #3224: Issue in the Foxx microservices examples

* fixed a deadlock in user privilege/permission change routine

* fixed a deadlock on server shutdown

* fixed some collection locking issues in MMFiles engine

* properly report commit errors in AQL write queries to the caller for the
  RocksDB engine

* UI: optimized error messages for invalid graph definitions. Also fixed a
  graph renderer cleanrenderer cleanup error.

* UI: document/edge editor now remembering their modes (e.g. code or tree)

* UI: added a delay within the graph viewer while changing the colors of the
  graph. Necessary due different browser behavior.

* fix removal of failed cluster nodes via web interface

* back port of ClusterComm::wait fix in devel
  among other things this fixes too eager dropping of other followers in case
  one of the followers does not respond in time

* transact interface in agency should not be inquired as of now

* inquiry tests and blocking of inquiry on AgencyGeneralTransaction

v3.2.8 (2017-11-18)
-------------------

* fixed a race condition occuring when upgrading via linux package manager

* fixed authentication issue during replication


v3.2.7 (2017-11-13)
-------------------

* Cluster customers, which have upgraded from 3.1 to 3.2 need to upgrade
  to 3.2.7. The cluster supervision is otherwise not operational.

* Fixed issue #3597: AQL with path filters returns unexpected results
  In some cases breadth first search in combination with vertex filters
  yields wrong result, the filter was not applied correctly.

* fixed some undefined behavior in some internal value caches for AQL GatherNodes
  and SortNodes, which could have led to sorted results being effectively not
  correctly sorted.

* make the replication applier for the RocksDB engine start automatically after a
  restart of the server if the applier was configured with its `autoStart` property
  set to `true`. previously the replication appliers were only automatically restarted
  at server start for the MMFiles engine.

* fixed arangodump batch size adaptivity in cluster mode and upped default batch size
  for arangodump

  these changes speed up arangodump in cluster context

* smart graphs now return a proper inventory in response to replication inventory
  requests

* fixed issue #3618: Inconsistent behavior of OR statement with object bind parameters

* only users with read/write rights on the "_system" database can now execute
  "_admin/shutdown" as well as modify properties of the write-ahead log (WAL)

* increase default maximum number of V8 contexts to at least 16 if not explicitly
  configured otherwise.
  the procedure for determining the actual maximum value of V8 contexts is unchanged
  apart from the value `16` and works as follows:
  - if explicitly set, the value of the configuration option `--javascript.v8-contexts`
    is used as the maximum number of V8 contexts
  - when the option is not set, the maximum number of V8 contexts is determined
    by the configuration option `--server.threads` if that option is set. if
    `--server.threads` is not set, then the maximum number of V8 contexts is the
    server's reported hardware concurrency (number of processors visible
    to the arangod process). if that would result in a maximum value of less than 16
    in any of these two cases, then the maximum value will be increased to 16.

* fixed issue #3447: ArangoError 1202: AQL: NotFound: (while executing) when
  updating collection

* potential fix for issue #3581: Unexpected "rocksdb unique constraint
  violated" with unique hash index

* fixed geo index optimizer rule for geo indexes with a single (array of coordinates)
  attribute.

* improved the speed of the shards overview in cluster (API endpoint /_api/cluster/shardDistribution API)
  It is now guaranteed to return after ~2 seconds even if the entire cluster is unresponsive.

* fix agency precondition check for complex objects
  this fixes issues with several CAS operations in the agency

* several fixes for agency restart and shutdown

* the cluster-internal representation of planned collection objects is now more
  lightweight than before, using less memory and not allocating any cache for indexes
  etc.

* fixed issue #3403: How to kill long running AQL queries with the browser console's
  AQL (display issue)

* fixed issue #3549: server reading ENGINE config file fails on common standard
  newline character

* UI: fixed error notifications for collection modifications

* several improvements for the truncate operation on collections:

  * the timeout for the truncate operation was increased in cluster mode in
    order to prevent too frequent "could not truncate collection" errors

  * after a truncate operation, collections in MMFiles still used disk space.
    to reclaim disk space used by truncated collection, the truncate actions
    in the web interface and from the ArangoShell now issue an extra WAL flush
    command (in cluster mode, this command is also propagated to all servers).
    the WAL flush allows all servers to write out any pending operations into the
    datafiles of the truncated collection. afterwards, a final journal rotate
    command is sent, which enables the compaction to entirely remove all datafiles
    and journals for the truncated collection, so that all disk space can be
    reclaimed

  * for MMFiles a special method will be called after a truncate operation so that
    all indexes of the collection can free most of their memory. previously some
    indexes (hash and skiplist indexes) partially kept already allocated memory
    in order to avoid future memory allocations

  * after a truncate operation in the RocksDB engine, an additional compaction
    will be triggered for the truncated collection. this compaction removes all
    deletions from the key space so that follow-up scans over the collection's key
    range do not have to filter out lots of already-removed values

  These changes make truncate operations potentially more time-consuming than before,
  but allow for memory/disk space savings afterwards.

* enable JEMalloc background threads for purging and returning unused memory
  back to the operating system (Linux only)

  JEMalloc will create its background threads on demand. The number of background
  threads is capped by the number of CPUs or active arenas. The background threads run
  periodically and purge unused memory pages, allowing memory to be returned to the
  operating system.

  This change will make the arangod process create several additional threads.
  It is accompanied by an increased `TasksMax` value in the systemd service configuration
  file for the arangodb3 service.

* upgraded bundled V8 engine to bugfix version v5.7.492.77

  this upgrade fixes a memory leak in upstream V8 described in
  https://bugs.chromium.org/p/v8/issues/detail?id=5945 that will result in memory
  chunks only getting uncommitted but not unmapped


v3.2.6 (2017-10-26)
-------------------

* UI: fixed event cleanup in cluster shards view

* UI: reduced cluster dashboard api calls

* fixed a permission problem that prevented collection contents to be displayed
  in the web interface

* removed posix_fadvise call from RocksDB's PosixSequentialFile::Read(). This is
  consistent with Facebook PR 2573 (#3505)

  this fix should improve the performance of the replication with the RocksDB
  storage engine

* allow changing of collection replication factor for existing collections

* UI: replicationFactor of a collection is now changeable in a cluster
  environment

* several fixes for the cluster agency

* fixed undefined behavior in the RocksDB-based geo index

* fixed Foxxmaster failover

* purging or removing the Debian/Ubuntu arangodb3 packages now properly stops
  the arangod instance before actuallying purging or removing


v3.2.5 (2017-10-16)
-------------------

* general-graph module and _api/gharial now accept cluster options
  for collection creation. It is now possible to set replicationFactor and
  numberOfShards for all collections created via this graph object.
  So adding a new collection will not result in a singleShard and
  no replication anymore.

* fixed issue #3408: Hard crash in query for pagination

* minimum number of V8 contexts in console mode must be 2, not 1. this is
  required to ensure the console gets one dedicated V8 context and all other
  operations have at least one extra context. This requirement was not enforced
  anymore.

* fixed issue #3395: AQL: cannot instantiate CollectBlock with undetermined
  aggregation method

* UI: fixed wrong user attribute name validation, issue #3228

* fix potential overflow in CRC marker check when a corrupted CRC marker
  is found at the very beginning of an MMFiles datafile

* UI: fixed unresponsive events in cluster shards view

* Add statistics about the V8 context counts and number of available/active/busy
  threads we expose through the server statistics interface.


v3.2.4 (2017-09-26)
-------------------

* UI: no default index selected during index creation

* UI: added replicationFactor option during SmartGraph creation

* make the MMFiles compactor perform less writes during normal compaction
  operation

  This partially fixes issue #3144

* make the MMFiles compactor configurable

  The following options have been added:

* `--compaction.db-sleep-time`: sleep interval between two compaction runs
    (in s)
  * `--compaction.min-interval"`: minimum sleep time between two compaction
     runs (in s)
  * `--compaction.min-small-data-file-size`: minimal filesize threshold
    original datafiles have to be below for a compaction
  * `--compaction.dead-documents-threshold`: minimum unused count of documents
    in a datafile
  * `--compaction.dead-size-threshold`: how many bytes of the source data file
    are allowed to be unused at most
  * `--compaction.dead-size-percent-threshold`: how many percent of the source
    datafile should be unused at least
  * `--compaction.max-files`: Maximum number of files to merge to one file
  * `--compaction.max-result-file-size`: how large may the compaction result
    file become (in bytes)
  * `--compaction.max-file-size-factor`: how large the resulting file may
    be in comparison to the collection's `--database.maximal-journal-size' setting`

* fix downwards-incompatibility in /_api/explain REST handler

* fix Windows implementation for fs.getTempPath() to also create a
  sub-directory as we do on linux

* fixed a multi-threading issue in cluster-internal communication

* performance improvements for traversals and edge lookups

* removed internal memory zone handling code. the memory zones were a leftover
  from the early ArangoDB days and did not provide any value in the current
  implementation.

* (Enterprise Edition only) added `skipInaccessibleCollections` option for AQL queries:
  if set, AQL queries (especially graph traversals) will treat collections to
  which a user has no access rights to as if these collections were empty.

* adjusted scheduler thread handling to start and stop less threads in
  normal operations

* leader-follower replication catchup code has been rewritten in C++

* early stage AQL optimization now also uses the C++ implementations of
  AQL functions if present. Previously it always referred to the JavaScript
  implementations and ignored the C++ implementations. This change gives
  more flexibility to the AQL optimizer.

* ArangoDB tty log output is now colored for log messages with levels
  FATAL, ERR and WARN.

* changed the return values of AQL functions `REGEX_TEST` and `REGEX_REPLACE`
  to `null` when the input regex is invalid. Previous versions of ArangoDB
  partly returned `false` for invalid regexes and partly `null`.

* added `--log.role` option for arangod

  When set to `true`, this option will make the ArangoDB logger print a single
  character with the server's role into each logged message. The roles are:

  - U: undefined/unclear (used at startup)
  - S: single server
  - C: coordinator
  - P: primary
  - A: agent

  The default value for this option is `false`, so no roles will be logged.


v3.2.3 (2017-09-07)
-------------------

* fixed issue #3106: orphan collections could not be registered in general-graph module

* fixed wrong selection of the database inside the internal cluster js api

* added startup option `--server.check-max-memory-mappings` to make arangod check
  the number of memory mappings currently used by the process and compare it with
  the maximum number of allowed mappings as determined by /proc/sys/vm/max_map_count

  The default value is `true`, so the checks will be performed. When the current
  number of mappings exceeds 90% of the maximum number of mappings, the creation
  of further V8 contexts will be deferred.

  Note that this option is effective on Linux systems only.

* arangoimp now has a `--remove-attribute` option

* added V8 context lifetime control options
  `--javascript.v8-contexts-max-invocations` and `--javascript.v8-contexts-max-age`

  These options allow specifying after how many invocations a used V8 context is
  disposed, or after what time a V8 context is disposed automatically after its
  creation. If either of the two thresholds is reached, an idl V8 context will be
  disposed.

  The default value of `--javascript.v8-contexts-max-invocations` is 0, meaning that
  the maximum number of invocations per context is unlimited. The default value
  for `--javascript.v8-contexts-max-age` is 60 seconds.

* fixed wrong UI cluster health information

* fixed issue #3070: Add index in _jobs collection

* fixed issue #3125: HTTP Foxx API JSON parsing

* fixed issue #3120: Foxx queue: job isn't running when server.authentication = true

* fixed supervision failure detection and handling, which happened with simultaneous
  agency leadership change


v3.2.2 (2017-08-23)
-------------------

* make "Rebalance shards" button work in selected database only, and not make
  it rebalance the shards of all databases

* fixed issue #2847: adjust the response of the DELETE `/_api/users/database/*` calls

* fixed issue #3075: Error when upgrading arangoDB on linux ubuntu 16.04

* fixed a buffer overrun in linenoise console input library for long input strings

* increase size of the linenoise input buffer to 8 KB

* abort compilation if the detected GCC or CLANG isn't in the range of compilers
  we support

* fixed spurious cluster hangups by always sending AQL-query related requests
  to the correct servers, even after failover or when a follower drops

  The problem with the previous shard-based approach was that responsibilities
  for shards may change from one server to another at runtime, after the query
  was already instanciated. The coordinator and other parts of the query then
  sent further requests for the query to the servers now responsible for the
  shards.
  However, an AQL query must send all further requests to the same servers on
  which the query was originally instanciated, even in case of failover.
  Otherwise this would potentially send requests to servers that do not know
  about the query, and would also send query shutdown requests to the wrong
  servers, leading to abandoned queries piling up and using resources until
  they automatically time out.

* fixed issue with RocksDB engine acquiring the collection count values too
  early, leading to the collection count values potentially being slightly off
  even in exclusive transactions (for which the exclusive access should provide
  an always-correct count value)

* fixed some issues in leader-follower catch-up code, specifically for the
  RocksDB engine

* make V8 log fatal errors to syslog before it terminates the process.
  This change is effective on Linux only.

* fixed issue with MMFiles engine creating superfluous collection journals
  on shutdown

* fixed issue #3067: Upgrade from 3.2 to 3.2.1 reset autoincrement keys

* fixed issue #3044: ArangoDB server shutdown unexpectedly

* fixed issue #3039: Incorrect filter interpretation

* fixed issue #3037: Foxx, internal server error when I try to add a new service

* improved MMFiles fulltext index document removal performance
  and fulltext index query performance for bigger result sets

* ui: fixed a display bug within the slow and running queries view

* ui: fixed a bug when success event triggers twice in a modal

* ui: fixed the appearance of the documents filter

* ui: graph vertex collections not restricted to 10 anymore

* fixed issue #2835: UI detection of JWT token in case of server restart or upgrade

* upgrade jemalloc version to 5.0.1

  This fixes problems with the memory allocator returing "out of memory" when
  calling munmap to free memory in order to return it to the OS.

  It seems that calling munmap on Linux can increase the number of mappings, at least
  when a region is partially unmapped. This can lead to the process exceeding its
  maximum number of mappings, and munmap and future calls to mmap returning errors.

  jemalloc version 5.0.1 does not have the `--enable-munmap` configure option anymore,
  so the problem is avoided. To return memory to the OS eventually, jemalloc 5's
  background purge threads are used on Linux.

* fixed issue #2978: log something more obvious when you log a Buffer

* fixed issue #2982: AQL parse error?

* fixed issue #3125: HTTP Foxx API Json parsing

v3.2.1 (2017-08-09)
-------------------

* added C++ implementations for AQL functions `LEFT()`, `RIGHT()` and `TRIM()`

* fixed docs for issue #2968: Collection _key autoincrement value increases on error

* fixed issue #3011: Optimizer rule reduce-extraction-to-projection breaks queries

* Now allowing to restore users in a sharded environment as well
  It is still not possible to restore collections that are sharded
  differently than by _key.

* fixed an issue with restoring of system collections and user rights.
  It was not possible to restore users into an authenticated server.

* fixed issue #2977: Documentation for db._createDatabase is wrong

* ui: added bind parameters to slow query history view

* fixed issue #1751: Slow Query API should provide bind parameters, webui should display them

* ui: fixed a bug when moving multiple documents was not possible

* fixed docs for issue #2968: Collection _key autoincrement value increases on error

* AQL CHAR_LENGTH(null) returns now 0. Since AQL TO_STRING(null) is '' (string of length 0)

* ui: now supports single js file upload for Foxx services in addition to zip files

* fixed a multi-threading issue in the agency when callElection was called
  while the Supervision was calling updateSnapshot

* added startup option `--query.tracking-with-bindvars`

  This option controls whether the list of currently running queries
  and the list of slow queries should contain the bind variables used
  in the queries or not.

  The option can be changed at runtime using the commands

      // enables tracking of bind variables
      // set to false to turn tracking of bind variables off
      var value = true;
      require("@arangodb/aql/queries").properties({
        trackBindVars: value
      });

* index selectivity estimates are now available in the cluster as well

* fixed issue #2943: loadIndexesIntoMemory not returning the same structure
  as the rest of the collection APIs

* fixed issue #2949: ArangoError 1208: illegal name

* fixed issue #2874: Collection properties do not return `isVolatile`
  attribute

* potential fix for issue #2939: Segmentation fault when starting
  coordinator node

* fixed issue #2810: out of memory error when running UPDATE/REPLACE
  on medium-size collection

* fix potential deadlock errors in collector thread

* disallow the usage of volatile collections in the RocksDB engine
  by throwing an error when a collection is created with attribute
  `isVolatile` set to `true`.
  Volatile collections are unsupported by the RocksDB engine, so
  creating them should not succeed and silently create a non-volatile
  collection

* prevent V8 from issuing SIGILL instructions when it runs out of memory

  Now arangod will attempt to log a FATAL error into its logfile in case V8
  runs out of memory. In case V8 runs out of memory, it will still terminate the
  entire process. But at least there should be something in the ArangoDB logs
  indicating what the problem was. Apart from that, the arangod process should
  now be exited with SIGABRT rather than SIGILL as it shouldn't return into the
  V8 code that aborted the process with `__builtin_trap`.

  this potentially fixes issue #2920: DBServer crashing automatically post upgrade to 3.2

* Foxx queues and tasks now ensure that the scripts in them run with the same
  permissions as the Foxx code who started the task / queue

* fixed issue #2928: Offset problems

* fixed issue #2876: wrong skiplist index usage in edge collection

* fixed issue #2868: cname missing from logger-follow results in rocksdb

* fixed issue #2889: Traversal query using incorrect collection id

* fixed issue #2884: AQL traversal uniqueness constraints "propagating" to other traversals? Weird results

* arangoexport: added `--query` option for passing an AQL query to export the result

* fixed issue #2879: No result when querying for the last record of a query

* ui: allows now to edit default access level for collections in database
  _system for all users except the root user.

* The _users collection is no longer accessible outside the arngod process, _queues is always read-only

* added new option "--rocksdb.max-background-jobs"

* removed options "--rocksdb.max-background-compactions", "--rocksdb.base-background-compactions" and "--rocksdb.max-background-flushes"

* option "--rocksdb.compaction-read-ahead-size" now defaults to 2MB

* change Windows build so that RocksDB doesn't enforce AVX optimizations by default
  This fixes startup crashes on servers that do not have AVX CPU extensions

* speed up RocksDB secondary index creation and dropping

* removed RocksDB note in Geo index docs


v3.2.0 (2017-07-20)
-------------------

* fixed UI issues

* fixed multi-threading issues in Pregel

* fixed Foxx resilience

* added command-line option `--javascript.allow-admin-execute`

  This option can be used to control whether user-defined JavaScript code
  is allowed to be executed on server by sending via HTTP to the API endpoint
  `/_admin/execute`  with an authenticated user account.
  The default value is `false`, which disables the execution of user-defined
  code. This is also the recommended setting for production. In test environments,
  it may be convenient to turn the option on in order to send arbitrary setup
  or teardown commands for execution on the server.


v3.2.beta6 (2017-07-18)
-----------------------

* various bugfixes


v3.2.beta5 (2017-07-16)
-----------------------

* numerous bugfixes


v3.2.beta4 (2017-07-04)
-----------------------

* ui: fixed document view _from and _to linking issue for special characters

* added function `db._parse(query)` for parsing an AQL query and returning information about it

* fixed one medium priority and two low priority security user interface
  issues found by owasp zap.

* ui: added index deduplicate options

* ui: fixed renaming of collections for the rocksdb storage engine

* documentation and js fixes for secondaries

* RocksDB storage format was changed, users of the previous beta/alpha versions
  must delete the database directory and re-import their data

* enabled permissions on database and collection level

* added and changed some user related REST APIs
    * added `PUT /_api/user/{user}/database/{database}/{collection}` to change collection permission
    * added `GET /_api/user/{user}/database/{database}/{collection}`
    * added optional `full` parameter to the `GET /_api/user/{user}/database/` REST call

* added user functions in the arangoshell `@arangodb/users` module
    * added `grantCollection` and `revokeCollection` functions
    * added `permission(user, database, collection)` to retrieve collection specific rights

* added "deduplicate" attribute for array indexes, which controls whether inserting
  duplicate index values from the same document into a unique array index will lead to
  an error or not:

      // with deduplicate = true, which is the default value:
      db._create("test");
      db.test.ensureIndex({ type: "hash", fields: ["tags[*]"], deduplicate: true });
      db.test.insert({ tags: ["a", "b"] });
      db.test.insert({ tags: ["c", "d", "c"] }); // will work, because deduplicate = true
      db.test.insert({ tags: ["a"] }); // will fail

      // with deduplicate = false
      db._create("test");
      db.test.ensureIndex({ type: "hash", fields: ["tags[*]"], deduplicate: false });
      db.test.insert({ tags: ["a", "b"] });
      db.test.insert({ tags: ["c", "d", "c"] }); // will not work, because deduplicate = false
      db.test.insert({ tags: ["a"] }); // will fail

  The "deduplicate" attribute is now also accepted by the index creation HTTP
  API endpoint POST /_api/index and is returned by GET /_api/index.

* added optimizer rule "remove-filters-covered-by-traversal"

* Debian/Ubuntu installer: make messages about future package upgrades more clear

* fix a hangup in VST

  The problem happened when the two first chunks of a VST message arrived
  together on a connection that was newly switched to VST.

* fix deletion of outdated WAL files in RocksDB engine

* make use of selectivity estimates in hash, skiplist and persistent indexes
  in RocksDB engine

* changed VM overcommit recommendation for user-friendliness

* fix a shutdown bug in the cluster: a destroyed query could still be active

* do not terminate the entire server process if a temp file cannot be created
  (Windows only)

* fix log output in the front-end, it stopped in case of too many messages


v3.2.beta3 (2017-06-27)
-----------------------

* numerous bugfixes


v3.2.beta2 (2017-06-20)
-----------------------

* potentially fixed issue #2559: Duplicate _key generated on insertion

* fix invalid results (too many) when a skipping LIMIT was used for a
  traversal. `LIMIT x` or `LIMIT 0, x` were not affected, but `LIMIT s, x`
  may have returned too many results

* fix races in SSL communication code

* fix invalid locking in JWT authentication cache, which could have
  crashed the server

* fix invalid first group results for sorted AQL COLLECT when LIMIT
  was used

* fix potential race, which could make arangod hang on startup

* removed `exception` field from transaction error result; users should throw
  explicit `Error` instances to return custom exceptions (addresses issue #2561)

* fixed issue #2613: Reduce log level when Foxx manager tries to self heal missing database

* add a read only mode for users and collection level authorization

* removed `exception` field from transaction error result; users should throw
  explicit `Error` instances to return custom exceptions (addresses issue #2561)

* fixed issue #2677: Foxx disabling development mode creates non-deterministic service bundle

* fixed issue #2684: Legacy service UI not working


v3.2.beta1 (2017-06-12)
-----------------------

* provide more context for index errors (addresses issue #342)

* arangod now validates several OS/environment settings on startup and warns if
  the settings are non-ideal. Most of the checks are executed on Linux systems only.

* fixed issue #2515: The replace-or-with-in optimization rule might prevent use of indexes

* added `REGEX_REPLACE` AQL function

* the RocksDB storage format was changed, users of the previous alpha versions
  must delete the database directory and re-import their data

* added server startup option `--query.fail-on-warning`

  setting this option to `true` will abort any AQL query with an exception if
  it causes a warning at runtime. The value can be overridden per query by
  setting the `failOnWarning` attribute in a query's options.

* added --rocksdb.num-uncompressed-levels to adjust number of non-compressed levels

* added checks for memory managment and warn (i. e. if hugepages are enabled)

* set default SSL cipher suite string to "HIGH:!EXPORT:!aNULL@STRENGTH"

* fixed issue #2469: Authentication = true does not protect foxx-routes

* fixed issue #2459: compile success but can not run with rocksdb

* `--server.maximal-queue-size` is now an absolute maximum. If the queue is
  full, then 503 is returned. Setting it to 0 means "no limit".

* (Enterprise Edition only) added authentication against an LDAP server

* fixed issue #2083: Foxx services aren't distributed to all coordinators

* fixed issue #2384: new coordinators don't pick up existing Foxx services

* fixed issue #2408: Foxx service validation causes unintended side-effects

* extended HTTP API with routes for managing Foxx services

* added distinction between hasUser and authorized within Foxx
  (cluster internal requests are authorized requests but don't have a user)

* arangoimp now has a `--threads` option to enable parallel imports of data

* PR #2514: Foxx services that can't be fixed by self-healing now serve a 503 error

* added `time` function to `@arangodb` module


v3.2.alpha4 (2017-04-25)
------------------------

* fixed issue #2450: Bad optimization plan on simple query

* fixed issue #2448: ArangoDB Web UI takes no action when Delete button is clicked

* fixed issue #2442: Frontend shows already deleted databases during login

* added 'x-content-type-options: nosniff' to avoid MSIE bug

* set default value for `--ssl.protocol` from TLSv1 to TLSv1.2.

* AQL breaking change in cluster:
  The SHORTEST_PATH statement using edge-collection names instead
  of a graph name now requires to explicitly name the vertex-collection names
  within the AQL query in the cluster. It can be done by adding `WITH <name>`
  at the beginning of the query.

  Example:
  ```
  FOR v,e IN OUTBOUND SHORTEST_PATH @start TO @target edges [...]
  ```

  Now has to be:

  ```
  WITH vertices
  FOR v,e IN OUTBOUND SHORTEST_PATH @start TO @target edges [...]
  ```

  This change is due to avoid dead-lock sitations in clustered case.
  An error stating the above is included.

* add implicit use of geo indexes when using SORT/FILTER in AQL, without
  the need to use the special-purpose geo AQL functions `NEAR` or `WITHIN`.

  the special purpose `NEAR` AQL function can now be substituted with the
  following AQL (provided there is a geo index present on the `doc.latitude`
  and `doc.longitude` attributes):

      FOR doc in geoSort
        SORT DISTANCE(doc.latitude, doc.longitude, 0, 0)
        LIMIT 5
        RETURN doc

  `WITHIN` can be substituted with the following AQL:

      FOR doc in geoFilter
        FILTER DISTANCE(doc.latitude, doc.longitude, 0, 0) < 2000
        RETURN doc

  Compared to using the special purpose AQL functions this approach has the
  advantage that it is more composable, and will also honor any `LIMIT` values
  used in the AQL query.

* potential fix for shutdown hangs on OSX

* added KB, MB, GB prefix for integer parameters, % for integer parameters
  with a base value

* added JEMALLOC 4.5.0

* added `--vm.resident-limit` and `--vm.path` for file-backed memory mapping
  after reaching a configurable maximum RAM size

* try recommended limit for file descriptors in case of unlimited
  hard limit

* issue #2413: improve logging in case of lock timeout and deadlocks

* added log topic attribute to /_admin/log api

* removed internal build option `USE_DEV_TIMERS`

  Enabling this option activated some proprietary timers for only selected
  events in arangod. Instead better use `perf` to gather timings.


v3.2.alpha3 (2017-03-22)
------------------------

* increase default collection lock timeout from 30 to 900 seconds

* added function `db._engine()` for retrieval of storage engine information at
  server runtime

  There is also an HTTP REST handler at GET /_api/engine that returns engine
  information.

* require at least cmake 3.2 for building ArangoDB

* make arangod start with less V8 JavaScript contexts

  This speeds up the server start (a little bit) and makes it use less memory.
  Whenever a V8 context is needed by a Foxx action or some other operation and
  there is no usable V8 context, a new one will be created dynamically now.

  Up to `--javascript.v8-contexts` V8 contexts will be created, so this option
  will change its meaning. Previously as many V8 contexts as specified by this
  option were created at server start, and the number of V8 contexts did not
  change at runtime. Now up to this number of V8 contexts will be in use at the
  same time, but the actual number of V8 contexts is dynamic.

  The garbage collector thread will automatically delete unused V8 contexts after
  a while. The number of spare contexts will go down to as few as configured in
  the new option `--javascript.v8-contexts-minimum`. Actually that many V8 contexts
  are also created at server start.

  The first few requests in new V8 contexts will take longer than in contexts
  that have been there already. Performance may therefore suffer a bit for the
  initial requests sent to ArangoDB or when there are only few but performance-
  critical situations in which new V8 contexts will be created. If this is a
  concern, it can easily be fixed by setting `--javascipt.v8-contexts-minimum`
  and `--javascript.v8-contexts` to a relatively high value, which will guarantee
  that many number of V8 contexts to be created at startup and kept around even
  when unused.

  Waiting for an unused V8 context will now also abort if no V8 context can be
  acquired/created after 120 seconds.

* improved diagnostic messages written to logfiles by supervisor process

* fixed issue #2367

* added "bindVars" to attributes of currently running and slow queries

* added "jsonl" as input file type for arangoimp

* upgraded version of bundled zlib library from 1.2.8 to 1.2.11

* added input file type `auto` for arangoimp so it can automatically detect the
  type of the input file from the filename extension

* fixed variables parsing in GraphQL

* added `--translate` option for arangoimp to translate attribute names from
  the input files to attriubte names expected by ArangoDB

  The `--translate` option can be specified multiple times (once per translation
  to be executed). The following example renames the "id" column from the input
  file to "_key", and the "from" column to "_from", and the "to" column to "_to":

      arangoimp --type csv --file data.csv --translate "id=_key" --translate "from=_from" --translate "to=_to"

  `--translate` works for CSV and TSV inputs only.

* changed default value for `--server.max-packet-size` from 128 MB to 256 MB

* fixed issue #2350

* fixed issue #2349

* fixed issue #2346

* fixed issue #2342

* change default string truncation length from 80 characters to 256 characters for
  `print`/`printShell` functions in ArangoShell and arangod. This will emit longer
  prefixes of string values before truncating them with `...`, which is helpful
  for debugging.

* always validate incoming JSON HTTP requests for duplicate attribute names

  Incoming JSON data with duplicate attribute names will now be rejected as
  invalid. Previous versions of ArangoDB only validated the uniqueness of
  attribute names inside incoming JSON for some API endpoints, but not
  consistently for all APIs.

* don't let read-only transactions block the WAL collector

* allow passing own `graphql-sync` module instance to Foxx GraphQL router

* arangoexport can now export to csv format

* arangoimp: fixed issue #2214

* Foxx: automatically add CORS response headers

* added "OPTIONS" to CORS `access-control-allow-methods` header

* Foxx: Fix arangoUser sometimes not being set correctly

* fixed issue #1974


v3.2.alpha2 (2017-02-20)
------------------------

* ui: fixed issue #2065

* ui: fixed a dashboard related memory issue

* Internal javascript rest actions will now hide their stack traces to the client
  unless maintainer mode is activated. Instead they will always log to the logfile

* Removed undocumented internal HTTP API:
  * PUT _api/edges

  The documented GET _api/edges and the undocumented POST _api/edges remains unmodified.

* updated V8 version to 5.7.0.0

* change undocumented behavior in case of invalid revision ids in
  If-Match and If-None-Match headers from 400 (BAD) to 412 (PRECONDITION
  FAILED).

* change undocumented behavior in case of invalid revision ids in
  JavaScript document operations from 1239 ("illegal document revision")
  to 1200 ("conflict").

* added data export tool, arangoexport.

  arangoexport can be used to export collections to json, jsonl or xml
  and export a graph or collections to xgmml.

* fixed a race condition when closing a connection

* raised default hard limit on threads for very small to 64

* fixed negative counting of http connection in UI


v3.2.alpha1 (2017-02-05)
------------------------

* added figure `httpRequests` to AQL query statistics

* removed revisions cache intermediate layer implementation

* obsoleted startup options `--database.revision-cache-chunk-size` and
  `--database.revision-cache-target-size`

* fix potential port number over-/underruns

* added startup option `--log.shorten-filenames` for controlling whether filenames
  in log messages should be shortened to just the filename with the absolute path

* removed IndexThreadFeature, made `--database.index-threads` option obsolete

* changed index filling to make it more parallel, dispatch tasks to boost::asio

* more detailed stacktraces in Foxx apps

* generated Foxx services now use swagger tags


v3.1.24 (XXXX-XX-XX)
--------------------

* fixed one more LIMIT issue in traversals


v3.1.23 (2017-06-19)
--------------------

* potentially fixed issue #2559: Duplicate _key generated on insertion

* fix races in SSL communication code

* fix invalid results (too many) when a skipping LIMIT was used for a
  traversal. `LIMIT x` or `LIMIT 0, x` were not affected, but `LIMIT s, x`
  may have returned too many results

* fix invalid first group results for sorted AQL COLLECT when LIMIT
  was used

* fix invalid locking in JWT authentication cache, which could have
  crashed the server

* fix undefined behavior in traverser when traversals were used inside
  a FOR loop


v3.1.22 (2017-06-07)
--------------------

* fixed issue #2505: Problem with export + report of a bug

* documented changed behavior of WITH

* fixed ui glitch in aardvark

* avoid agency compaction bug

* fixed issue #2283: disabled proxy communication internally


v3.1.21 (2017-05-22)
--------------------

* fixed issue #2488:  AQL operator IN error when data use base64 chars

* more randomness in seeding RNG

v3.1.20 (2016-05-16)
--------------------

* fixed incorrect sorting for distributeShardsLike

* improve reliability of AgencyComm communication with Agency

* fixed shard numbering bug, where ids were erouneously incremented by 1

* remove an unnecessary precondition in createCollectionCoordinator

* funny fail rotation fix

* fix in SimpleHttpClient for correct advancement of readBufferOffset

* forward SIG_HUP in supervisor process to the server process to fix logrotaion
  You need to stop the remaining arangod server process manually for the upgrade to work.


v3.1.19 (2017-04-28)
--------------------

* Fixed a StackOverflow issue in Traversal and ShortestPath. Occured if many (>1000) input
  values in a row do not return any result. Fixes issue: #2445

* fixed issue #2448

* fixed issue #2442

* added 'x-content-type-options: nosniff' to avoid MSIE bug

* fixed issue #2441

* fixed issue #2440

* Fixed a StackOverflow issue in Traversal and ShortestPath. Occured if many (>1000) input
  values in a row do not return any result. Fixes issue: #2445

* fix occasional hanging shutdowns on OS X


v3.1.18 (2017-04-18)
--------------------

* fixed error in continuous synchronization of collections

* fixed spurious hangs on server shutdown

* better error messages during restore collection

* completely overhaul supervision. More detailed tests

* Fixed a dead-lock situation in cluster traversers, it could happen in
  rare cases if the computation on one DBServer could be completed much earlier
  than the other server. It could also be restricted to SmartGraphs only.

* (Enterprise Edition only) Fixed a bug in SmartGraph DepthFirstSearch. In some
  more complicated queries, the maxDepth limit of 1 was not considered strictly
  enough, causing the traverser to do unlimited depth searches.

* fixed issue #2415

* fixed issue #2422

* fixed issue #1974


v3.1.17 (2017-04-04)
--------------------

* (Enterprise Edition only) fixed a bug where replicationFactor was not correctly
  forwarded in SmartGraph creation.

* fixed issue #2404

* fixed issue #2397

* ui - fixed smart graph option not appearing

* fixed issue #2389

* fixed issue #2400


v3.1.16 (2017-03-27)
--------------------

* fixed issue #2392

* try to raise file descriptors to at least 8192, warn otherwise

* ui - aql editor improvements + updated ace editor version (memory leak)

* fixed lost HTTP requests

* ui - fixed some event issues

* avoid name resolution when given connection string is a valid ip address

* helps with issue #1842, bug in COLLECT statement in connection with LIMIT.

* fix locking bug in cluster traversals

* increase lock timeout defaults

* increase various cluster timeouts

* limit default target size for revision cache to 1GB, which is better for
  tight RAM situations (used to be 40% of (totalRAM - 1GB), use
  --database.revision-cache-target-size <VALUEINBYTES> to get back the
  old behavior

* fixed a bug with restarted servers indicating status as "STARTUP"
  rather that "SERVING" in Nodes UI.


v3.1.15 (2017-03-20)
--------------------

* add logrotate configuration as requested in #2355

* fixed issue #2376

* ui - changed document api due a chrome bug

* ui - fixed a submenu bug

* added endpoint /_api/cluster/endpoints in cluster case to get all
  coordinator endpoints

* fix documentation of /_api/endpoint, declaring this API obsolete.

* Foxx response objects now have a `type` method for manipulating the content-type header

* Foxx tests now support `xunit` and `tap` reporters


v3.1.14 (2017-03-13)
--------------------

* ui - added feature request (multiple start nodes within graph viewer) #2317

* added missing locks to authentication cache methods

* ui - added feature request (multiple start nodes within graph viewer) #2317

* ui - fixed wrong merge of statistics information from different coordinators

* ui - fixed issue #2316

* ui - fixed wrong protocol usage within encrypted environment

* fixed compile error on Mac Yosemite

* minor UI fixes


v3.1.13 (2017-03-06)
--------------------

* fixed variables parsing in GraphQL

* fixed issue #2214

* fixed issue #2342

* changed thread handling to queue only user requests on coordinator

* use exponential backoff when waiting for collection locks

* repair short name server lookup in cluster in the case of a removed
  server


v3.1.12 (2017-02-28)
--------------------

* disable shell color escape sequences on Windows

* fixed issue #2326

* fixed issue #2320

* fixed issue #2315

* fixed a race condition when closing a connection

* raised default hard limit on threads for very small to 64

* fixed negative counting of http connection in UI

* fixed a race when renaming collections

* fixed a race when dropping databases


v3.1.11 (2017-02-17)
--------------------

* fixed a race between connection closing and sending out last chunks of data to clients
  when the "Connection: close" HTTP header was set in requests

* ui: optimized smart graph creation usability

* ui: fixed #2308

* fixed a race in async task cancellation via `require("@arangodb/tasks").unregisterTask()`

* fixed spuriously hanging threads in cluster AQL that could sit idle for a few minutes

* fixed potential numeric overflow for big index ids in index deletion API

* fixed sort issue in cluster, occurring when one of the local sort buffers of a
  GatherNode was empty

* reduce number of HTTP requests made for certain kinds of join queries in cluster,
  leading to speedup of some join queries

* supervision deals with demised coordinators correctly again

* implement a timeout in TraverserEngineRegistry

* agent communication reduced in large batches of append entries RPCs

* inception no longer estimates RAFT timings

* compaction in agents has been moved to a separate thread

* replicated logs hold local timestamps

* supervision jobs failed leader and failed follower revisited for
  function in precarious stability situations

* fixed bug in random number generator for 64bit int


v3.1.10 (2017-02-02)
--------------------

* updated versions of bundled node modules:
  - joi: from 8.4.2 to 9.2.0
  - joi-to-json-schema: from 2.2.0 to 2.3.0
  - sinon: from 1.17.4 to 1.17.6
  - lodash: from 4.13.1 to 4.16.6

* added shortcut for AQL ternary operator
  instead of `condition ? true-part : false-part` it is now possible to also use a
  shortcut variant `condition ? : false-part`, e.g.

      FOR doc IN docs RETURN doc.value ?: 'not present'

  instead of

      FOR doc IN docs RETURN doc.value ? doc.value : 'not present'

* fixed wrong sorting order in cluster, if an index was used to sort with many
  shards.

* added --replication-factor, --number-of-shards and --wait-for-sync to arangobench

* turn on UTF-8 string validation for VelocyPack values received via VST connections

* fixed issue #2257

* upgraded Boost version to 1.62.0

* added optional detail flag for db.<collection>.count()
  setting the flag to `true` will make the count operation returned the per-shard
  counts for the collection:

      db._create("test", { numberOfShards: 10 });
      for (i = 0; i < 1000; ++i) {
        db.test.insert({value: i});
      }
      db.test.count(true);

      {
        "s100058" : 99,
        "s100057" : 103,
        "s100056" : 100,
        "s100050" : 94,
        "s100055" : 90,
        "s100054" : 122,
        "s100051" : 109,
        "s100059" : 99,
        "s100053" : 95,
        "s100052" : 89
      }

* added optional memory limit for AQL queries:

      db._query("FOR i IN 1..100000 SORT i RETURN i", {}, { options: { memoryLimit: 100000 } });

  This option limits the default maximum amount of memory (in bytes) that a single
  AQL query can use.
  When a single AQL query reaches the specified limit value, the query will be
  aborted with a *resource limit exceeded* exception. In a cluster, the memory
  accounting is done per shard, so the limit value is effectively a memory limit per
  query per shard.

  The global limit value can be overriden per query by setting the *memoryLimit*
  option value for individual queries when running an AQL query.

* added server startup option `--query.memory-limit`

* added convenience function to create vertex-centric indexes.

  Usage: `db.collection.ensureVertexCentricIndex("label", {type: "hash", direction: "outbound"})`
  That will create an index that can be used on OUTBOUND with filtering on the
  edge attribute `label`.

* change default log output for tools to stdout (instead of stderr)

* added option -D to define a configuration file environment key=value

* changed encoding behavior for URLs encoded in the C++ code of ArangoDB:
  previously the special characters `-`, `_`, `~` and `.` were returned as-is
  after URL-encoding, now `.` will be encoded to be `%2e`.
  This also changes the behavior of how incoming URIs are processed: previously
  occurrences of `..` in incoming request URIs were collapsed (e.g. `a/../b/` was
  collapsed to a plain `b/`). Now `..` in incoming request URIs are not collapsed.

* Foxx request URL suffix is no longer unescaped

* @arangodb/request option json now defaults to `true` if the response body is not empty and encoding is not explicitly set to `null` (binary).
  The option can still be set to `false` to avoid unnecessary attempts at parsing the response as JSON.

* Foxx configuration values for unknown options will be discarded when saving the configuration in production mode using the web interface

* module.context.dependencies is now immutable

* process.stdout.isTTY now returns `true` in arangosh and when running arangod with the `--console` flag

* add support for Swagger tags in Foxx


v3.1.9 (XXXX-XX-XX)
-------------------

* macos CLI package: store databases and apps in the users home directory

* ui: fixed re-login issue within a non system db, when tab was closed

* fixed a race in the VelocyStream Commtask implementation

* fixed issue #2256


v3.1.8 (2017-01-09)
-------------------

* add Windows silent installer

* add handling of debug symbols during Linux & windows release builds.

* fixed issue #2181

* fixed issue #2248: reduce V8 max old space size from 3 GB to 1 GB on 32 bit systems

* upgraded Boost version to 1.62.0

* fixed issue #2238

* fixed issue #2234

* agents announce new endpoints in inception phase to leader

* agency leadership accepts updatet endpoints to given uuid

* unified endpoints replace localhost with 127.0.0.1

* fix several problems within an authenticated cluster


v3.1.7 (2016-12-29)
-------------------

* fixed one too many elections in RAFT

* new agency comm backported from devel


v3.1.6 (2016-12-20)
-------------------

* fixed issue #2227

* fixed issue #2220

* agency constituent/agent bug fixes in race conditions picking up
  leadership

* supervision does not need waking up anymore as it is running
  regardless

* agents challenge their leadership more rigorously


v3.1.5 (2016-12-16)
-------------------

* lowered default value of `--database.revision-cache-target-size` from 75% of
  RAM to less than 40% of RAM

* fixed issue #2218

* fixed issue #2217

* Foxx router.get/post/etc handler argument can no longer accidentally omitted

* fixed issue #2223


v3.1.4 (2016-12-08)
-------------------

* fixed issue #2211

* fixed issue #2204

* at cluster start, coordinators wait until at least one DBserver is there,
  and either at least two DBservers are there or 15s have passed, before they
  initiate the bootstrap of system collections.

* more robust agency startup from devel

* supervision's AddFollower adds many followers at once

* supervision has new FailedFollower job

* agency's Node has new method getArray

* agency RAFT timing estimates more conservative in waitForSync
  scenario

* agency RAFT timing estimates capped at maximum 2.0/10.0 for low/high


v3.1.3 (2016-12-02)
-------------------

* fix a traversal bug when using skiplist indexes:
  if we have a skiplist of ["a", "unused", "_from"] and a traversal like:
  FOR v,e,p IN OUTBOUND @start @@edges
    FILTER p.edges[0].a == 'foo'
    RETURN v
  And the above index applied on "a" is considered better than EdgeIndex, than
  the executor got into undefined behavior.

* fix endless loop when trying to create a collection with replicationFactor: -1


v3.1.2 (2016-11-24)
-------------------

* added support for descriptions field in Foxx dependencies

* (Enterprise Edition only) fixed a bug in the statistic report for SmartGraph traversals.
Now they state correctly how many documents were fetched from the index and how many
have been filtered.

* Prevent uniform shard distribution when replicationFactor == numServers

v3.1.1 (2016-11-15)
-------------------

* fixed issue #2176

* fixed issue #2168

* display index usage of traversals in AQL explainer output (previously missing)

* fixed issue #2163

* preserve last-used HLC value across server starts

* allow more control over handling of pre-3.1 _rev values

  this changes the server startup option `--database.check-30-revisions` from a boolean (true/false)
  parameter to a string parameter with the following possible values:

  - "fail":
    will validate _rev values of 3.0 collections on collection loading and throw an exception when invalid _rev values are found.
    in this case collections with invalid _rev values are marked as corrupted and cannot be used in the ArangoDB 3.1 instance.
    the fix procedure for such collections is to export the collections from 3.0 database with arangodump and restore them in 3.1 with arangorestore.
    collections that do not contain invalid _rev values are marked as ok and will not be re-checked on following loads.
    collections that contain invalid _rev values will be re-checked on following loads.

  - "true":
    will validate _rev values of 3.0 collections on collection loading and print a warning when invalid _rev values are found.
    in this case collections with invalid _rev values can be used in the ArangoDB 3.1 instance.
    however, subsequent operations on documents with invalid _rev values may silently fail or fail with explicit errors.
    the fix procedure for such collections is to export the collections from 3.0 database with arangodump and restore them in 3.1 with arangorestore.
    collections that do not contain invalid _rev values are marked as ok and will not be re-checked on following loads.
    collections that contain invalid _rev values will be re-checked on following loads.

  - "false":
    will not validate _rev values on collection loading and not print warnings.
    no hint is given when invalid _rev values are found.
    subsequent operations on documents with invalid _rev values may silently fail or fail with explicit errors.
    this setting does not affect whether collections are re-checked later.
    collections will be re-checked on following loads if `--database.check-30-revisions` is later set to either `true` or `fail`.

  The change also suppresses warnings that were printed when collections were restored using arangorestore, and the restore
  data contained invalid _rev values. Now these warnings are suppressed, and new HLC _rev values are generated for these documents
  as before.

* added missing functions to AQL syntax highlighter in web interface

* fixed display of `ANY` direction in traversal explainer output (direction `ANY` was shown as either
  `INBOUND` or `OUTBOUND`)

* changed behavior of toJSON() function when serializing an object before saving it in the database

  if an object provides a toJSON() function, this function is still called for serializing it.
  the change is that the result of toJSON() is not stringified anymore, but saved as is. previous
  versions of ArangoDB called toJSON() and after that additionally stringified its result.

  This change will affect the saving of JS Buffer objects, which will now be saved as arrays of
  bytes instead of a comma-separated string of the Buffer's byte contents.

* allow creating unique indexes on more attributes than present in shardKeys

  The following combinations of shardKeys and indexKeys are allowed/not allowed:

  shardKeys     indexKeys
      a             a        ok
      a             b    not ok
      a           a b        ok
    a b             a    not ok
    a b             b    not ok
    a b           a b        ok
    a b         a b c        ok
  a b c           a b    not ok
  a b c         a b c        ok

* fixed wrong version in web interface login screen (EE only)

* make web interface not display an exclamation mark next to ArangoDB version number 3.1

* fixed search for arbitrary document attributes in web interface in case multiple
  search values were used on different attribute names. in this case, the search always
  produced an empty result

* disallow updating `_from` and `_to` values of edges in SmartGraphs. Updating these
  attributes would lead to potential redistribution of edges to other shards, which must be
  avoided.

* fixed issue #2148

* updated graphql-sync dependency to 0.6.2

* fixed issue #2156

* fixed CRC4 assembly linkage


v3.1.0 (2016-10-29)
-------------------

* AQL breaking change in cluster:

  from ArangoDB 3.1 onwards `WITH` is required for traversals in a
  clustered environment in order to avoid deadlocks.

  Note that for queries that access only a single collection or that have all
  collection names specified somewhere else in the query string, there is no
  need to use *WITH*. *WITH* is only useful when the AQL query parser cannot
  automatically figure out which collections are going to be used by the query.
  *WITH* is only useful for queries that dynamically access collections, e.g.
  via traversals, shortest path operations or the *DOCUMENT()* function.

  more info can be found [here](https://github.com/arangodb/arangodb/blob/devel/Documentation/Books/AQL/Operations/With.md)

* added AQL function `DISTANCE` to calculate the distance between two arbitrary
  coordinates (haversine formula)

* fixed issue #2110

* added Auto-aptation of RAFT timings as calculations only


v3.1.rc2 (2016-10-10)
---------------------

* second release candidate


v3.1.rc1 (2016-09-30)
---------------------

* first release candidate


v3.1.alpha2 (2016-09-01)
------------------------

* added module.context.createDocumentationRouter to replace module.context.apiDocumentation

* bug in RAFT implementation of reads. dethroned leader still answered requests in isolation

* ui: added new graph viewer

* ui: aql-editor added tabular & graph display

* ui: aql-editor improved usability

* ui: aql-editor: query profiling support

* fixed issue #2109

* fixed issue #2111

* fixed issue #2075

* added AQL function `DISTANCE` to calculate the distance between two arbitrary
  coordinates (haversine formula)

* rewrote scheduler and dispatcher based on boost::asio

  parameters changed:
    `--scheduler.threads` and `--server.threads` are now merged into a single one: `--server.threads`

    hidden `--server.extra-threads` has been removed

    hidden `--server.aql-threads` has been removed

    hidden `--server.backend` has been removed

    hidden `--server.show-backends` has been removed

    hidden `--server.thread-affinity` has been removed

* fixed issue #2086

* fixed issue #2079

* fixed issue #2071

  make the AQL query optimizer inject filter condition expressions referred to
  by variables during filter condition aggregation.
  For example, in the following query

      FOR doc IN collection
        LET cond1 = (doc.value == 1)
        LET cond2 = (doc.value == 2)
        FILTER cond1 || cond2
        RETURN { doc, cond1, cond2 }

  the optimizer will now inject the conditions for `cond1` and `cond2` into the filter
  condition `cond1 || cond2`, expanding it to `(doc.value == 1) || (doc.value == 2)`
  and making these conditions available for index searching.

  Note that the optimizer previously already injected some conditions into other
  conditions, but only if the variable that defined the condition was not used
  elsewhere. For example, the filter condition in the query

      FOR doc IN collection
        LET cond = (doc.value == 1)
        FILTER cond
        RETURN { doc }

  already got optimized before because `cond` was only used once in the query and
  the optimizer decided to inject it into the place where it was used.

  This only worked for variables that were referred to once in the query.
  When a variable was used multiple times, the condition was not injected as
  in the following query:

      FOR doc IN collection
        LET cond = (doc.value == 1)
        FILTER cond
        RETURN { doc, cond }

  The fix for #2070 now will enable this optimization so that the query can
  use an index on `doc.value` if available.

* changed behavior of AQL array comparison operators for empty arrays:
  * `ALL` and `ANY` now always return `false` when the left-hand operand is an
    empty array. The behavior for non-empty arrays does not change:
    * `[] ALL == 1` will return `false`
    * `[1] ALL == 1` will return `true`
    * `[1, 2] ALL == 1` will return `false`
    * `[2, 2] ALL == 1` will return `false`
    * `[] ANY == 1` will return `false`
    * `[1] ANY == 1` will return `true`
    * `[1, 2] ANY == 1` will return `true`
    * `[2, 2] ANY == 1` will return `false`
  * `NONE` now always returns `true` when the left-hand operand is an empty array.
    The behavior for non-empty arrays does not change:
    * `[] NONE == 1` will return `true`
    * `[1] NONE == 1` will return `false`
    * `[1, 2] NONE == 1` will return `false`
    * `[2, 2] NONE == 1` will return `true`

* added experimental AQL functions `JSON_STRINGIFY` and `JSON_PARSE`

* added experimental support for incoming gzip-compressed requests

* added HTTP REST APIs for online log level adjustments:

  - GET `/_admin/log/level` returns the current log level settings
  - PUT `/_admin/log/level` modifies the current log level settings

* PATCH /_api/gharial/{graph-name}/vertex/{collection-name}/{vertex-key}
  - changed default value for keepNull to true

* PATCH /_api/gharial/{graph-name}/edge/{collection-name}/{edge-key}
  - changed default value for keepNull to true

* renamed `maximalSize` attribute in parameter.json files to `journalSize`

  The `maximalSize` attribute will still be picked up from collections that
  have not been adjusted. Responses from the replication API will now also use
  `journalSize` instead of `maximalSize`.

* added `--cluster.system-replication-factor` in order to adjust the
  replication factor for new system collections

* fixed issue #2012

* added a memory expection in case V8 memory gets too low

* added Optimizer Rule for other indexes in Traversals
  this allows AQL traversals to use other indexes than the edge index.
  So traversals with filters on edges can now make use of more specific
  indexes, e.g.

      FOR v, e, p IN 2 OUTBOUND @start @@edge FILTER p.edges[0].foo == "bar"

  will prefer a Hash Index on [_from, foo] above the EdgeIndex.

* fixed epoch computation in hybrid logical clock

* fixed thread affinity

* replaced require("internal").db by require("@arangodb").db

* added option `--skip-lines` for arangoimp
  this allows skipping the first few lines from the import file in case the
  CSV or TSV import are used

* fixed periodic jobs: there should be only one instance running - even if it
  runs longer than the period

* improved performance of primary index and edge index lookups

* optimizations for AQL `[*]` operator in case no filter, no projection and
  no offset/limit are used

* added AQL function `OUTERSECTION` to return the symmetric difference of its
  input arguments

* Foxx manifests of installed services are now saved to disk with indentation

* Foxx tests and scripts in development mode should now always respect updated
  files instead of loading stale modules

* When disabling Foxx development mode the setup script is now re-run

* Foxx now provides an easy way to directly serve GraphQL requests using the
  `@arangodb/foxx/graphql` module and the bundled `graphql-sync` dependency

* Foxx OAuth2 module now correctly passes the `access_token` to the OAuth2 server

* added iconv-lite and timezone modules

* web interface now allows installing GitHub and zip services in legacy mode

* added module.context.createDocumentationRouter to replace module.context.apiDocumentation

* bug in RAFT implementation of reads. dethroned leader still answered
  requests in isolation

* all lambdas in ClusterInfo might have been left with dangling references.

* Agency bug fix for handling of empty json objects as values.

* Foxx tests no longer support the Mocha QUnit interface as this resulted in weird
  inconsistencies in the BDD and TDD interfaces. This fixes the TDD interface
  as well as out-of-sequence problems when using the BDD before/after functions.

* updated bundled JavaScript modules to latest versions; joi has been updated from 8.4 to 9.2
  (see [joi 9.0.0 release notes](https://github.com/hapijs/joi/issues/920) for information on
  breaking changes and new features)

* fixed issue #2139

* updated graphql-sync dependency to 0.6.2

* fixed issue #2156


v3.0.13 (XXXX-XX-XX)
--------------------

* fixed issue #2315

* fixed issue #2210


v3.0.12 (2016-11-23)
--------------------

* fixed issue #2176

* fixed issue #2168

* fixed issues #2149, #2159

* fixed error reporting for issue #2158

* fixed assembly linkage bug in CRC4 module

* added support for descriptions field in Foxx dependencies


v3.0.11 (2016-11-08)
--------------------

* fixed issue #2140: supervisor dies instead of respawning child

* fixed issue #2131: use shard key value entered by user in web interface

* fixed issue #2129: cannot kill a long-run query

* fixed issue #2110

* fixed issue #2081

* fixed issue #2038

* changes to Foxx service configuration or dependencies should now be
  stored correctly when options are cleared or omitted

* Foxx tests no longer support the Mocha QUnit interface as this resulted in weird
  inconsistencies in the BDD and TDD interfaces. This fixes the TDD interface
  as well as out-of-sequence problems when using the BDD before/after functions.

* fixed issue #2148


v3.0.10 (2016-09-26)
--------------------

* fixed issue #2072

* fixed issue #2070

* fixed slow cluster starup issues. supervision will demonstrate more
  patience with db servers


v3.0.9 (2016-09-21)
-------------------

* fixed issue #2064

* fixed issue #2060

* speed up `collection.any()` and skiplist index creation

* fixed multiple issues where ClusterInfo bug hung agency in limbo
  timeouting on multiple collection and database callbacks


v3.0.8 (2016-09-14)
-------------------

* fixed issue #2052

* fixed issue #2005

* fixed issue #2039

* fixed multiple issues where ClusterInfo bug hung agency in limbo
  timeouting on multiple collection and database callbacks


v3.0.7 (2016-09-05)
-------------------

* new supervision job handles db server failure during collection creation.


v3.0.6 (2016-09-02)
-------------------

* fixed issue #2026

* slightly better error diagnostics for AQL query compilation and replication

* fixed issue #2018

* fixed issue #2015

* fixed issue #2012

* fixed wrong default value for arangoimp's `--on-duplicate` value

* fix execution of AQL traversal expressions when there are multiple
  conditions that refer to variables set outside the traversal

* properly return HTTP 503 in JS actions when backend is gone

* supervision creates new key in agency for failed servers

* new shards will not be allocated on failed or cleaned servers


v3.0.5 (2016-08-18)
-------------------

* execute AQL ternary operator via C++ if possible

* fixed issue #1977

* fixed extraction of _id attribute in AQL traversal conditions

* fix SSL agency endpoint

* Minimum RAFT timeout was one order of magnitude to short.

* Optimized RAFT RPCs from leader to followers for efficiency.

* Optimized RAFT RPC handling on followers with respect to compaction.

* Fixed bug in handling of duplicates and overlapping logs

* Fixed bug in supervision take over after leadership change.

v3.0.4 (2016-08-01)
-------------------

* added missing lock for periodic jobs access

* fix multiple Foxx related cluster issues

* fix handling of empty AQL query strings

* fixed issue in `INTERSECTION` AQL function with duplicate elements
  in the source arrays

* fixed issue #1970

* fixed issue #1968

* fixed issue #1967

* fixed issue #1962

* fixed issue #1959

* replaced require("internal").db by require("@arangodb").db

* fixed issue #1954

* fixed issue #1953

* fixed issue #1950

* fixed issue #1949

* fixed issue #1943

* fixed segfault in V8, by backporting https://bugs.chromium.org/p/v8/issues/detail?id=5033

* Foxx OAuth2 module now correctly passes the `access_token` to the OAuth2 server

* fixed credentialed CORS requests properly respecting --http.trusted-origin

* fixed a crash in V8Periodic task (forgotten lock)

* fixed two bugs in synchronous replication (syncCollectionFinalize)


v3.0.3 (2016-07-17)
-------------------

* fixed issue #1942

* fixed issue #1941

* fixed array index batch insertion issues for hash indexes that caused problems when
  no elements remained for insertion

* fixed AQL MERGE() function with External objects originating from traversals

* fixed some logfile recovery errors with error message "document not found"

* fixed issue #1937

* fixed issue #1936

* improved performance of arangorestore in clusters with synchronous
  replication

* Foxx tests and scripts in development mode should now always respect updated
  files instead of loading stale modules

* When disabling Foxx development mode the setup script is now re-run

* Foxx manifests of installed services are now saved to disk with indentation


v3.0.2 (2016-07-09)
-------------------

* fixed assertion failure in case multiple remove operations were used in the same query

* fixed upsert behavior in case upsert was used in a loop with the same document example

* fixed issue #1930

* don't expose local file paths in Foxx error messages.

* fixed issue #1929

* make arangodump dump the attribute `isSystem` when dumping the structure
  of a collection, additionally make arangorestore not fail when the attribute
  is missing

* fixed "Could not extract custom attribute" issue when using COLLECT with
  MIN/MAX functions in some contexts

* honor presence of persistent index for sorting

* make AQL query optimizer not skip "use-indexes-rule", even if enough
  plans have been created already

* make AQL optimizer not skip "use-indexes-rule", even if enough execution plans
  have been created already

* fix double precision value loss in VelocyPack JSON parser

* added missing SSL support for arangorestore

* improved cluster import performance

* fix Foxx thumbnails on DC/OS

* fix Foxx configuration not being saved

* fix Foxx app access from within the frontend on DC/OS

* add option --default-replication-factor to arangorestore and simplify
  the control over the number of shards when restoring

* fix a bug in the VPack -> V8 conversion if special attributes _key,
  _id, _rev, _from and _to had non-string values, which is allowed
  below the top level

* fix malloc_usable_size for darwin


v3.0.1 (2016-06-30)
-------------------

* fixed periodic jobs: there should be only one instance running - even if it
  runs longer than the period

* increase max. number of collections in AQL queries from 32 to 256

* fixed issue #1916: header "authorization" is required" when opening
  services page

* fixed issue #1915: Explain: member out of range

* fixed issue #1914: fix unterminated buffer

* don't remove lockfile if we are the same (now stale) pid
  fixes docker setups (our pid will always be 1)

* do not use revision id comparisons in compaction for determining whether a
  revision is obsolete, but marker memory addresses
  this ensures revision ids don't matter when compacting documents

* escape Unicode characters in JSON HTTP responses
  this converts UTF-8 characters in HTTP responses of arangod into `\uXXXX`
  escape sequences. This makes the HTTP responses fit into the 7 bit ASCII
  character range, which speeds up HTTP response parsing for some clients,
  namely node.js/v8

* add write before read collections when starting a user transaction
  this allows specifying the same collection in both read and write mode without
  unintended side effects

* fixed buffer overrun that occurred when building very large result sets

* index lookup optimizations for primary index and edge index

* fixed "collection is a nullptr" issue when starting a traversal from a transaction

* enable /_api/import on coordinator servers


v3.0.0 (2016-06-22)
-------------------

* minor GUI fixxes

* fix for replication and nonces


v3.0.0-rc3 (2016-06-19)
-----------------------

* renamed various Foxx errors to no longer refer to Foxx services as apps

* adjusted various error messages in Foxx to be more informative

* specifying "files" in a Foxx manifest to be mounted at the service root
  no longer results in 404s when trying to access non-file routes

* undeclared path parameters in Foxx no longer break the service

* trusted reverse proxy support is now handled more consistently

* ArangoDB request compatibility and user are now exposed in Foxx

* all bundled NPM modules have been upgraded to their latest versions


v3.0.0-rc2 (2016-06-12)
-----------------------

* added option `--server.max-packet-size` for client tools

* renamed option `--server.ssl-protocol` to `--ssl.protocol` in client tools
  (was already done for arangod, but overlooked for client tools)

* fix handling of `--ssl.protocol` value 5 (TLS v1.2) in client tools, which
  claimed to support it but didn't

* config file can use '@include' to include a different config file as base


v3.0.0-rc1 (2016-06-10)
-----------------------

* the user management has changed: it now has users that are independent of
  databases. A user can have one or more database assigned to the user.

* forward ported V8 Comparator bugfix for inline heuristics from
  https://github.com/v8/v8/commit/5ff7901e24c2c6029114567de5a08ed0f1494c81

* changed to-string conversion for AQL objects and arrays, used by the AQL
  function `TO_STRING()` and implicit to-string casts in AQL

  - arrays are now converted into their JSON-stringify equivalents, e.g.

    - `[ ]` is now converted to `[]`
    - `[ 1, 2, 3 ]` is now converted to `[1,2,3]`
    - `[ "test", 1, 2 ] is now converted to `["test",1,2]`

    Previous versions of ArangoDB converted arrays with no members into the
    empty string, and non-empty arrays into a comma-separated list of member
    values, without the surrounding angular brackets. Additionally, string
    array members were not enclosed in quotes in the result string:

    - `[ ]` was converted to ``
    - `[ 1, 2, 3 ]` was converted to `1,2,3`
    - `[ "test", 1, 2 ] was converted to `test,1,2`

  - objects are now converted to their JSON-stringify equivalents, e.g.

    - `{ }` is converted to `{}`
    - `{ a: 1, b: 2 }` is converted to `{"a":1,"b":2}`
    - `{ "test" : "foobar" }` is converted to `{"test":"foobar"}`

    Previous versions of ArangoDB always converted objects into the string
    `[object Object]`

  This change affects also the AQL functions `CONCAT()` and `CONCAT_SEPARATOR()`
  which treated array values differently in previous versions. Previous versions
  of ArangoDB automatically flattened array values on the first level of the array,
  e.g. `CONCAT([1, 2, 3, [ 4, 5, 6 ]])` produced `1,2,3,4,5,6`. Now this will produce
  `[1,2,3,[4,5,6]]`. To flatten array members on the top level, you can now use
  the more explicit `CONCAT(FLATTEN([1, 2, 3, [4, 5, 6]], 1))`.

* added C++ implementations for AQL functions `SLICE()`, `CONTAINS()` and
  `RANDOM_TOKEN()`

* as a consequence of the upgrade to V8 version 5, the implementation of the
  JavaScript `Buffer` object had to be changed. JavaScript `Buffer` objects in
  ArangoDB now always store their data on the heap. There is no shared pool
  for small Buffer values, and no pointing into existing Buffer data when
  extracting slices. This change may increase the cost of creating Buffers with
  short contents or when peeking into existing Buffers, but was required for
  safer memory management and to prevent leaks.

* the `db` object's function `_listDatabases()` was renamed to just `_databases()`
  in order to make it more consistent with the existing `_collections()` function.
  Additionally the `db` object's `_listEndpoints()` function was renamed to just
  `_endpoints()`.

* changed default value of `--server.authentication` from `false` to `true` in
  configuration files etc/relative/arangod.conf and etc/arangodb/arangod.conf.in.
  This means the server will be started with authentication enabled by default,
  requiring all client connections to provide authentication data when connecting
  to ArangoDB. Authentication can still be turned off via setting the value of
  `--server.authentication` to `false` in ArangoDB's configuration files or by
  specifying the option on the command-line.

* Changed result format for querying all collections via the API GET `/_api/collection`.

  Previous versions of ArangoDB returned an object with an attribute named `collections`
  and an attribute named `names`. Both contained all available collections, but
  `collections` contained the collections as an array, and `names` contained the
  collections again, contained in an object in which the attribute names were the
  collection names, e.g.

  ```
  {
    "collections": [
      {"id":"5874437","name":"test","isSystem":false,"status":3,"type":2},
      {"id":"17343237","name":"something","isSystem":false,"status":3,"type":2},
      ...
    ],
    "names": {
      "test": {"id":"5874437","name":"test","isSystem":false,"status":3,"type":2},
      "something": {"id":"17343237","name":"something","isSystem":false,"status":3,"type":2},
      ...
    }
  }
  ```
  This result structure was redundant, and therefore has been simplified to just

  ```
  {
    "result": [
      {"id":"5874437","name":"test","isSystem":false,"status":3,"type":2},
      {"id":"17343237","name":"something","isSystem":false,"status":3,"type":2},
      ...
    ]
  }
  ```

  in ArangoDB 3.0.

* added AQL functions `TYPENAME()` and `HASH()`

* renamed arangob tool to arangobench

* added AQL string comparison operator `LIKE`

  The operator can be used to compare strings like this:

      value LIKE search

  The operator is currently implemented by calling the already existing AQL
  function `LIKE`.

  This change also makes `LIKE` an AQL keyword. Using `LIKE` in either case as
  an attribute or collection name in AQL thus requires quoting.

* make AQL optimizer rule "remove-unnecessary-calculations" fire in more cases

  The rule will now remove calculations that are used exactly once in other
  expressions (e.g. `LET a = doc RETURN a.value`) and calculations,
  or calculations that are just references (e.g. `LET a = b`).

* renamed AQL optimizer rule "merge-traversal-filter" to "optimize-traversals"
  Additionally, the optimizer rule will remove unused edge and path result variables
  from the traversal in case they are specified in the `FOR` section of the traversal,
  but not referenced later in the query. This saves constructing edges and paths
  results.

* added AQL optimizer rule "inline-subqueries"

  This rule can pull out certain subqueries that are used as an operand to a `FOR`
  loop one level higher, eliminating the subquery completely. For example, the query

      FOR i IN (FOR j IN [1,2,3] RETURN j) RETURN i

  will be transformed by the rule to:

      FOR i IN [1,2,3] RETURN i

  The query

      FOR name IN (FOR doc IN _users FILTER doc.status == 1 RETURN doc.name) LIMIT 2 RETURN name

  will be transformed into

      FOR tmp IN _users FILTER tmp.status == 1 LIMIT 2 RETURN tmp.name

  The rule will only fire when the subquery is used as an operand to a `FOR` loop, and
  if the subquery does not contain a `COLLECT` with an `INTO` variable.

* added new endpoint "srv://" for DNS service records

* The result order of the AQL functions VALUES and ATTRIBUTES has never been
  guaranteed and it only had the "correct" ordering by accident when iterating
  over objects that were not loaded from the database. This accidental behavior
  is now changed by introduction of VelocyPack. No ordering is guaranteed unless
  you specify the sort parameter.

* removed configure option `--enable-logger`

* added AQL array comparison operators

  All AQL comparison operators now also exist in an array variant. In the
  array variant, the operator is preceded with one of the keywords *ALL*, *ANY*
  or *NONE*. Using one of these keywords changes the operator behavior to
  execute the comparison operation for all, any, or none of its left hand
  argument values. It is therefore expected that the left hand argument
  of an array operator is an array.

  Examples:

      [ 1, 2, 3 ] ALL IN [ 2, 3, 4 ]   // false
      [ 1, 2, 3 ] ALL IN [ 1, 2, 3 ]   // true
      [ 1, 2, 3 ] NONE IN [ 3 ]        // false
      [ 1, 2, 3 ] NONE IN [ 23, 42 ]   // true
      [ 1, 2, 3 ] ANY IN [ 4, 5, 6 ]   // false
      [ 1, 2, 3 ] ANY IN [ 1, 42 ]     // true
      [ 1, 2, 3 ] ANY == 2             // true
      [ 1, 2, 3 ] ANY == 4             // false
      [ 1, 2, 3 ] ANY > 0              // true
      [ 1, 2, 3 ] ANY <= 1             // true
      [ 1, 2, 3 ] NONE < 99            // false
      [ 1, 2, 3 ] NONE > 10            // true
      [ 1, 2, 3 ] ALL > 2              // false
      [ 1, 2, 3 ] ALL > 0              // true
      [ 1, 2, 3 ] ALL >= 3             // false
      ["foo", "bar"] ALL != "moo"      // true
      ["foo", "bar"] NONE == "bar"     // false
      ["foo", "bar"] ANY == "foo"      // true

* improved AQL optimizer to remove unnecessary sort operations in more cases

* allow enclosing AQL identifiers in forward ticks in addition to using
  backward ticks

  This allows for convenient writing of AQL queries in JavaScript template strings
  (which are delimited with backticks themselves), e.g.

      var q = `FOR doc IN ´collection´ RETURN doc.´name´`;

* allow to set `print.limitString` to configure the number of characters
  to output before truncating

* make logging configurable per log "topic"

  `--log.level <level>` sets the global log level to <level>, e.g. `info`,
  `debug`, `trace`.

  `--log.level topic=<level>` sets the log level for a specific topic.
  Currently, the following topics exist: `collector`, `compactor`, `mmap`,
  `performance`, `queries`, and `requests`. `performance` and `requests` are
  set to FATAL by default. `queries` is set to info. All others are
  set to the global level by default.

  The new log option `--log.output <definition>` allows directing the global
  or per-topic log output to different outputs. The output definition
  "<definition>" can be one of

    "-" for stdin
    "+" for stderr
    "syslog://<syslog-facility>"
    "syslog://<syslog-facility>/<application-name>"
    "file://<relative-path>"

  The option can be specified multiple times in order to configure the output
  for different log topics. To set up a per-topic output configuration, use
  `--log.output <topic>=<definition>`, e.g.

    queries=file://queries.txt

  logs all queries to the file "queries.txt".

* the option `--log.requests-file` is now deprecated. Instead use

    `--log.level requests=info`
    `--log.output requests=file://requests.txt`

* the option `--log.facility` is now deprecated. Instead use

    `--log.output requests=syslog://facility`

* the option `--log.performance` is now deprecated. Instead use

    `--log.level performance=trace`

* removed option `--log.source-filter`

* removed configure option `--enable-logger`

* change collection directory names to include a random id component at the end

  The new pattern is `collection-<id>-<random>`, where `<id>` is the collection
  id and `<random>` is a random number. Previous versions of ArangoDB used a
  pattern `collection-<id>` without the random number.

  ArangoDB 3.0 understands both the old and name directory name patterns.

* removed mostly unused internal spin-lock implementation

* removed support for pre-Windows 7-style locks. This removes compatibility for
  Windows versions older than Windows 7 (e.g. Windows Vista, Windows XP) and
  Windows 2008R2 (e.g. Windows 2008).

* changed names of sub-threads started by arangod

* added option `--default-number-of-shards` to arangorestore, allowing creating
  collections with a specifiable number of shards from a non-cluster dump

* removed support for CoffeeScript source files

* removed undocumented SleepAndRequeue

* added WorkMonitor to inspect server threads

* when downloading a Foxx service from the web interface the suggested filename
  is now based on the service's mount path instead of simply "app.zip"

* the `@arangodb/request` response object now stores the parsed JSON response
  body in a property `json` instead of `body` when the request was made using the
  `json` option. The `body` instead contains the response body as a string.

* the Foxx API has changed significantly, 2.8 services are still supported
  using a backwards-compatible "legacy mode"


v2.8.12 (XXXX-XX-XX)
--------------------

* issue #2091: decrease connect timeout to 5 seconds on startup

* fixed issue #2072

* slightly better error diagnostics for some replication errors

* fixed issue #1977

* fixed issue in `INTERSECTION` AQL function with duplicate elements
  in the source arrays

* fixed issue #1962

* fixed issue #1959

* export aqlQuery template handler as require('org/arangodb').aql for forwards-compatibility


v2.8.11 (2016-07-13)
--------------------

* fixed array index batch insertion issues for hash indexes that caused problems when
  no elements remained for insertion

* fixed issue #1937


v2.8.10 (2016-07-01)
--------------------

* make sure next local _rev value used for a document is at least as high as the
  _rev value supplied by external sources such as replication

* make adding a collection in both read- and write-mode to a transaction behave as
  expected (write includes read). This prevents the `unregister collection used in
  transaction` error

* fixed sometimes invalid result for `byExample(...).count()` when an index plus
  post-filtering was used

* fixed "collection is a nullptr" issue when starting a traversal from a transaction

* honor the value of startup option `--database.wait-for-sync` (that is used to control
  whether new collections are created with `waitForSync` set to `true` by default) also
  when creating collections via the HTTP API (and thus the ArangoShell). When creating
  a collection via these mechanisms, the option was ignored so far, which was inconsistent.

* fixed issue #1826: arangosh --javascript.execute: internal error (geo index issue)

* fixed issue #1823: Arango crashed hard executing very simple query on windows


v2.8.9 (2016-05-13)
-------------------

* fixed escaping and quoting of extra parameters for executables in Mac OS X App

* added "waiting for" status variable to web interface collection figures view

* fixed undefined behavior in query cache invaldation

* fixed access to /_admin/statistics API in case statistics are disable via option
  `--server.disable-statistics`

* Foxx manager will no longer fail hard when Foxx store is unreachable unless installing
  a service from the Foxx store (e.g. when behind a firewall or GitHub is unreachable).


v2.8.8 (2016-04-19)
-------------------

* fixed issue #1805: Query: internal error (location: arangod/Aql/AqlValue.cpp:182).
  Please report this error to arangodb.com (while executing)

* allow specifying collection name prefixes for `_from` and `_to` in arangoimp:

  To avoid specifying complete document ids (consisting of collection names and document
  keys) for *_from* and *_to* values when importing edges with arangoimp, there are now
  the options *--from-collection-prefix* and *--to-collection-prefix*.

  If specified, these values will be automatically prepended to each value in *_from*
  (or *_to* resp.). This allows specifying only document keys inside *_from* and/or *_to*.

  *Example*

      > arangoimp --from-collection-prefix users --to-collection-prefix products ...

  Importing the following document will then create an edge between *users/1234* and
  *products/4321*:

  ```js
  { "_from" : "1234", "_to" : "4321", "desc" : "users/1234 is connected to products/4321" }
  ```

* requests made with the interactive system API documentation in the web interface
  (Swagger) will now respect the active database instead of always using `_system`


v2.8.7 (2016-04-07)
-------------------

* optimized primary=>secondary failover

* fix to-boolean conversion for documents in AQL

* expose the User-Agent HTTP header from the ArangoShell since Github seems to
  require it now, and we use the ArangoShell for fetching Foxx repositories from Github

* work with http servers that only send

* fixed potential race condition between compactor and collector threads

* fix removal of temporary directories on arangosh exit

* javadoc-style comments in Foxx services are no longer interpreted as
  Foxx comments outside of controller/script/exports files (#1748)

* removed remaining references to class syntax for Foxx Model and Repository
  from the documentation

* added a safe-guard for corrupted master-pointer


v2.8.6 (2016-03-23)
-------------------

* arangosh can now execute JavaScript script files that contain a shebang
  in the first line of the file. This allows executing script files directly.

  Provided there is a script file `/path/to/script.js` with the shebang
  `#!arangosh --javascript.execute`:

      > cat /path/to/script.js
      #!arangosh --javascript.execute
      print("hello from script.js");

  If the script file is made executable

      > chmod a+x /path/to/script.js

  it can be invoked on the shell directly and use arangosh for its execution:

      > /path/to/script.js
      hello from script.js

  This did not work in previous versions of ArangoDB, as the whole script contents
  (including the shebang) were treated as JavaScript code.
  Now shebangs in script files will now be ignored for all files passed to arangosh's
  `--javascript.execute` parameter.

  The alternative way of executing a JavaScript file with arangosh still works:

      > arangosh --javascript.execute /path/to/script.js
      hello from script.js

* added missing reset of traversal state for nested traversals.
  The state of nested traversals (a traversal in an AQL query that was
  located in a repeatedly executed subquery or inside another FOR loop)
  was not reset properly, so that multiple invocations of the same nested
  traversal with different start vertices led to the nested traversal
  always using the start vertex provided on the first invocation.

* fixed issue #1781: ArangoDB startup time increased tremendously

* fixed issue #1783: SIGHUP should rotate the log


v2.8.5 (2016-03-11)
-------------------

* Add OpenSSL handler for TLS V1.2 as sugested by kurtkincaid in #1771

* fixed issue #1765 (The webinterface should display the correct query time)
  and #1770 (Display ACTUAL query time in aardvark's AQL editor)

* Windows: the unhandled exception handler now calls the windows logging
  facilities directly without locks.
  This fixes lockups on crashes from the logging framework.

* improve nullptr handling in logger.

* added new endpoint "srv://" for DNS service records

* `org/arangodb/request` no longer sets the content-type header to the
  string "undefined" when no content-type header should be sent (issue #1776)


v2.8.4 (2016-03-01)
-------------------

* global modules are no longer incorrectly resolved outside the ArangoDB
  JavaScript directory or the Foxx service's root directory (issue #1577)

* improved error messages from Foxx and JavaScript (issues #1564, #1565, #1744)


v2.8.3 (2016-02-22)
-------------------

* fixed AQL filter condition collapsing for deeply-nested cases, potentially
  enabling usage of indexes in some dedicated cases

* added parentheses in AQL explain command output to correctly display precedence
  of logical and arithmetic operators

* Foxx Model event listeners defined on the model are now correctly invoked by
  the Repository methods (issue #1665)

* Deleting a Foxx service in the frontend should now always succeed even if the
  files no longer exist on the file system (issue #1358)

* Routing actions loaded from the database no longer throw exceptions when
  trying to load other modules using "require"

* The `org/arangodb/request` response object now sets a property `json` to the
  parsed JSON response body in addition to overwriting the `body` property when
  the request was made using the `json` option.

* Improved Windows stability

* Fixed a bug in the interactive API documentation that would escape slashes
  in document-handle fields. Document handles are now provided as separate
  fields for collection name and document key.


v2.8.2 (2016-02-09)
-------------------

* the continuous replication applier will now prevent the master's WAL logfiles
  from being removed if they are still needed by the applier on the slave. This
  should help slaves that suffered from masters garbage collection WAL logfiles
  which would have been needed by the slave later.

  The initial synchronization will block removal of still needed WAL logfiles
  on the master for 10 minutes initially, and will extend this period when further
  requests are made to the master. Initial synchronization hands over its handle
  for blocking logfile removal to the continuous replication when started via
  the *setupReplication* function. In this case, continuous replication will
  extend the logfile removal blocking period for the required WAL logfiles when
  the slave makes additional requests.

  All handles that block logfile removal will time out automatically after at
  most 5 minutes should a master not be contacted by the slave anymore (e.g. in
  case the slave's replication is turned off, the slaves loses the connection
  to the master or the slave goes down).

* added all-in-one function *setupReplication* to synchronize data from master
  to slave and start the continuous replication:

      require("@arangodb/replication").setupReplication(configuration);

  The command will return when the initial synchronization is finished and the
  continuous replication has been started, or in case the initial synchronization
  has failed.

  If the initial synchronization is successful, the command will store the given
  configuration on the slave. It also configures the continuous replication to start
  automatically if the slave is restarted, i.e. *autoStart* is set to *true*.

  If the command is run while the slave's replication applier is already running,
  it will first stop the running applier, drop its configuration and do a
  resynchronization of data with the master. It will then use the provided configration,
  overwriting any previously existing replication configuration on the slave.

  The following example demonstrates how to use the command for setting up replication
  for the *_system* database. Note that it should be run on the slave and not the
  master:

      db._useDatabase("_system");
      require("@arangodb/replication").setupReplication({
        endpoint: "tcp://master.domain.org:8529",
        username: "myuser",
        password: "mypasswd",
        verbose: false,
        includeSystem: false,
        incremental: true,
        autoResync: true
      });

* the *sync* and *syncCollection* functions now always start the data synchronization
  as an asynchronous server job. The call to *sync* or *syncCollection* will block
  until synchronization is either complete or has failed with an error. The functions
  will automatically poll the slave periodically for status updates.

  The main benefit is that the connection to the slave does not need to stay open
  permanently and is thus not affected by timeout issues. Additionally the caller does
  not need to query the synchronization status from the slave manually as this is
  now performed automatically by these functions.

* fixed undefined behavior when explaining some types of AQL traversals, fixed
  display of some types of traversals in AQL explain output


v2.8.1 (2016-01-29)
-------------------

* Improved AQL Pattern matching by allowing to specify a different traversal
  direction for one or many of the edge collections.

      FOR v, e, p IN OUTBOUND @start @@ec1, INBOUND @@ec2, @@ec3

  will traverse *ec1* and *ec3* in the OUTBOUND direction and for *ec2* it will use
  the INBOUND direction. These directions can be combined in arbitrary ways, the
  direction defined after *IN [steps]* will we used as default direction and can
  be overriden for specific collections.
  This feature is only available for collection lists, it is not possible to
  combine it with graph names.

* detect more types of transaction deadlocks early

* fixed display of relational operators in traversal explain output

* fixed undefined behavior in AQL function `PARSE_IDENTIFIER`

* added "engines" field to Foxx services generated in the admin interface

* added AQL function `IS_SAME_COLLECTION`:

  *IS_SAME_COLLECTION(collection, document)*: Return true if *document* has the same
  collection id as the collection specified in *collection*. *document* can either be
  a [document handle](../Glossary/README.md#document-handle) string, or a document with
  an *_id* attribute. The function does not validate whether the collection actually
  contains the specified document, but only compares the name of the specified collection
  with the collection name part of the specified document.
  If *document* is neither an object with an *id* attribute nor a *string* value,
  the function will return *null* and raise a warning.

      /* true */
      IS_SAME_COLLECTION('_users', '_users/my-user')
      IS_SAME_COLLECTION('_users', { _id: '_users/my-user' })

      /* false */
      IS_SAME_COLLECTION('_users', 'foobar/baz')
      IS_SAME_COLLECTION('_users', { _id: 'something/else' })


v2.8.0 (2016-01-25)
-------------------

* avoid recursive locking


v2.8.0-beta8 (2016-01-19)
-------------------------

* improved internal datafile statistics for compaction and compaction triggering
  conditions, preventing excessive growth of collection datafiles under some
  workloads. This should also fix issue #1596.

* renamed AQL optimizer rule `remove-collect-into` to `remove-collect-variables`

* fixed primary and edge index lookups prematurely aborting searches when the
  specified id search value contained a different collection than the collection
  the index was created for


v2.8.0-beta7 (2016-01-06)
-------------------------

* added vm.runInThisContext

* added AQL keyword `AGGREGATE` for use in AQL `COLLECT` statement

  Using `AGGREGATE` allows more efficient aggregation (incrementally while building
  the groups) than previous versions of AQL, which built group aggregates afterwards
  from the total of all group values.

  `AGGREGATE` can be used inside a `COLLECT` statement only. If used, it must follow
  the declaration of grouping keys:

      FOR doc IN collection
        COLLECT gender = doc.gender AGGREGATE minAge = MIN(doc.age), maxAge = MAX(doc.age)
        RETURN { gender, minAge, maxAge }

  or, if no grouping keys are used, it can follow the `COLLECT` keyword:

      FOR doc IN collection
        COLLECT AGGREGATE minAge = MIN(doc.age), maxAge = MAX(doc.age)
        RETURN {
  minAge, maxAge
}

  Only specific expressions are allowed on the right-hand side of each `AGGREGATE`
  assignment:

  - on the top level the expression must be a call to one of the supported aggregation
    functions `LENGTH`, `MIN`, `MAX`, `SUM`, `AVERAGE`, `STDDEV_POPULATION`, `STDDEV_SAMPLE`,
    `VARIANCE_POPULATION`, or `VARIANCE_SAMPLE`

  - the expression must not refer to variables introduced in the `COLLECT` itself

* Foxx: mocha test paths with wildcard characters (asterisks) now work on Windows

* reserved AQL keyword `NONE` for future use

* web interface: fixed a graph display bug concerning dashboard view

* web interface: fixed several bugs during the dashboard initialize process

* web interface: included several bugfixes: #1597, #1611, #1623

* AQL query optimizer now converts `LENGTH(collection-name)` to an optimized
  expression that returns the number of documents in a collection

* adjusted the behavior of the expansion (`[*]`) operator in AQL for non-array values

  In ArangoDB 2.8, calling the expansion operator on a non-array value will always
  return an empty array. Previous versions of ArangoDB expanded non-array values by
  calling the `TO_ARRAY()` function for the value, which for example returned an
  array with a single value for boolean, numeric and string input values, and an array
  with the object's values for an object input value. This behavior was inconsistent
  with how the expansion operator works for the array indexes in 2.8, so the behavior
  is now unified:

  - if the left-hand side operand of `[*]` is an array, the array will be returned as
    is when calling `[*]` on it
  - if the left-hand side operand of `[*]` is not an array, an empty array will be
    returned by `[*]`

  AQL queries that rely on the old behavior can be changed by either calling `TO_ARRAY`
  explicitly or by using the `[*]` at the correct position.

  The following example query will change its result in 2.8 compared to 2.7:

      LET values = "foo" RETURN values[*]

  In 2.7 the query has returned the array `[ "foo" ]`, but in 2.8 it will return an
  empty array `[ ]`. To make it return the array `[ "foo" ]` again, an explicit
  `TO_ARRAY` function call is needed in 2.8 (which in this case allows the removal
  of the `[*]` operator altogether). This also works in 2.7:

      LET values = "foo" RETURN TO_ARRAY(values)

  Another example:

      LET values = [ { name: "foo" }, { name: "bar" } ]
      RETURN values[*].name[*]

  The above returned `[ [ "foo" ], [ "bar" ] ] in 2.7. In 2.8 it will return
  `[ [ ], [ ] ]`, because the value of `name` is not an array. To change the results
  to the 2.7 style, the query can be changed to

      LET values = [ { name: "foo" }, { name: "bar" } ]
      RETURN values[* RETURN TO_ARRAY(CURRENT.name)]

  The above also works in 2.7.
  The following types of queries won't change:

      LET values = [ 1, 2, 3 ] RETURN values[*]
      LET values = [ { name: "foo" }, { name: "bar" } ] RETURN values[*].name
      LET values = [ { names: [ "foo", "bar" ] }, { names: [ "baz" ] } ] RETURN values[*].names[*]
      LET values = [ { names: [ "foo", "bar" ] }, { names: [ "baz" ] } ] RETURN values[*].names[**]

* slightly adjusted V8 garbage collection strategy so that collection eventually
  happens in all contexts that hold V8 external references to documents and
  collections.

  also adjusted default value of `--javascript.gc-frequency` from 10 seconds to
  15 seconds, as less internal operations are carried out in JavaScript.

* fixes for AQL optimizer and traversal

* added `--create-collection-type` option to arangoimp

  This allows specifying the type of the collection to be created when
  `--create-collection` is set to `true`.

* Foxx export cache should no longer break if a broken app is loaded in the
  web admin interface.


v2.8.0-beta2 (2015-12-16)
-------------------------

* added AQL query optimizer rule "sort-in-values"

  This rule pre-sorts the right-hand side operand of the `IN` and `NOT IN`
  operators so the operation can use a binary search with logarithmic complexity
  instead of a linear search. The rule is applied when the right-hand side
  operand of an `IN` or `NOT IN` operator in a filter condition is a variable that
  is defined in a different loop/scope than the operator itself. Additionally,
  the filter condition must consist of solely the `IN` or `NOT IN` operation
  in order to avoid any side-effects.

* changed collection status terminology in web interface for collections for
  which an unload request has been issued from `in the process of being unloaded`
  to `will be unloaded`.

* unloading a collection via the web interface will now trigger garbage collection
  in all v8 contexts and force a WAL flush. This increases the chances of perfoming
  the unload faster.

* added the following attributes to the result of `collection.figures()` and the
  corresponding HTTP API at `PUT /_api/collection/<name>/figures`:

  - `documentReferences`: The number of references to documents in datafiles
    that JavaScript code currently holds. This information can be used for
    debugging compaction and unload issues.
  - `waitingFor`: An optional string value that contains information about
    which object type is at the head of the collection's cleanup queue. This
    information can be used for debugging compaction and unload issues.
  - `compactionStatus.time`: The point in time the compaction for the collection
    was last executed. This information can be used for debugging compaction
    issues.
  - `compactionStatus.message`: The action that was performed when the compaction
    was last run for the collection. This information can be used for debugging
    compaction issues.

  Note: `waitingFor` and `compactionStatus` may be empty when called on a coordinator
  in a cluster.

* the compaction will now provide queryable status info that can be used to track
  its progress. The compaction status is displayed in the web interface, too.

* better error reporting for arangodump and arangorestore

* arangodump will now fail by default when trying to dump edges that
  refer to already dropped collections. This can be circumvented by
  specifying the option `--force true` when invoking arangodump

* fixed cluster upgrade procedure

* the AQL functions `NEAR` and `WITHIN` now have stricter validations
  for their input parameters `limit`, `radius` and `distance`. They may now throw
  exceptions when invalid parameters are passed that may have not led
  to exceptions in previous versions.

* deprecation warnings now log stack traces

* Foxx: improved backwards compatibility with 2.5 and 2.6

  - reverted Model and Repository back to non-ES6 "classes" because of
    compatibility issues when using the extend method with a constructor

  - removed deprecation warnings for extend and controller.del

  - restored deprecated method Model.toJSONSchema

  - restored deprecated `type`, `jwt` and `sessionStorageApp` options
    in Controller#activateSessions

* Fixed a deadlock problem in the cluster


v2.8.0-beta1 (2015-12-06)
-------------------------

* added AQL function `IS_DATESTRING(value)`

  Returns true if *value* is a string that can be used in a date function.
  This includes partial dates such as *2015* or *2015-10* and strings containing
  invalid dates such as *2015-02-31*. The function will return false for all
  non-string values, even if some of them may be usable in date functions.


v2.8.0-alpha1 (2015-12-03)
--------------------------

* added AQL keywords `GRAPH`, `OUTBOUND`, `INBOUND` and `ANY` for use in graph
  traversals, reserved AQL keyword `ALL` for future use

  Usage of these keywords as collection names, variable names or attribute names
  in AQL queries will not be possible without quoting. For example, the following
  AQL query will still work as it uses a quoted collection name and a quoted
  attribute name:

      FOR doc IN `OUTBOUND`
        RETURN doc.`any`

* issue #1593: added AQL `POW` function for exponentation

* added cluster execution site info in explain output for AQL queries

* replication improvements:

  - added `autoResync` configuration parameter for continuous replication.

    When set to `true`, a replication slave will automatically trigger a full data
    re-synchronization with the master when the master cannot provide the log data
    the slave had asked for. Note that `autoResync` will only work when the option
    `requireFromPresent` is also set to `true` for the continuous replication, or
    when the continuous syncer is started and detects that no start tick is present.

    Automatic re-synchronization may transfer a lot of data from the master to the
    slave and may be expensive. It is therefore turned off by default.
    When turned off, the slave will never perform an automatic re-synchronization
    with the master.

  - added `idleMinWaitTime` and `idleMaxWaitTime` configuration parameters for
    continuous replication.

    These parameters can be used to control the minimum and maximum wait time the
    slave will (intentionally) idle and not poll for master log changes in case the
    master had sent the full logs already.
    The `idleMaxWaitTime` value will only be used when `adapativePolling` is set
    to `true`. When `adaptivePolling` is disable, only `idleMinWaitTime` will be
    used as a constant time span in which the slave will not poll the master for
    further changes. The default values are 0.5 seconds for `idleMinWaitTime` and
    2.5 seconds for `idleMaxWaitTime`, which correspond to the hard-coded values
    used in previous versions of ArangoDB.

  - added `initialSyncMaxWaitTime` configuration parameter for initial and continuous
    replication

    This option controls the maximum wait time (in seconds) that the initial
    synchronization will wait for a response from the master when fetching initial
    collection data. If no response is received within this time period, the initial
    synchronization will give up and fail. This option is also relevant for
    continuous replication in case *autoResync* is set to *true*, as then the
    continuous replication may trigger a full data re-synchronization in case
    the master cannot the log data the slave had asked for.

  - HTTP requests sent from the slave to the master during initial synchronization
    will now be retried if they fail with connection problems.

  - the initial synchronization now logs its progress so it can be queried using
    the regular replication status check APIs.

  - added `async` attribute for `sync` and `syncCollection` operations called from
    the ArangoShell. Setthing this attribute to `true` will make the synchronization
    job on the server go into the background, so that the shell does not block. The
    status of the started asynchronous synchronization job can be queried from the
    ArangoShell like this:

        /* starts initial synchronization */
        var replication = require("@arangodb/replication");
        var id = replication.sync({
          endpoint: "tcp://master.domain.org:8529",
          username: "myuser",
          password: "mypasswd",
          async: true
       });

       /* now query the id of the returned async job and print the status */
       print(replication.getSyncResult(id));

    The result of `getSyncResult()` will be `false` while the server-side job
    has not completed, and different to `false` if it has completed. When it has
    completed, all job result details will be returned by the call to `getSyncResult()`.


* fixed non-deterministic query results in some cluster queries

* fixed issue #1589

* return HTTP status code 410 (gone) instead of HTTP 408 (request timeout) for
  server-side operations that are canceled / killed. Sending 410 instead of 408
  prevents clients from re-starting the same (canceled) operation. Google Chrome
  for example sends the HTTP request again in case it is responded with an HTTP
  408, and this is exactly the opposite of the desired behavior when an operation
  is canceled / killed by the user.

* web interface: queries in AQL editor now cancelable

* web interface: dashboard - added replication information

* web interface: AQL editor now supports bind parameters

* added startup option `--server.hide-product-header` to make the server not send
  the HTTP response header `"Server: ArangoDB"` in its HTTP responses. By default,
  the option is turned off so the header is still sent as usual.

* added new AQL function `UNSET_RECURSIVE` to recursively unset attritutes from
  objects/documents

* switched command-line editor in ArangoShell and arangod to linenoise-ng

* added automatic deadlock detection for transactions

  In case a deadlock is detected, a multi-collection operation may be rolled back
  automatically and fail with error 29 (`deadlock detected`). Client code for
  operations containing more than one collection should be aware of this potential
  error and handle it accordingly, either by giving up or retrying the transaction.

* Added C++ implementations for the AQL arithmetic operations and the following
  AQL functions:
  - ABS
  - APPEND
  - COLLECTIONS
  - CURRENT_DATABASE
  - DOCUMENT
  - EDGES
  - FIRST
  - FIRST_DOCUMENT
  - FIRST_LIST
  - FLATTEN
  - FLOOR
  - FULLTEXT
  - LAST
  - MEDIAN
  - MERGE_RECURSIVE
  - MINUS
  - NEAR
  - NOT_NULL
  - NTH
  - PARSE_IDENTIFIER
  - PERCENTILE
  - POP
  - POSITION
  - PUSH
  - RAND
  - RANGE
  - REMOVE_NTH
  - REMOVE_VALUE
  - REMOVE_VALUES
  - ROUND
  - SHIFT
  - SQRT
  - STDDEV_POPULATION
  - STDDEV_SAMPLE
  - UNSHIFT
  - VARIANCE_POPULATION
  - VARIANCE_SAMPLE
  - WITHIN
  - ZIP

* improved performance of skipping over many documents in an AQL query when no
  indexes and no filters are used, e.g.

      FOR doc IN collection
        LIMIT 1000000, 10
        RETURN doc

* Added array indexes

  Hash indexes and skiplist indexes can now optionally be defined for array values
  so they index individual array members.

  To define an index for array values, the attribute name is extended with the
  expansion operator `[*]` in the index definition:

      arangosh> db.colName.ensureHashIndex("tags[*]");

  When given the following document

      { tags: [ "AQL", "ArangoDB", "Index" ] }

  the index will now contain the individual values `"AQL"`, `"ArangoDB"` and `"Index"`.

  Now the index can be used for finding all documents having `"ArangoDB"` somewhere in their
  tags array using the following AQL query:

      FOR doc IN colName
        FILTER "ArangoDB" IN doc.tags[*]
        RETURN doc

* rewrote AQL query optimizer rule `use-index-range` and renamed it to `use-indexes`.
  The name change affects rule names in the optimizer's output.

* rewrote AQL execution node `IndexRangeNode` and renamed it to `IndexNode`. The name
  change affects node names in the optimizer's explain output.

* added convenience function `db._explain(query)` for human-readable explanation
  of AQL queries

* module resolution as used by `require` now behaves more like in node.js

* the `org/arangodb/request` module now returns response bodies for error responses
  by default. The old behavior of not returning bodies for error responses can be
  re-enabled by explicitly setting the option `returnBodyOnError` to `false` (#1437)


v2.7.6 (2016-01-30)
-------------------

* detect more types of transaction deadlocks early


v2.7.5 (2016-01-22)
-------------------

* backported added automatic deadlock detection for transactions

  In case a deadlock is detected, a multi-collection operation may be rolled back
  automatically and fail with error 29 (`deadlock detected`). Client code for
  operations containing more than one collection should be aware of this potential
  error and handle it accordingly, either by giving up or retrying the transaction.

* improved internal datafile statistics for compaction and compaction triggering
  conditions, preventing excessive growth of collection datafiles under some
  workloads. This should also fix issue #1596.

* Foxx export cache should no longer break if a broken app is loaded in the
  web admin interface.

* Foxx: removed some incorrect deprecation warnings.

* Foxx: mocha test paths with wildcard characters (asterisks) now work on Windows


v2.7.4 (2015-12-21)
-------------------

* slightly adjusted V8 garbage collection strategy so that collection eventually
  happens in all contexts that hold V8 external references to documents and
  collections.

* added the following attributes to the result of `collection.figures()` and the
  corresponding HTTP API at `PUT /_api/collection/<name>/figures`:

  - `documentReferences`: The number of references to documents in datafiles
    that JavaScript code currently holds. This information can be used for
    debugging compaction and unload issues.
  - `waitingFor`: An optional string value that contains information about
    which object type is at the head of the collection's cleanup queue. This
    information can be used for debugging compaction and unload issues.
  - `compactionStatus.time`: The point in time the compaction for the collection
    was last executed. This information can be used for debugging compaction
    issues.
  - `compactionStatus.message`: The action that was performed when the compaction
    was last run for the collection. This information can be used for debugging
    compaction issues.

  Note: `waitingFor` and `compactionStatus` may be empty when called on a coordinator
  in a cluster.

* the compaction will now provide queryable status info that can be used to track
  its progress. The compaction status is displayed in the web interface, too.


v2.7.3 (2015-12-17)
-------------------

* fixed some replication value conversion issues when replication applier properties
  were set via ArangoShell

* fixed disappearing of documents for collections transferred via `sync` or
  `syncCollection` if the collection was dropped right before synchronization
  and drop and (re-)create collection markers were located in the same WAL file

* fixed an issue where overwriting the system sessions collection would break
  the web interface when authentication is enabled


v2.7.2 (2015-12-01)
-------------------

* replication improvements:

  - added `autoResync` configuration parameter for continuous replication.

    When set to `true`, a replication slave will automatically trigger a full data
    re-synchronization with the master when the master cannot provide the log data
    the slave had asked for. Note that `autoResync` will only work when the option
    `requireFromPresent` is also set to `true` for the continuous replication, or
    when the continuous syncer is started and detects that no start tick is present.

    Automatic re-synchronization may transfer a lot of data from the master to the
    slave and may be expensive. It is therefore turned off by default.
    When turned off, the slave will never perform an automatic re-synchronization
    with the master.

  - added `idleMinWaitTime` and `idleMaxWaitTime` configuration parameters for
    continuous replication.

    These parameters can be used to control the minimum and maximum wait time the
    slave will (intentionally) idle and not poll for master log changes in case the
    master had sent the full logs already.
    The `idleMaxWaitTime` value will only be used when `adapativePolling` is set
    to `true`. When `adaptivePolling` is disable, only `idleMinWaitTime` will be
    used as a constant time span in which the slave will not poll the master for
    further changes. The default values are 0.5 seconds for `idleMinWaitTime` and
    2.5 seconds for `idleMaxWaitTime`, which correspond to the hard-coded values
    used in previous versions of ArangoDB.

  - added `initialSyncMaxWaitTime` configuration parameter for initial and continuous
    replication

    This option controls the maximum wait time (in seconds) that the initial
    synchronization will wait for a response from the master when fetching initial
    collection data. If no response is received within this time period, the initial
    synchronization will give up and fail. This option is also relevant for
    continuous replication in case *autoResync* is set to *true*, as then the
    continuous replication may trigger a full data re-synchronization in case
    the master cannot the log data the slave had asked for.

  - HTTP requests sent from the slave to the master during initial synchronization
    will now be retried if they fail with connection problems.

  - the initial synchronization now logs its progress so it can be queried using
    the regular replication status check APIs.

* fixed non-deterministic query results in some cluster queries

* added missing lock instruction for primary index in compactor size calculation

* fixed issue #1589

* fixed issue #1583

* fixed undefined behavior when accessing the top level of a document with the `[*]`
  operator

* fixed potentially invalid pointer access in shaper when the currently accessed
  document got re-located by the WAL collector at the very same time

* Foxx: optional configuration options no longer log validation errors when assigned
  empty values (#1495)

* Foxx: constructors provided to Repository and Model sub-classes via extend are
  now correctly called (#1592)


v2.7.1 (2015-11-07)
-------------------

* switch to linenoise next generation

* exclude `_apps` collection from replication

  The slave has its own `_apps` collection which it populates on server start.
  When replicating data from the master to the slave, the data from the master may
  clash with the slave's own data in the `_apps` collection. Excluding the `_apps`
  collection from replication avoids this.

* disable replication appliers when starting in modes `--upgrade`, `--no-server`
  and `--check-upgrade`

* more detailed output in arango-dfdb

* fixed "no start tick" issue in replication applier

  This error could occur after restarting a slave server after a shutdown
  when no data was ever transferred from the master to the slave via the
  continuous replication

* fixed problem during SSL client connection abort that led to scheduler thread
  staying at 100% CPU saturation

* fixed potential segfault in AQL `NEIGHBORS` function implementation when C++ function
  variant was used and collection names were passed as strings

* removed duplicate target for some frontend JavaScript files from the Makefile

* make AQL function `MERGE()` work on a single array parameter, too.
  This allows combining the attributes of multiple objects from an array into
  a single object, e.g.

      RETURN MERGE([
        { foo: 'bar' },
        { quux: 'quetzalcoatl', ruled: true },
        { bar: 'baz', foo: 'done' }
      ])

  will now return:

      {
        "foo": "done",
        "quux": "quetzalcoatl",
        "ruled": true,
        "bar": "baz"
      }

* fixed potential deadlock in collection status changing on Windows

* fixed hard-coded `incremental` parameter in shell implementation of
  `syncCollection` function in replication module

* fix for GCC5: added check for '-stdlib' option


v2.7.0 (2015-10-09)
-------------------

* fixed request statistics aggregation
  When arangod was started in supervisor mode, the request statistics always showed
  0 requests, as the statistics aggregation thread did not run then.

* read server configuration files before dropping privileges. this ensures that
  the SSL keyfile specified in the configuration can be read with the server's start
  privileges (i.e. root when using a standard ArangoDB package).

* fixed replication with a 2.6 replication configuration and issues with a 2.6 master

* raised default value of `--server.descriptors-minimum` to 1024

* allow Foxx apps to be installed underneath URL path `/_open/`, so they can be
  (intentionally) accessed without authentication.

* added *allowImplicit* sub-attribute in collections declaration of transactions.
  The *allowImplicit* attributes allows making transactions fail should they
  read-access a collection that was not explicitly declared in the *collections*
  array of the transaction.

* added "special" password ARANGODB_DEFAULT_ROOT_PASSWORD. If you pass
  ARANGODB_DEFAULT_ROOT_PASSWORD as password, it will read the password
  from the environment variable ARANGODB_DEFAULT_ROOT_PASSWORD


v2.7.0-rc2 (2015-09-22)
-----------------------

* fix over-eager datafile compaction

  This should reduce the need to compact directly after loading a collection when a
  collection datafile contained many insertions and updates for the same documents. It
  should also prevent from re-compacting already merged datafiles in case not many
  changes were made. Compaction will also make fewer index lookups than before.

* added `syncCollection()` function in module `org/arangodb/replication`

  This allows synchronizing the data of a single collection from a master to a slave
  server. Synchronization can either restore the whole collection by transferring all
  documents from the master to the slave, or incrementally by only transferring documents
  that differ. This is done by partitioning the collection's entire key space into smaller
  chunks and comparing the data chunk-wise between master and slave. Only chunks that are
  different will be re-transferred.

  The `syncCollection()` function can be used as follows:

      require("org/arangodb/replication").syncCollection(collectionName, options);

  e.g.

      require("org/arangodb/replication").syncCollection("myCollection", {
        endpoint: "tcp://127.0.0.1:8529",  /* master */
        username: "root",                  /* username for master */
        password: "secret",                /* password for master */
        incremental: true                  /* use incremental mode */
      });


* additionally allow the following characters in document keys:

  `(` `)` `+` `,` `=` `;` `$` `!` `*` `'` `%`


v2.7.0-rc1 (2015-09-17)
-----------------------

* removed undocumented server-side-only collection functions:
  * collection.OFFSET()
  * collection.NTH()
  * collection.NTH2()
  * collection.NTH3()

* upgraded Swagger to version 2.0 for the Documentation

  This gives the user better prepared test request structures.
  More conversions will follow so finally client libraries can be auto-generated.

* added extra AQL functions for date and time calculation and manipulation.
  These functions were contributed by GitHub users @CoDEmanX and @friday.
  A big thanks for their work!

  The following extra date functions are available from 2.7 on:

  * `DATE_DAYOFYEAR(date)`: Returns the day of year number of *date*.
    The return values range from 1 to 365, or 366 in a leap year respectively.

  * `DATE_ISOWEEK(date)`: Returns the ISO week date of *date*.
    The return values range from 1 to 53. Monday is considered the first day of the week.
    There are no fractional weeks, thus the last days in December may belong to the first
    week of the next year, and the first days in January may be part of the previous year's
    last week.

  * `DATE_LEAPYEAR(date)`: Returns whether the year of *date* is a leap year.

  * `DATE_QUARTER(date)`: Returns the quarter of the given date (1-based):
    * 1: January, February, March
    * 2: April, May, June
    * 3: July, August, September
    * 4: October, November, December

  - *DATE_DAYS_IN_MONTH(date)*: Returns the number of days in *date*'s month (28..31).

  * `DATE_ADD(date, amount, unit)`: Adds *amount* given in *unit* to *date* and
    returns the calculated date.

    *unit* can be either of the following to specify the time unit to add or
    subtract (case-insensitive):
    - y, year, years
    - m, month, months
    - w, week, weeks
    - d, day, days
    - h, hour, hours
    - i, minute, minutes
    - s, second, seconds
    - f, millisecond, milliseconds

    *amount* is the number of *unit*s to add (positive value) or subtract
    (negative value).

  * `DATE_SUBTRACT(date, amount, unit)`: Subtracts *amount* given in *unit* from
    *date* and returns the calculated date.

    It works the same as `DATE_ADD()`, except that it subtracts. It is equivalent
    to calling `DATE_ADD()` with a negative amount, except that `DATE_SUBTRACT()`
    can also subtract ISO durations. Note that negative ISO durations are not
    supported (i.e. starting with `-P`, like `-P1Y`).

  * `DATE_DIFF(date1, date2, unit, asFloat)`: Calculate the difference
    between two dates in given time *unit*, optionally with decimal places.
    Returns a negative value if *date1* is greater than *date2*.

  * `DATE_COMPARE(date1, date2, unitRangeStart, unitRangeEnd)`: Compare two
    partial dates and return true if they match, false otherwise. The parts to
    compare are defined by a range of time units.

    The full range is: years, months, days, hours, minutes, seconds, milliseconds.
    Pass the unit to start from as *unitRangeStart*, and the unit to end with as
    *unitRangeEnd*. All units in between will be compared. Leave out *unitRangeEnd*
    to only compare *unitRangeStart*.

  * `DATE_FORMAT(date, format)`: Format a date according to the given format string.
    It supports the following placeholders (case-insensitive):
    - %t: timestamp, in milliseconds since midnight 1970-01-01
    - %z: ISO date (0000-00-00T00:00:00.000Z)
    - %w: day of week (0..6)
    - %y: year (0..9999)
    - %yy: year (00..99), abbreviated (last two digits)
    - %yyyy: year (0000..9999), padded to length of 4
    - %yyyyyy: year (-009999 .. +009999), with sign prefix and padded to length of 6
    - %m: month (1..12)
    - %mm: month (01..12), padded to length of 2
    - %d: day (1..31)
    - %dd: day (01..31), padded to length of 2
    - %h: hour (0..23)
    - %hh: hour (00..23), padded to length of 2
    - %i: minute (0..59)
    - %ii: minute (00..59), padded to length of 2
    - %s: second (0..59)
    - %ss: second (00..59), padded to length of 2
    - %f: millisecond (0..999)
    - %fff: millisecond (000..999), padded to length of 3
    - %x: day of year (1..366)
    - %xxx: day of year (001..366), padded to length of 3
    - %k: ISO week date (1..53)
    - %kk: ISO week date (01..53), padded to length of 2
    - %l: leap year (0 or 1)
    - %q: quarter (1..4)
    - %a: days in month (28..31)
    - %mmm: abbreviated English name of month (Jan..Dec)
    - %mmmm: English name of month (January..December)
    - %www: abbreviated English name of weekday (Sun..Sat)
    - %wwww: English name of weekday (Sunday..Saturday)
    - %&: special escape sequence for rare occasions
    - %%: literal %
    - %: ignored

* new WAL logfiles and datafiles are now created non-sparse

  This prevents SIGBUS signals being raised when memory of a sparse datafile is accessed
  and the disk is full and the accessed file part is not actually disk-backed. In
  this case the mapped memory region is not necessarily backed by physical memory, and
  accessing the memory may raise SIGBUS and crash arangod.

* the `internal.download()` function and the module `org/arangodb/request` used some
  internal library function that handled the sending of HTTP requests from inside of
  ArangoDB. This library unconditionally set an HTTP header `Accept-Encoding: gzip`
  in all outgoing HTTP requests.

  This has been fixed in 2.7, so `Accept-Encoding: gzip` is not set automatically anymore.
  Additionally, the header `User-Agent: ArangoDB` is not set automatically either. If
  client applications desire to send these headers, they are free to add it when
  constructing the requests using the `download` function or the request module.

* fixed issue #1436: org/arangodb/request advertises deflate without supporting it

* added template string generator function `aqlQuery` for generating AQL queries

  This can be used to generate safe AQL queries with JavaScript parameter
  variables or expressions easily:

      var name = 'test';
      var attributeName = '_key';
      var query = aqlQuery`FOR u IN users FILTER u.name == ${name} RETURN u.${attributeName}`;
      db._query(query);

* report memory usage for document header data (revision id, pointer to data etc.)
  in `db.collection.figures()`. The memory used for document headers will now
  show up in the already existing attribute `indexes.size`. Due to that, the index
  sizes reported by `figures()` in 2.7 will be higher than those reported by 2.6,
  but the 2.7 values are more accurate.

* IMPORTANT CHANGE: the filenames in dumps created by arangodump now contain
  not only the name of the dumped collection, but also an additional 32-digit hash
  value. This is done to prevent overwriting dump files in case-insensitive file
  systems when there exist multiple collections with the same name (but with
  different cases).

  For example, if a database has two collections: `test` and `Test`, previous
  versions of ArangoDB created the files

  * `test.structure.json` and `test.data.json` for collection `test`
  * `Test.structure.json` and `Test.data.json` for collection `Test`

  This did not work for case-insensitive filesystems, because the files for the
  second collection would have overwritten the files of the first. arangodump in
  2.7 will create the following filenames instead:

  * `test_098f6bcd4621d373cade4e832627b4f6.structure.json` and `test_098f6bcd4621d373cade4e832627b4f6.data.json`
  * `Test_0cbc6611f5540bd0809a388dc95a615b.structure.json` and `Test_0cbc6611f5540bd0809a388dc95a615b.data.json`

  These filenames will be unambiguous even in case-insensitive filesystems.

* IMPORTANT CHANGE: make arangod actually close lingering client connections
  when idle for at least the duration specified via `--server.keep-alive-timeout`.
  In previous versions of ArangoDB, connections were not closed by the server
  when the timeout was reached and the client was still connected. Now the
  connection is properly closed by the server in case of timeout. Client
  applications relying on the old behavior may now need to reconnect to the
  server when their idle connections time out and get closed (note: connections
  being idle for a long time may be closed by the OS or firewalls anyway -
  client applications should be aware of that and try to reconnect).

* IMPORTANT CHANGE: when starting arangod, the server will drop the process
  privileges to the specified values in options `--server.uid` and `--server.gid`
  instantly after parsing the startup options.

  That means when either `--server.uid` or `--server.gid` are set, the privilege
  change will happen earlier. This may prevent binding the server to an endpoint
  with a port number lower than 1024 if the arangodb user has no privileges
  for that. Previous versions of ArangoDB changed the privileges later, so some
  startup actions were still carried out under the invoking user (i.e. likely
  *root* when started via init.d or system scripts) and especially binding to
  low port numbers was still possible there.

  The default privileges for user *arangodb* will not be sufficient for binding
  to port numbers lower than 1024. To have an ArangoDB 2.7 bind to a port number
  lower than 1024, it needs to be started with either a different privileged user,
  or the privileges of the *arangodb* user have to raised manually beforehand.

* added AQL optimizer rule `patch-update-statements`

* Linux startup scripts and systemd configuration for arangod now try to
  adjust the NOFILE (number of open files) limits for the process. The limit
  value is set to 131072 (128k) when ArangoDB is started via start/stop
  commands

* When ArangoDB is started/stopped manually via the start/stop commands, the
  main process will wait for up to 10 seconds after it forks the supervisor
  and arangod child processes. If the startup fails within that period, the
  start/stop script will fail with an exit code other than zero. If the
  startup of the supervisor or arangod is still ongoing after 10 seconds,
  the main program will still return with exit code 0. The limit of 10 seconds
  is arbitrary because the time required for a startup is not known in advance.

* added startup option `--database.throw-collection-not-loaded-error`

  Accessing a not-yet loaded collection will automatically load a collection
  on first access. This flag controls what happens in case an operation
  would need to wait for another thread to finalize loading a collection. If
  set to *true*, then the first operation that accesses an unloaded collection
  will load it. Further threads that try to access the same collection while
  it is still loading immediately fail with an error (1238, *collection not loaded*).
  This is to prevent all server threads from being blocked while waiting on the
  same collection to finish loading. When the first thread has completed loading
  the collection, the collection becomes regularly available, and all operations
  from that point on can be carried out normally, and error 1238 will not be
  thrown anymore for that collection.

  If set to *false*, the first thread that accesses a not-yet loaded collection
  will still load it. Other threads that try to access the collection while
  loading will not fail with error 1238 but instead block until the collection
  is fully loaded. This configuration might lead to all server threads being
  blocked because they are all waiting for the same collection to complete
  loading. Setting the option to *true* will prevent this from happening, but
  requires clients to catch error 1238 and react on it (maybe by scheduling
  a retry for later).

  The default value is *false*.

* added better control-C support in arangosh

  When CTRL-C is pressed in arangosh, it will now print a `^C` first. Pressing
  CTRL-C again will reset the prompt if something was entered before, or quit
  arangosh if no command was entered directly before.

  This affects the arangosh version build with Readline-support only (Linux
  and MacOS).

  The MacOS version of ArangoDB for Homebrew now depends on Readline, too. The
  Homebrew formula has been changed accordingly.
  When self-compiling ArangoDB on MacOS without Homebrew, Readline now is a
  prerequisite.

* increased default value for collection-specific `indexBuckets` value from 1 to 8

  Collections created from 2.7 on will use the new default value of `8` if not
  overridden on collection creation or later using
  `collection.properties({ indexBuckets: ... })`.

  The `indexBuckets` value determines the number of buckets to use for indexes of
  type `primary`, `hash` and `edge`. Having multiple index buckets allows splitting
  an index into smaller components, which can be filled in parallel when a collection
  is loading. Additionally, resizing and reallocation of indexes are faster and
  less intrusive if the index uses multiple buckets, because resize and reallocation
  will affect only data in a single bucket instead of all index values.

  The index buckets will be filled in parallel when loading a collection if the collection
  has an `indexBuckets` value greater than 1 and the collection contains a significant
  amount of documents/edges (the current threshold is 256K documents but this value
  may change in future versions of ArangoDB).

* changed HTTP client to use poll instead of select on Linux and MacOS

  This affects the ArangoShell and user-defined JavaScript code running inside
  arangod that initiates its own HTTP calls.

  Using poll instead of select allows using arbitrary high file descriptors
  (bigger than the compiled in FD_SETSIZE). Server connections are still handled using
  epoll, which has never been affected by FD_SETSIZE.

* implemented AQL `LIKE` function using ICU regexes

* added `RETURN DISTINCT` for AQL queries to return unique results:

      FOR doc IN collection
        RETURN DISTINCT doc.status

  This change also introduces `DISTINCT` as an AQL keyword.

* removed `createNamedQueue()` and `addJob()` functions from org/arangodb/tasks

* use less locks and more atomic variables in the internal dispatcher
  and V8 context handling implementations. This leads to improved throughput in
  some ArangoDB internals and allows for higher HTTP request throughput for
  many operations.

  A short overview of the improvements can be found here:

  https://www.arangodb.com/2015/08/throughput-enhancements/

* added shorthand notation for attribute names in AQL object literals:

      LET name = "Peter"
      LET age = 42
      RETURN { name, age }

  The above is the shorthand equivalent of the generic form

      LET name = "Peter"
      LET age = 42
      RETURN { name : name, age : age }

* removed configure option `--enable-timings`

  This option did not have any effect.

* removed configure option `--enable-figures`

  This option previously controlled whether HTTP request statistics code was
  compiled into ArangoDB or not. The previous default value was `true` so
  statistics code was available in official packages. Setting the option to
  `false` led to compile errors so it is doubtful the default value was
  ever changed. By removing the option some internal statistics code was also
  simplified.

* removed run-time manipulation methods for server endpoints:

  * `db._removeEndpoint()`
  * `db._configureEndpoint()`
  * HTTP POST `/_api/endpoint`
  * HTTP DELETE `/_api/endpoint`

* AQL query result cache

  The query result cache can optionally cache the complete results of all or selected AQL queries.
  It can be operated in the following modes:

  * `off`: the cache is disabled. No query results will be stored
  * `on`: the cache will store the results of all AQL queries unless their `cache`
    attribute flag is set to `false`
  * `demand`: the cache will store the results of AQL queries that have their
    `cache` attribute set to `true`, but will ignore all others

  The mode can be set at server startup using the `--database.query-cache-mode` configuration
  option and later changed at runtime.

  The following HTTP REST APIs have been added for controlling the query cache:

  * HTTP GET `/_api/query-cache/properties`: returns the global query cache configuration
  * HTTP PUT `/_api/query-cache/properties`: modifies the global query cache configuration
  * HTTP DELETE `/_api/query-cache`: invalidates all results in the query cache

  The following JavaScript functions have been added for controlling the query cache:

  * `require("org/arangodb/aql/cache").properties()`: returns the global query cache configuration
  * `require("org/arangodb/aql/cache").properties(properties)`: modifies the global query cache configuration
  * `require("org/arangodb/aql/cache").clear()`: invalidates all results in the query cache

* do not link arangoimp against V8

* AQL function call arguments optimization

  This will lead to arguments in function calls inside AQL queries not being copied but passed
  by reference. This may speed up calls to functions with bigger argument values or queries that
  call functions a lot of times.

* upgraded V8 version to 4.3.61

* removed deprecated AQL `SKIPLIST` function.

  This function was introduced in older versions of ArangoDB with a less powerful query optimizer to
  retrieve data from a skiplist index using a `LIMIT` clause. It was marked as deprecated in ArangoDB
  2.6.

  Since ArangoDB 2.3 the behavior of the `SKIPLIST` function can be emulated using regular AQL
  constructs, e.g.

      FOR doc IN @@collection
        FILTER doc.value >= @value
        SORT doc.value DESC
        LIMIT 1
        RETURN doc

* the `skip()` function for simple queries does not accept negative input any longer.
  This feature was deprecated in 2.6.0.

* fix exception handling

  In some cases JavaScript exceptions would re-throw without information of the original problem.
  Now the original exception is logged for failure analysis.

* based REST API method PUT `/_api/simple/all` on the cursor API and make it use AQL internally.

  The change speeds up this REST API method and will lead to additional query information being
  returned by the REST API. Clients can use this extra information or ignore it.

* Foxx Queue job success/failure handlers arguments have changed from `(jobId, jobData, result, jobFailures)` to `(result, jobData, job)`.

* added Foxx Queue job options `repeatTimes`, `repeatUntil` and `repeatDelay` to automatically re-schedule jobs when they are completed.

* added Foxx manifest configuration type `password` to mask values in the web interface.

* fixed default values in Foxx manifest configurations sometimes not being used as defaults.

* fixed optional parameters in Foxx manifest configurations sometimes not being cleared correctly.

* Foxx dependencies can now be marked as optional using a slightly more verbose syntax in your manifest file.

* converted Foxx constructors to ES6 classes so you can extend them using class syntax.

* updated aqb to 2.0.

* updated chai to 3.0.

* Use more madvise calls to speed up things when memory is tight, in particular
  at load time but also for random accesses later.

* Overhauled web interface

  The web interface now has a new design.

  The API documentation for ArangoDB has been moved from "Tools" to "Links" in the web interface.

  The "Applications" tab in the web interfaces has been renamed to "Services".


v2.6.12 (2015-12-02)
--------------------

* fixed disappearing of documents for collections transferred via `sync` if the
  the collection was dropped right before synchronization and drop and (re-)create
  collection markers were located in the same WAL file

* added missing lock instruction for primary index in compactor size calculation

* fixed issue #1589

* fixed issue #1583

* Foxx: optional configuration options no longer log validation errors when assigned
  empty values (#1495)


v2.6.11 (2015-11-18)
--------------------

* fixed potentially invalid pointer access in shaper when the currently accessed
  document got re-located by the WAL collector at the very same time


v2.6.10 (2015-11-10)
--------------------

* disable replication appliers when starting in modes `--upgrade`, `--no-server`
  and `--check-upgrade`

* more detailed output in arango-dfdb

* fixed potential deadlock in collection status changing on Windows

* issue #1521: Can't dump/restore with user and password


v2.6.9 (2015-09-29)
-------------------

* added "special" password ARANGODB_DEFAULT_ROOT_PASSWORD. If you pass
  ARANGODB_DEFAULT_ROOT_PASSWORD as password, it will read the password
  from the environment variable ARANGODB_DEFAULT_ROOT_PASSWORD

* fixed failing AQL skiplist, sort and limit combination

  When using a Skiplist index on an attribute (say "a") and then using sort
  and skip on this attribute caused the result to be empty e.g.:

    require("internal").db.test.ensureSkiplist("a");
    require("internal").db._query("FOR x IN test SORT x.a LIMIT 10, 10");

  Was always empty no matter how many documents are stored in test.
  This is now fixed.

v2.6.8 (2015-09-09)
-------------------

* ARM only:

  The ArangoDB packages for ARM require the kernel to allow unaligned memory access.
  How the kernel handles unaligned memory access is configurable at runtime by
  checking and adjusting the contents `/proc/cpu/alignment`.

  In order to operate on ARM, ArangoDB requires the bit 1 to be set. This will
  make the kernel trap and adjust unaligned memory accesses. If this bit is not
  set, the kernel may send a SIGBUS signal to ArangoDB and terminate it.

  To set bit 1 in `/proc/cpu/alignment` use the following command as a privileged
  user (e.g. root):

      echo "2" > /proc/cpu/alignment

  Note that this setting affects all user processes and not just ArangoDB. Setting
  the alignment with the above command will also not make the setting permanent,
  so it will be lost after a restart of the system. In order to make the setting
  permanent, it should be executed during system startup or before starting arangod.

  The ArangoDB start/stop scripts do not adjust the alignment setting, but rely on
  the environment to have the correct alignment setting already. The reason for this
  is that the alignment settings also affect all other user processes (which ArangoDB
  is not aware of) and thus may have side-effects outside of ArangoDB. It is therefore
  more reasonable to have the system administrator carry out the change.


v2.6.7 (2015-08-25)
-------------------

* improved AssocMulti index performance when resizing.

  This makes the edge index perform less I/O when under memory pressure.


v2.6.6 (2015-08-23)
-------------------

* added startup option `--server.additional-threads` to create separate queues
  for slow requests.


v2.6.5 (2015-08-17)
-------------------

* added startup option `--database.throw-collection-not-loaded-error`

  Accessing a not-yet loaded collection will automatically load a collection
  on first access. This flag controls what happens in case an operation
  would need to wait for another thread to finalize loading a collection. If
  set to *true*, then the first operation that accesses an unloaded collection
  will load it. Further threads that try to access the same collection while
  it is still loading immediately fail with an error (1238, *collection not loaded*).
  This is to prevent all server threads from being blocked while waiting on the
  same collection to finish loading. When the first thread has completed loading
  the collection, the collection becomes regularly available, and all operations
  from that point on can be carried out normally, and error 1238 will not be
  thrown anymore for that collection.

  If set to *false*, the first thread that accesses a not-yet loaded collection
  will still load it. Other threads that try to access the collection while
  loading will not fail with error 1238 but instead block until the collection
  is fully loaded. This configuration might lead to all server threads being
  blocked because they are all waiting for the same collection to complete
  loading. Setting the option to *true* will prevent this from happening, but
  requires clients to catch error 1238 and react on it (maybe by scheduling
  a retry for later).

  The default value is *false*.

* fixed busy wait loop in scheduler threads that sometimes consumed 100% CPU while
  waiting for events on connections closed unexpectedly by the client side

* handle attribute `indexBuckets` when restoring collections via arangorestore.
  Previously the `indexBuckets` attribute value from the dump was ignored, and the
   server default value for `indexBuckets` was used when restoring a collection.

* fixed "EscapeValue already set error" crash in V8 actions that might have occurred when
  canceling V8-based operations.


v2.6.4 (2015-08-01)
-------------------

* V8: Upgrade to version 4.1.0.27 - this is intended to be the stable V8 version.

* fixed issue #1424: Arango shell should not processing arrows pushing on keyboard


v2.6.3 (2015-07-21)
-------------------

* issue #1409: Document values with null character truncated


v2.6.2 (2015-07-04)
-------------------

* fixed issue #1383: bindVars for HTTP API doesn't work with empty string

* fixed handling of default values in Foxx manifest configurations

* fixed handling of optional parameters in Foxx manifest configurations

* fixed a reference error being thrown in Foxx queues when a function-based job type is used that is not available and no options object is passed to queue.push


v2.6.1 (2015-06-24)
-------------------

* Add missing swagger files to cmake build. fixes #1368

* fixed documentation errors


v2.6.0 (2015-06-20)
-------------------

* using negative values for `SimpleQuery.skip()` is deprecated.
  This functionality will be removed in future versions of ArangoDB.

* The following simple query functions are now deprecated:

  * collection.near
  * collection.within
  * collection.geo
  * collection.fulltext
  * collection.range
  * collection.closedRange

  This also lead to the following REST API methods being deprecated from now on:

  * PUT /_api/simple/near
  * PUT /_api/simple/within
  * PUT /_api/simple/fulltext
  * PUT /_api/simple/range

  It is recommended to replace calls to these functions or APIs with equivalent AQL queries,
  which are more flexible because they can be combined with other operations:

      FOR doc IN NEAR(@@collection, @latitude, @longitude, @limit)
        RETURN doc

      FOR doc IN WITHIN(@@collection, @latitude, @longitude, @radius, @distanceAttributeName)
        RETURN doc

      FOR doc IN FULLTEXT(@@collection, @attributeName, @queryString, @limit)
        RETURN doc

      FOR doc IN @@collection
        FILTER doc.value >= @left && doc.value < @right
        LIMIT @skip, @limit
        RETURN doc`

  The above simple query functions and REST API methods may be removed in future versions
  of ArangoDB.

* deprecated now-obsolete AQL `SKIPLIST` function

  The function was introduced in older versions of ArangoDB with a less powerful query optimizer to
  retrieve data from a skiplist index using a `LIMIT` clause.

  Since 2.3 the same goal can be achieved by using regular AQL constructs, e.g.

      FOR doc IN collection FILTER doc.value >= @value SORT doc.value DESC LIMIT 1 RETURN doc

* fixed issues when switching the database inside tasks and during shutdown of database cursors

  These features were added during 2.6 alpha stage so the fixes affect devel/2.6-alpha builds only

* issue #1360: improved foxx-manager help

* added `--enable-tcmalloc` configure option.

  When this option is set, arangod and the client tools will be linked against tcmalloc, which replaces
  the system allocator. When the option is set, a tcmalloc library must be present on the system under
  one of the names `libtcmalloc`, `libtcmalloc_minimal` or `libtcmalloc_debug`.

  As this is a configure option, it is supported for manual builds on Linux-like systems only. tcmalloc
  support is currently experimental.

* issue #1353: Windows: HTTP API - incorrect path in errorMessage

* issue #1347: added option `--create-database` for arangorestore.

  Setting this option to `true` will now create the target database if it does not exist. When creating
  the target database, the username and passwords passed to arangorestore will be used to create an
  initial user for the new database.

* issue #1345: advanced debug information for User Functions

* issue #1341: Can't use bindvars in UPSERT

* fixed vulnerability in JWT implementation.

* changed default value of option `--database.ignore-datafile-errors` from `true` to `false`

  If the new default value of `false` is used, then arangod will refuse loading collections that contain
  datafiles with CRC mismatches or other errors. A collection with datafile errors will then become
  unavailable. This prevents follow up errors from happening.

  The only way to access such collection is to use the datafile debugger (arango-dfdb) and try to repair
  or truncate the datafile with it.

  If `--database.ignore-datafile-errors` is set to `true`, then collections will become available
  even if parts of their data cannot be loaded. This helps availability, but may cause (partial) data
  loss and follow up errors.

* added server startup option `--server.session-timeout` for controlling the timeout of user sessions
  in the web interface

* add sessions and cookie authentication for ArangoDB's web interface

  ArangoDB's built-in web interface now uses sessions. Session information ids are stored in cookies,
  so clients using the web interface must accept cookies in order to use it

* web interface: display query execution time in AQL editor

* web interface: renamed AQL query *submit* button to *execute*

* web interface: added query explain feature in AQL editor

* web interface: demo page added. only working if demo data is available, hidden otherwise

* web interface: added support for custom app scripts with optional arguments and results

* web interface: mounted apps that need to be configured are now indicated in the app overview

* web interface: added button for running tests to app details

* web interface: added button for configuring app dependencies to app details

* web interface: upgraded API documentation to use Swagger 2

* INCOMPATIBLE CHANGE

  removed startup option `--log.severity`

  The docs for `--log.severity` mentioned lots of severities (e.g. `exception`, `technical`, `functional`, `development`)
  but only a few severities (e.g. `all`, `human`) were actually used, with `human` being the default and `all` enabling the
  additional logging of requests. So the option pretended to control a lot of things which it actually didn't. Additionally,
  the option `--log.requests-file` was around for a long time already, also controlling request logging.

  Because the `--log.severity` option effectively did not control that much, it was removed. A side effect of removing the
  option is that 2.5 installations which used `--log.severity all` will not log requests after the upgrade to 2.6. This can
  be adjusted by setting the `--log.requests-file` option.

* add backtrace to fatal log events

* added optional `limit` parameter for AQL function `FULLTEXT`

* make fulltext index also index text values contained in direct sub-objects of the indexed
  attribute.

  Previous versions of ArangoDB only indexed the attribute value if it was a string. Sub-attributes
  of the index attribute were ignored when fulltext indexing.

  Now, if the index attribute value is an object, the object's values will each be included in the
  fulltext index if they are strings. If the index attribute value is an array, the array's values
  will each be included in the fulltext index if they are strings.

  For example, with a fulltext index present on the `translations` attribute, the following text
  values will now be indexed:

      var c = db._create("example");
      c.ensureFulltextIndex("translations");
      c.insert({ translations: { en: "fox", de: "Fuchs", fr: "renard", ru: "лиса" } });
      c.insert({ translations: "Fox is the English translation of the German word Fuchs" });
      c.insert({ translations: [ "ArangoDB", "document", "database", "Foxx" ] });

      c.fulltext("translations", "лиса").toArray();       // returns only first document
      c.fulltext("translations", "Fox").toArray();        // returns first and second documents
      c.fulltext("translations", "prefix:Fox").toArray(); // returns all three documents

* added batch document removal and lookup commands:

      collection.lookupByKeys(keys)
      collection.removeByKeys(keys)

  These commands can be used to perform multi-document lookup and removal operations efficiently
  from the ArangoShell. The argument to these operations is an array of document keys.

  Also added HTTP APIs for batch document commands:

  * PUT /_api/simple/lookup-by-keys
  * PUT /_api/simple/remove-by-keys

* properly prefix document address URLs with the current database name for calls to the REST
  API method GET `/_api/document?collection=...` (that method will return partial URLs to all
  documents in the collection).

  Previous versions of ArangoDB returned the URLs starting with `/_api/` but without the current
  database name, e.g. `/_api/document/mycollection/mykey`. Starting with 2.6, the response URLs
  will include the database name as well, e.g. `/_db/_system/_api/document/mycollection/mykey`.

* added dedicated collection export HTTP REST API

  ArangoDB now provides a dedicated collection export API, which can take snapshots of entire
  collections more efficiently than the general-purpose cursor API. The export API is useful
  to transfer the contents of an entire collection to a client application. It provides optional
  filtering on specific attributes.

  The export API is available at endpoint `POST /_api/export?collection=...`. The API has the
  same return value structure as the already established cursor API (`POST /_api/cursor`).

  An introduction to the export API is given in this blog post:
  http://jsteemann.github.io/blog/2015/04/04/more-efficient-data-exports/

* subquery optimizations for AQL queries

  This optimization avoids copying intermediate results into subqueries that are not required
  by the subquery.

  A brief description can be found here:
  http://jsteemann.github.io/blog/2015/05/04/subquery-optimizations/

* return value optimization for AQL queries

  This optimization avoids copying the final query result inside the query's main `ReturnNode`.

  A brief description can be found here:
  http://jsteemann.github.io/blog/2015/05/04/return-value-optimization-for-aql/

* speed up AQL queries containing big `IN` lists for index lookups

  `IN` lists used for index lookups had performance issues in previous versions of ArangoDB.
  These issues have been addressed in 2.6 so using bigger `IN` lists for filtering is much
  faster.

  A brief description can be found here:
  http://jsteemann.github.io/blog/2015/05/07/in-list-improvements/

* allow `@` and `.` characters in document keys, too

  This change also leads to document keys being URL-encoded when returned in HTTP `location`
  response headers.

* added alternative implementation for AQL COLLECT

  The alternative method uses a hash table for grouping and does not require its input elements
  to be sorted. It will be taken into account by the optimizer for `COLLECT` statements that do
  not use an `INTO` clause.

  In case a `COLLECT` statement can use the hash table variant, the optimizer will create an extra
  plan for it at the beginning of the planning phase. In this plan, no extra `SORT` node will be
  added in front of the `COLLECT` because the hash table variant of `COLLECT` does not require
  sorted input. Instead, a `SORT` node will be added after it to sort its output. This `SORT` node
  may be optimized away again in later stages. If the sort order of the result is irrelevant to
  the user, adding an extra `SORT null` after a hash `COLLECT` operation will allow the optimizer to
  remove the sorts altogether.

  In addition to the hash table variant of `COLLECT`, the optimizer will modify the original plan
  to use the regular `COLLECT` implementation. As this implementation requires sorted input, the
  optimizer will insert a `SORT` node in front of the `COLLECT`. This `SORT` node may be optimized
  away in later stages.

  The created plans will then be shipped through the regular optimization pipeline. In the end,
  the optimizer will pick the plan with the lowest estimated total cost as usual. The hash table
  variant does not require an up-front sort of the input, and will thus be preferred over the
  regular `COLLECT` if the optimizer estimates many input elements for the `COLLECT` node and
  cannot use an index to sort them.

  The optimizer can be explicitly told to use the regular *sorted* variant of `COLLECT` by
  suffixing a `COLLECT` statement with `OPTIONS { "method" : "sorted" }`. This will override the
  optimizer guesswork and only produce the *sorted* variant of `COLLECT`.

  A blog post on the new `COLLECT` implementation can be found here:
  http://jsteemann.github.io/blog/2015/04/22/collecting-with-a-hash-table/

* refactored HTTP REST API for cursors

  The HTTP REST API for cursors (`/_api/cursor`) has been refactored to improve its performance
  and use less memory.

  A post showing some of the performance improvements can be found here:
  http://jsteemann.github.io/blog/2015/04/01/improvements-for-the-cursor-api/

* simplified return value syntax for data-modification AQL queries

  ArangoDB 2.4 since version allows to return results from data-modification AQL queries. The
  syntax for this was quite limited and verbose:

      FOR i IN 1..10
        INSERT { value: i } IN test
        LET inserted = NEW
        RETURN inserted

  The `LET inserted = NEW RETURN inserted` was required literally to return the inserted
  documents. No calculations could be made using the inserted documents.

  This is now more flexible. After a data-modification clause (e.g. `INSERT`, `UPDATE`, `REPLACE`,
  `REMOVE`, `UPSERT`) there can follow any number of `LET` calculations. These calculations can
  refer to the pseudo-values `OLD` and `NEW` that are created by the data-modification statements.

  This allows returning projections of inserted or updated documents, e.g.:

      FOR i IN 1..10
        INSERT { value: i } IN test
        RETURN { _key: NEW._key, value: i }

  Still not every construct is allowed after a data-modification clause. For example, no functions
  can be called that may access documents.

  More information can be found here:
  http://jsteemann.github.io/blog/2015/03/27/improvements-for-data-modification-queries/

* added AQL `UPSERT` statement

  This adds an `UPSERT` statement to AQL that is a combination of both `INSERT` and `UPDATE` /
  `REPLACE`. The `UPSERT` will search for a matching document using a user-provided example.
  If no document matches the example, the *insert* part of the `UPSERT` statement will be
  executed. If there is a match, the *update* / *replace* part will be carried out:

      UPSERT { page: 'index.html' }                 /* search example */
        INSERT { page: 'index.html', pageViews: 1 } /* insert part */
        UPDATE { pageViews: OLD.pageViews + 1 }     /* update part */
        IN pageViews

  `UPSERT` can be used with an `UPDATE` or `REPLACE` clause. The `UPDATE` clause will perform
  a partial update of the found document, whereas the `REPLACE` clause will replace the found
  document entirely. The `UPDATE` or `REPLACE` parts can refer to the pseudo-value `OLD`, which
  contains all attributes of the found document.

  `UPSERT` statements can optionally return values. In the following query, the return
  attribute `found` will return the found document before the `UPDATE` was applied. If no
  document was found, `found` will contain a value of `null`. The `updated` result attribute will
  contain the inserted / updated document:

      UPSERT { page: 'index.html' }                 /* search example */
        INSERT { page: 'index.html', pageViews: 1 } /* insert part */
        UPDATE { pageViews: OLD.pageViews + 1 }     /* update part */
        IN pageViews
        RETURN { found: OLD, updated: NEW }

  A more detailed description of `UPSERT` can be found here:
  http://jsteemann.github.io/blog/2015/03/27/preview-of-the-upsert-command/

* adjusted default configuration value for `--server.backlog-size` from 10 to 64.

* issue #1231: bug xor feature in AQL: LENGTH(null) == 4

  This changes the behavior of the AQL `LENGTH` function as follows:

  - if the single argument to `LENGTH()` is `null`, then the result will now be `0`. In previous
    versions of ArangoDB, the result of `LENGTH(null)` was `4`.

  - if the single argument to `LENGTH()` is `true`, then the result will now be `1`. In previous
    versions of ArangoDB, the result of `LENGTH(true)` was `4`.

  - if the single argument to `LENGTH()` is `false`, then the result will now be `0`. In previous
    versions of ArangoDB, the result of `LENGTH(false)` was `5`.

  The results of `LENGTH()` with string, numeric, array object argument values do not change.

* issue #1298: Bulk import if data already exists (#1298)

  This change extends the HTTP REST API for bulk imports as follows:

  When documents are imported and the `_key` attribute is specified for them, the import can be
  used for inserting and updating/replacing documents. Previously, the import could be used for
  inserting new documents only, and re-inserting a document with an existing key would have failed
  with a *unique key constraint violated* error.

  The above behavior is still the default. However, the API now allows controlling the behavior
  in case of a unique key constraint error via the optional URL parameter `onDuplicate`.

  This parameter can have one of the following values:

  - `error`: when a unique key constraint error occurs, do not import or update the document but
    report an error. This is the default.

  - `update`: when a unique key constraint error occurs, try to (partially) update the existing
    document with the data specified in the import. This may still fail if the document would
    violate secondary unique indexes. Only the attributes present in the import data will be
    updated and other attributes already present will be preserved. The number of updated documents
    will be reported in the `updated` attribute of the HTTP API result.

  - `replace`: when a unique key constraint error occurs, try to fully replace the existing
    document with the data specified in the import. This may still fail if the document would
    violate secondary unique indexes. The number of replaced documents will be reported in the
    `updated` attribute of the HTTP API result.

  - `ignore`: when a unique key constraint error occurs, ignore this error. There will be no
    insert, update or replace for the particular document. Ignored documents will be reported
    separately in the `ignored` attribute of the HTTP API result.

  The result of the HTTP import API will now contain the attributes `ignored` and `updated`, which
  contain the number of ignored and updated documents respectively. These attributes will contain a
  value of zero unless the `onDuplicate` URL parameter is set to either `update` or `replace`
  (in this case the `updated` attribute may contain non-zero values) or `ignore` (in this case the
  `ignored` attribute may contain a non-zero value).

  To support the feature, arangoimp also has a new command line option `--on-duplicate` which can
  have one of the values `error`, `update`, `replace`, `ignore`. The default value is `error`.

  A few examples for using arangoimp with the `--on-duplicate` option can be found here:
  http://jsteemann.github.io/blog/2015/04/14/updating-documents-with-arangoimp/

* changed behavior of `db._query()` in the ArangoShell:

  if the command's result is printed in the shell, the first 10 results will be printed. Previously
  only a basic description of the underlying query result cursor was printed. Additionally, if the
  cursor result contains more than 10 results, the cursor is assigned to a global variable `more`,
  which can be used to iterate over the cursor result.

  Example:

      arangosh [_system]> db._query("FOR i IN 1..15 RETURN i")
      [object ArangoQueryCursor, count: 15, hasMore: true]

      [
        1,
        2,
        3,
        4,
        5,
        6,
        7,
        8,
        9,
        10
      ]

      type 'more' to show more documents


      arangosh [_system]> more
      [object ArangoQueryCursor, count: 15, hasMore: false]

      [
        11,
        12,
        13,
        14,
        15
      ]

* Disallow batchSize value 0 in HTTP `POST /_api/cursor`:

  The HTTP REST API `POST /_api/cursor` does not accept a `batchSize` parameter value of
  `0` any longer. A batch size of 0 never made much sense, but previous versions of ArangoDB
  did not check for this value. Now creating a cursor using a `batchSize` value 0 will
  result in an HTTP 400 error response

* REST Server: fix memory leaks when failing to add jobs

* 'EDGES' AQL Function

  The AQL function `EDGES` got a new fifth option parameter.
  Right now only one option is available: 'includeVertices'. This is a boolean parameter
  that allows to modify the result of the `EDGES` function.
  Default is 'includeVertices: false' which does not have any effect.
  'includeVertices: true' modifies the result, such that
  {vertex: <vertexDocument>, edge: <edgeDocument>} is returned.

* INCOMPATIBLE CHANGE:

  The result format of the AQL function `NEIGHBORS` has been changed.
  Before it has returned an array of objects containing 'vertex' and 'edge'.
  Now it will only contain the vertex directly.
  Also an additional option 'includeData' has been added.
  This is used to define if only the 'vertex._id' value should be returned (false, default),
  or if the vertex should be looked up in the collection and the complete JSON should be returned
  (true).
  Using only the id values can lead to significantly improved performance if this is the only information
  required.

  In order to get the old result format prior to ArangoDB 2.6, please use the function EDGES instead.
  Edges allows for a new option 'includeVertices' which, set to true, returns exactly the format of NEIGHBORS.
  Example:

      NEIGHBORS(<vertexCollection>, <edgeCollection>, <vertex>, <direction>, <example>)

  This can now be achieved by:

      EDGES(<edgeCollection>, <vertex>, <direction>, <example>, {includeVertices: true})

  If you are nesting several NEIGHBORS steps you can speed up their performance in the following way:

  Old Example:

  FOR va IN NEIGHBORS(Users, relations, 'Users/123', 'outbound') FOR vc IN NEIGHBORS(Products, relations, va.vertex._id, 'outbound') RETURN vc

  This can now be achieved by:

  FOR va IN NEIGHBORS(Users, relations, 'Users/123', 'outbound') FOR vc IN NEIGHBORS(Products, relations, va, 'outbound', null, {includeData: true}) RETURN vc
                                                                                                          ^^^^                  ^^^^^^^^^^^^^^^^^^^
                                                                                                  Use intermediate directly     include Data for final

* INCOMPATIBLE CHANGE:

  The AQL function `GRAPH_NEIGHBORS` now provides an additional option `includeData`.
  This option allows controlling whether the function should return the complete vertices
  or just their IDs. Returning only the IDs instead of the full vertices can lead to
  improved performance .

  If provided, `includeData` is set to `true`, all vertices in the result will be returned
  with all their attributes. The default value of `includeData` is `false`.
  This makes the default function results incompatible with previous versions of ArangoDB.

  To get the old result style in ArangoDB 2.6, please set the options as follows in calls
  to `GRAPH_NEIGHBORS`:

      GRAPH_NEIGHBORS(<graph>, <vertex>, { includeData: true })

* INCOMPATIBLE CHANGE:

  The AQL function `GRAPH_COMMON_NEIGHBORS` now provides an additional option `includeData`.
  This option allows controlling whether the function should return the complete vertices
  or just their IDs. Returning only the IDs instead of the full vertices can lead to
  improved performance .

  If provided, `includeData` is set to `true`, all vertices in the result will be returned
  with all their attributes. The default value of `includeData` is `false`.
  This makes the default function results incompatible with previous versions of ArangoDB.

  To get the old result style in ArangoDB 2.6, please set the options as follows in calls
  to `GRAPH_COMMON_NEIGHBORS`:

      GRAPH_COMMON_NEIGHBORS(<graph>, <vertexExamples1>, <vertexExamples2>, { includeData: true }, { includeData: true })

* INCOMPATIBLE CHANGE:

  The AQL function `GRAPH_SHORTEST_PATH` now provides an additional option `includeData`.
  This option allows controlling whether the function should return the complete vertices
  and edges or just their IDs. Returning only the IDs instead of full vertices and edges
  can lead to improved performance .

  If provided, `includeData` is set to `true`, all vertices and edges in the result will
  be returned with all their attributes. There is also an optional parameter `includePath` of
  type object.
  It has two optional sub-attributes `vertices` and `edges`, both of type boolean.
  Both can be set individually and the result will include all vertices on the path if
  `includePath.vertices == true` and all edges if `includePath.edges == true` respectively.

  The default value of `includeData` is `false`, and paths are now excluded by default.
  This makes the default function results incompatible with previous versions of ArangoDB.

  To get the old result style in ArangoDB 2.6, please set the options as follows in calls
  to `GRAPH_SHORTEST_PATH`:

      GRAPH_SHORTEST_PATH(<graph>, <source>, <target>, { includeData: true, includePath: { edges: true, vertices: true } })

  The attributes `startVertex` and `vertex` that were present in the results of `GRAPH_SHORTEST_PATH`
  in previous versions of ArangoDB will not be produced in 2.6. To calculate these attributes in 2.6,
  please extract the first and last elements from the `vertices` result attribute.

* INCOMPATIBLE CHANGE:

  The AQL function `GRAPH_DISTANCE_TO` will now return only the id the destination vertex
  in the `vertex` attribute, and not the full vertex data with all vertex attributes.

* INCOMPATIBLE CHANGE:

  All graph measurements functions in JavaScript module `general-graph` that calculated a
  single figure previously returned an array containing just the figure. Now these functions
  will return the figure directly and not put it inside an array.

  The affected functions are:

  * `graph._absoluteEccentricity`
  * `graph._eccentricity`
  * `graph._absoluteCloseness`
  * `graph._closeness`
  * `graph._absoluteBetweenness`
  * `graph._betweenness`
  * `graph._radius`
  * `graph._diameter`

* Create the `_graphs` collection in new databases with `waitForSync` attribute set to `false`

  The previous `waitForSync` value was `true`, so default the behavior when creating and dropping
  graphs via the HTTP REST API changes as follows if the new settings are in effect:

  * `POST /_api/graph` by default returns `HTTP 202` instead of `HTTP 201`
  * `DELETE /_api/graph/graph-name` by default returns `HTTP 202` instead of `HTTP 201`

  If the `_graphs` collection still has its `waitForSync` value set to `true`, then the HTTP status
  code will not change.

* Upgraded ICU to version 54; this increases performance in many places.
  based on https://code.google.com/p/chromium/issues/detail?id=428145

* added support for HTTP push aka chunked encoding

* issue #1051: add info whether server is running in service or user mode?

  This will add a "mode" attribute to the result of the result of HTTP GET `/_api/version?details=true`

  "mode" can have the following values:

  - `standalone`: server was started manually (e.g. on command-line)
  - `service`: service is running as Windows service, in daemon mode or under the supervisor

* improve system error messages in Windows port

* increased default value of `--server.request-timeout` from 300 to 1200 seconds for client tools
  (arangosh, arangoimp, arangodump, arangorestore)

* increased default value of `--server.connect-timeout` from 3 to 5 seconds for client tools
  (arangosh, arangoimp, arangodump, arangorestore)

* added startup option `--server.foxx-queues-poll-interval`

  This startup option controls the frequency with which the Foxx queues manager is checking
  the queue (or queues) for jobs to be executed.

  The default value is `1` second. Lowering this value will result in the queue manager waking
  up and checking the queues more frequently, which may increase CPU usage of the server.
  When not using Foxx queues, this value can be raised to save some CPU time.

* added startup option `--server.foxx-queues`

  This startup option controls whether the Foxx queue manager will check queue and job entries.
  Disabling this option can reduce server load but will prevent jobs added to Foxx queues from
  being processed at all.

  The default value is `true`, enabling the Foxx queues feature.

* make Foxx queues really database-specific.

  Foxx queues were and are stored in a database-specific collection `_queues`. However, a global
  cache variable for the queues led to the queue names being treated database-independently, which
  was wrong.

  Since 2.6, Foxx queues names are truly database-specific, so the same queue name can be used in
  two different databases for two different queues. Until then, it is advisable to think of queues
  as already being database-specific, and using the database name as a queue name prefix to be
  avoid name conflicts, e.g.:

      var queueName = "myQueue";
      var Foxx = require("org/arangodb/foxx");
      Foxx.queues.create(db._name() + ":" + queueName);

* added support for Foxx queue job types defined as app scripts.

  The old job types introduced in 2.4 are still supported but are known to cause issues in 2.5
  and later when the server is restarted or the job types are not defined in every thread.

  The new job types avoid this issue by storing an explicit mount path and script name rather
  than an assuming the job type is defined globally. It is strongly recommended to convert your
  job types to the new script-based system.

* renamed Foxx sessions option "sessionStorageApp" to "sessionStorage". The option now also accepts session storages directly.

* Added the following JavaScript methods for file access:
  * fs.copyFile() to copy single files
  * fs.copyRecursive() to copy directory trees
  * fs.chmod() to set the file permissions (non-Windows only)

* Added process.env for accessing the process environment from JavaScript code

* Cluster: kickstarter shutdown routines will more precisely follow the shutdown of its nodes.

* Cluster: don't delete agency connection objects that are currently in use.

* Cluster: improve passing along of HTTP errors

* fixed issue #1247: debian init script problems

* multi-threaded index creation on collection load

  When a collection contains more than one secondary index, they can be built in memory in
  parallel when the collection is loaded. How many threads are used for parallel index creation
  is determined by the new configuration parameter `--database.index-threads`. If this is set
  to 0, indexes are built by the opening thread only and sequentially. This is equivalent to
  the behavior in 2.5 and before.

* speed up building up primary index when loading collections

* added `count` attribute to `parameters.json` files of collections. This attribute indicates
  the number of live documents in the collection on unload. It is read when the collection is
  (re)loaded to determine the initial size for the collection's primary index

* removed remainders of MRuby integration, removed arangoirb

* simplified `controllers` property in Foxx manifests. You can now specify a filename directly
  if you only want to use a single file mounted at the base URL of your Foxx app.

* simplified `exports` property in Foxx manifests. You can now specify a filename directly if
  you only want to export variables from a single file in your Foxx app.

* added support for node.js-style exports in Foxx exports. Your Foxx exports file can now export
  arbitrary values using the `module.exports` property instead of adding properties to the
  `exports` object.

* added `scripts` property to Foxx manifests. You should now specify the `setup` and `teardown`
  files as properties of the `scripts` object in your manifests and can define custom,
  app-specific scripts that can be executed from the web interface or the CLI.

* added `tests` property to Foxx manifests. You can now define test cases using the `mocha`
  framework which can then be executed inside ArangoDB.

* updated `joi` package to 6.0.8.

* added `extendible` package.

* added Foxx model lifecycle events to repositories. See #1257.

* speed up resizing of edge index.

* allow to split an edge index into buckets which are resized individually.
  This is controlled by the `indexBuckets` attribute in the `properties`
  of the collection.

* fix a cluster deadlock bug in larger clusters by marking a thread waiting
  for a lock on a DBserver as blocked


v2.5.7 (2015-08-02)
-------------------

* V8: Upgrade to version 4.1.0.27 - this is intended to be the stable V8 version.


v2.5.6 (2015-07-21)
-------------------

* alter Windows build infrastructure so we can properly store pdb files.

* potentially fixed issue #1313: Wrong metric calculation at dashboard

  Escape whitespace in process name when scanning /proc/pid/stats

  This fixes statistics values read from that file

* Fixed variable naming in AQL `COLLECT INTO` results in case the COLLECT is placed
  in a subquery which itself is followed by other constructs that require variables


v2.5.5 (2015-05-29)
-------------------

* fixed vulnerability in JWT implementation.

* fixed format string for reading /proc/pid/stat

* take into account barriers used in different V8 contexts


v2.5.4 (2015-05-14)
-------------------

* added startup option `--log.performance`: specifying this option at startup will log
  performance-related info messages, mainly timings via the regular logging mechanisms

* cluster fixes

* fix for recursive copy under Windows


v2.5.3 (2015-04-29)
-------------------

* Fix fs.move to work across filesystem borders; Fixes Foxx app installation problems;
  issue #1292.

* Fix Foxx app install when installed on a different drive on Windows

* issue #1322: strange AQL result

* issue #1318: Inconsistent db._create() syntax

* issue #1315: queries to a collection fail with an empty response if the
  collection contains specific JSON data

* issue #1300: Make arangodump not fail if target directory exists but is empty

* allow specifying higher values than SOMAXCONN for `--server.backlog-size`

  Previously, arangod would not start when a `--server.backlog-size` value was
  specified that was higher than the platform's SOMAXCONN header value.

  Now, arangod will use the user-provided value for `--server.backlog-size` and
  pass it to the listen system call even if the value is higher than SOMAXCONN.
  If the user-provided value is higher than SOMAXCONN, arangod will log a warning
  on startup.

* Fixed a cluster deadlock bug. Mark a thread that is in a RemoteBlock as
  blocked to allow for additional dispatcher threads to be started.

* Fix locking in cluster by using another ReadWriteLock class for collections.

* Add a second DispatcherQueue for AQL in the cluster. This fixes a
  cluster-AQL thread explosion bug.


v2.5.2 (2015-04-11)
-------------------

* modules stored in _modules are automatically flushed when changed

* added missing query-id parameter in documentation of HTTP DELETE `/_api/query` endpoint

* added iterator for edge index in AQL queries

  this change may lead to less edges being read when used together with a LIMIT clause

* make graph viewer in web interface issue less expensive queries for determining
  a random vertex from the graph, and for determining vertex attributes

* issue #1285: syntax error, unexpected $undefined near '@_to RETURN obj

  this allows AQL bind parameter names to also start with underscores

* moved /_api/query to C++

* issue #1289: Foxx models created from database documents expose an internal method

* added `Foxx.Repository#exists`

* parallelize initialization of V8 context in multiple threads

* fixed a possible crash when the debug-level was TRACE

* cluster: do not initialize statistics collection on each
  coordinator, this fixes a race condition at startup

* cluster: fix a startup race w.r.t. the _configuration collection

* search for db:// JavaScript modules only after all local files have been
  considered, this speeds up the require command in a cluster considerably

* general cluster speedup in certain areas


v2.5.1 (2015-03-19)
-------------------

* fixed bug that caused undefined behavior when an AQL query was killed inside
  a calculation block

* fixed memleaks in AQL query cleanup in case out-of-memory errors are thrown

* by default, Debian and RedHat packages are built with debug symbols

* added option `--database.ignore-logfile-errors`

  This option controls how collection datafiles with a CRC mismatch are treated.

  If set to `false`, CRC mismatch errors in collection datafiles will lead
  to a collection not being loaded at all. If a collection needs to be loaded
  during WAL recovery, the WAL recovery will also abort (if not forced with
  `--wal.ignore-recovery-errors true`). Setting this flag to `false` protects
  users from unintentionally using a collection with corrupted datafiles, from
  which only a subset of the original data can be recovered.

  If set to `true`, CRC mismatch errors in collection datafiles will lead to
  the datafile being partially loaded. All data up to until the mismatch will
  be loaded. This will enable users to continue with collection datafiles
  that are corrupted, but will result in only a partial load of the data.
  The WAL recovery will still abort when encountering a collection with a
  corrupted datafile, at least if `--wal.ignore-recovery-errors` is not set to
  `true`.

  The default value is *true*, so for collections with corrupted datafiles
  there might be partial data loads once the WAL recovery has finished. If
  the WAL recovery will need to load a collection with a corrupted datafile,
  it will still stop when using the default values.

* INCOMPATIBLE CHANGE:

  make the arangod server refuse to start if during startup it finds a non-readable
  `parameter.json` file for a database or a collection.

  Stopping the startup process in this case requires manual intervention (fixing
  the unreadable files), but prevents follow-up errors due to ignored databases or
  collections from happening.

* datafiles and `parameter.json` files written by arangod are now created with read and write
  privileges for the arangod process user, and with read and write privileges for the arangod
  process group.

  Previously, these files were created with user read and write permissions only.

* INCOMPATIBLE CHANGE:

  abort WAL recovery if one of the collection's datafiles cannot be opened

* INCOMPATIBLE CHANGE:

  never try to raise the privileges after dropping them, this can lead to a race condition while
  running the recovery

  If you require to run ArangoDB on a port lower than 1024, you must run ArangoDB as root.

* fixed inefficiencies in `remove` methods of general-graph module

* added option `--database.slow-query-threshold` for controlling the default AQL slow query
  threshold value on server start

* add system error strings for Windows on many places

* rework service startup so we announce 'RUNNING' only when we're finished starting.

* use the Windows eventlog for FATAL and ERROR - log messages

* fix service handling in NSIS Windows installer, specify human readable name

* add the ICU_DATA environment variable to the fatal error messages

* fixed issue #1265: arangod crashed with SIGSEGV

* fixed issue #1241: Wildcards in examples


v2.5.0 (2015-03-09)
-------------------

* installer fixes for Windows

* fix for downloading Foxx

* fixed issue #1258: http pipelining not working?


v2.5.0-beta4 (2015-03-05)
-------------------------

* fixed issue #1247: debian init script problems


v2.5.0-beta3 (2015-02-27)
-------------------------

* fix Windows install path calculation in arango

* fix Windows logging of long strings

* fix possible undefinedness of const strings in Windows


v2.5.0-beta2 (2015-02-23)
-------------------------

* fixed issue #1256: agency binary not found #1256

* fixed issue #1230: API: document/col-name/_key and cursor return different floats

* front-end: dashboard tries not to (re)load statistics if user has no access

* V8: Upgrade to version 3.31.74.1

* etcd: Upgrade to version 2.0 - This requires go 1.3 to compile at least.

* refuse to startup if ICU wasn't initialized, this will i.e. prevent errors from being printed,
  and libraries from being loaded.

* front-end: unwanted removal of index table header after creating new index

* fixed issue #1248: chrome: applications filtering not working

* fixed issue #1198: queries remain in aql editor (front-end) if you navigate through different tabs

* Simplify usage of Foxx

  Thanks to our user feedback we learned that Foxx is a powerful, yet rather complicated concept.
  With this release we tried to make it less complicated while keeping all its strength.
  That includes a rewrite of the documentation as well as some code changes as listed below:

  * Moved Foxx applications to a different folder.

    The naming convention now is: <app-path>/_db/<dbname>/<mountpoint>/APP
    Before it was: <app-path>/databases/<dbname>/<appname>:<appversion>
    This caused some trouble as apps where cached based on name and version and updates did not apply.
    Hence the path on filesystem and the app's access URL had no relation to one another.
    Now the path on filesystem is identical to the URL (except for slashes and the appended APP)

  * Rewrite of Foxx routing

    The routing of Foxx has been exposed to major internal changes we adjusted because of user feedback.
    This allows us to set the development mode per mount point without having to change paths and hold
    apps at separate locations.

  * Foxx Development mode

    The development mode used until 2.4 is gone. It has been replaced by a much more mature version.
    This includes the deprecation of the javascript.dev-app-path parameter, which is useless since 2.5.
    Instead of having two separate app directories for production and development, apps now reside in
    one place, which is used for production as well as for development.
    Apps can still be put into development mode, changing their behavior compared to production mode.
    Development mode apps are still reread from disk at every request, and still they ship more debug
    output.

    This change has also made the startup options `--javascript.frontend-development-mode` and
    `--javascript.dev-app-path` obsolete. The former option will not have any effect when set, and the
    latter option is only read and used during the upgrade to 2.5 and does not have any effects later.

  * Foxx install process

    Installing Foxx apps has been a two step process: import them into ArangoDB and mount them at a
    specific mount point. These operations have been joined together. You can install an app at one
    mount point, that's it. No fetch, mount, unmount, purge cycle anymore. The commands have been
    simplified to just:

    * install: get your Foxx app up and running
    * uninstall: shut it down and erase it from disk

  * Foxx error output

    Until 2.4 the errors produced by Foxx were not optimal. Often, the error message was just
    `unable to parse manifest` and contained only an internal stack trace.
    In 2.5 we made major improvements there, including a much more fine-grained error output that
    helps you debug your Foxx apps. The error message printed is now much closer to its source and
    should help you track it down.

    Also we added the default handlers for unhandled errors in Foxx apps:

    * You will get a nice internal error page whenever your Foxx app is called but was not installed
      due to any error
    * You will get a proper error message when having an uncaught error appears in any app route

    In production mode the messages above will NOT contain any information about your Foxx internals
    and are safe to be exposed to third party users.
    In development mode the messages above will contain the stacktrace (if available), making it easier for
    your in-house devs to track down errors in the application.

* added `console` object to Foxx apps. All Foxx apps now have a console object implementing
  the familiar Console API in their global scope, which can be used to log diagnostic
  messages to the database.

* added `org/arangodb/request` module, which provides a simple API for making HTTP requests
  to external services.

* added optimizer rule `propagate-constant-attributes`

  This rule will look inside `FILTER` conditions for constant value equality comparisons,
  and insert the constant values in other places in `FILTER`s. For example, the rule will
  insert `42` instead of `i.value` in the second `FILTER` of the following query:

      FOR i IN c1 FOR j IN c2 FILTER i.value == 42 FILTER j.value == i.value RETURN 1

* added `filtered` value to AQL query execution statistics

  This value indicates how many documents were filtered by `FilterNode`s in the AQL query.
  Note that `IndexRangeNode`s can also filter documents by selecting only the required ranges
  from the index. The `filtered` value will not include the work done by `IndexRangeNode`s,
  but only the work performed by `FilterNode`s.

* added support for sparse hash and skiplist indexes

  Hash and skiplist indexes can optionally be made sparse. Sparse indexes exclude documents
  in which at least one of the index attributes is either not set or has a value of `null`.

  As such documents are excluded from sparse indexes, they may contain fewer documents than
  their non-sparse counterparts. This enables faster indexing and can lead to reduced memory
  usage in case the indexed attribute does occur only in some, but not all documents of the
  collection. Sparse indexes will also reduce the number of collisions in non-unique hash
  indexes in case non-existing or optional attributes are indexed.

  In order to create a sparse index, an object with the attribute `sparse` can be added to
  the index creation commands:

      db.collection.ensureHashIndex(attributeName, { sparse: true });
      db.collection.ensureHashIndex(attributeName1, attributeName2, { sparse: true });
      db.collection.ensureUniqueConstraint(attributeName, { sparse: true });
      db.collection.ensureUniqueConstraint(attributeName1, attributeName2, { sparse: true });

      db.collection.ensureSkiplist(attributeName, { sparse: true });
      db.collection.ensureSkiplist(attributeName1, attributeName2, { sparse: true });
      db.collection.ensureUniqueSkiplist(attributeName, { sparse: true });
      db.collection.ensureUniqueSkiplist(attributeName1, attributeName2, { sparse: true });

  Note that in place of the above specialized index creation commands, it is recommended to use
  the more general index creation command `ensureIndex`:

  ```js
  db.collection.ensureIndex({ type: "hash", sparse: true, unique: true, fields: [ attributeName ] });
  db.collection.ensureIndex({ type: "skiplist", sparse: false, unique: false, fields: [ "a", "b" ] });
  ```

  When not explicitly set, the `sparse` attribute defaults to `false` for new indexes.

  This causes a change in behavior when creating a unique hash index without specifying the
  sparse flag: in 2.4, unique hash indexes were implicitly sparse, always excluding `null` values.
  There was no option to control this behavior, and sparsity was neither supported for non-unique
  hash indexes nor skiplists in 2.4. This implicit sparsity of unique hash indexes was considered
  an inconsistency, and therefore the behavior was cleaned up in 2.5. As of 2.5, indexes will
  only be created sparse if sparsity is explicitly requested. Existing unique hash indexes from 2.4
  or before will automatically be migrated so they are still sparse after the upgrade to 2.5.

  Geo indexes are implicitly sparse, meaning documents without the indexed location attribute or
  containing invalid location coordinate values will be excluded from the index automatically. This
  is also a change when compared to pre-2.5 behavior, when documents with missing or invalid
  coordinate values may have caused errors on insertion when the geo index' `unique` flag was set
  and its `ignoreNull` flag was not.

  This was confusing and has been rectified in 2.5. The method `ensureGeoConstaint()` now does the
  same as `ensureGeoIndex()`. Furthermore, the attributes `constraint`, `unique`, `ignoreNull` and
  `sparse` flags are now completely ignored when creating geo indexes.

  The same is true for fulltext indexes. There is no need to specify non-uniqueness or sparsity for
  geo or fulltext indexes. They will always be non-unique and sparse.

  As sparse indexes may exclude some documents, they cannot be used for every type of query.
  Sparse hash indexes cannot be used to find documents for which at least one of the indexed
  attributes has a value of `null`. For example, the following AQL query cannot use a sparse
  index, even if one was created on attribute `attr`:

      FOR doc In collection
        FILTER doc.attr == null
        RETURN doc

  If the lookup value is non-constant, a sparse index may or may not be used, depending on
  the other types of conditions in the query. If the optimizer can safely determine that
  the lookup value cannot be `null`, a sparse index may be used. When uncertain, the optimizer
  will not make use of a sparse index in a query in order to produce correct results.

  For example, the following queries cannot use a sparse index on `attr` because the optimizer
  will not know beforehand whether the comparison values for `doc.attr` will include `null`:

      FOR doc In collection
        FILTER doc.attr == SOME_FUNCTION(...)
        RETURN doc

      FOR other IN otherCollection
        FOR doc In collection
          FILTER doc.attr == other.attr
          RETURN doc

  Sparse skiplist indexes can be used for sorting if the optimizer can safely detect that the
  index range does not include `null` for any of the index attributes.

* inspection of AQL data-modification queries will now detect if the data-modification part
  of the query can run in lockstep with the data retrieval part of the query, or if the data
  retrieval part must be executed before the data modification can start.

  Executing the two in lockstep allows using much smaller buffers for intermediate results
  and starts the actual data-modification operations much earlier than if the two phases
  were executed separately.

* Allow dynamic attribute names in AQL object literals

  This allows using arbitrary expressions to construct attribute names in object
  literals specified in AQL queries. To disambiguate expressions and other unquoted
  attribute names, dynamic attribute names need to be enclosed in brackets (`[` and `]`).
  Example:

      FOR i IN 1..100
        RETURN { [ CONCAT('value-of-', i) ] : i }

* make AQL optimizer rule "use-index-for-sort" remove sort also in case a non-sorted
  index (e.g. a hash index) is used for only equality lookups and all sort attributes
  are covered by the index.

  Example that does not require an extra sort (needs hash index on `value`):

      FOR doc IN collection FILTER doc.value == 1 SORT doc.value RETURN doc

  Another example that does not require an extra sort (with hash index on `value1`, `value2`):

      FOR doc IN collection FILTER doc.value1 == 1 && doc.value2 == 2 SORT doc.value1, doc.value2 RETURN doc

* make AQL optimizer rule "use-index-for-sort" remove sort also in case the sort criteria
  excludes the left-most index attributes, but the left-most index attributes are used
  by the index for equality-only lookups.

  Example that can use the index for sorting (needs skiplist index on `value1`, `value2`):

      FOR doc IN collection FILTER doc.value1 == 1 SORT doc.value2 RETURN doc

* added selectivity estimates for primary index, edge index, and hash index

  The selectivity estimates are returned by the `GET /_api/index` REST API method
  in a sub-attribute `selectivityEstimate` for each index that supports it. This
  attribute will be omitted for indexes that do not provide selectivity estimates.
  If provided, the selectivity estimate will be a numeric value between 0 and 1.

  Selectivity estimates will also be reported in the result of `collection.getIndexes()`
  for all indexes that support this. If no selectivity estimate can be determined for
  an index, the attribute `selectivityEstimate` will be omitted here, too.

  The web interface also shows selectivity estimates for each index that supports this.

  Currently the following index types can provide selectivity estimates:
  - primary index
  - edge index
  - hash index (unique and non-unique)

  No selectivity estimates will be provided when running in cluster mode.

* fixed issue #1226: arangod log issues

* added additional logger if arangod is started in foreground mode on a tty

* added AQL optimizer rule "move-calculations-down"

* use exclusive native SRWLocks on Windows instead of native mutexes

* added AQL functions `MD5`, `SHA1`, and `RANDOM_TOKEN`.

* reduced number of string allocations when parsing certain AQL queries

  parsing numbers (integers or doubles) does not require a string allocation
  per number anymore

* RequestContext#bodyParam now accepts arbitrary joi schemas and rejects invalid (but well-formed) request bodies.

* enforce that AQL user functions are wrapped inside JavaScript function () declarations

  AQL user functions were always expected to be wrapped inside a JavaScript function, but previously
  this was not enforced when registering a user function. Enforcing the AQL user functions to be contained
  inside functions prevents functions from doing some unexpected things that may have led to undefined
  behavior.

* Windows service uninstalling: only remove service if it points to the currently running binary,
  or --force was specified.

* Windows (debug only): print stacktraces on crash and run minidump

* Windows (cygwin): if you run arangosh in a cygwin shell or via ssh we will detect this and use
  the appropriate output functions.

* Windows: improve process management

* fix IPv6 reverse ip lookups - so far we only did IPv4 addresses.

* improve join documentation, add outer join example

* run jslint for unit tests too, to prevent "memory leaks" by global js objects with native code.

* fix error logging for exceptions - we wouldn't log the exception message itself so far.

* improve error reporting in the http client (Windows & *nix)

* improve error reports in cluster

* Standard errors can now contain custom messages.


v2.4.7 (XXXX-XX-XX)
-------------------

* fixed issue #1282: Geo WITHIN_RECTANGLE for nested lat/lng


v2.4.6 (2015-03-18)
-------------------

* added option `--database.ignore-logfile-errors`

  This option controls how collection datafiles with a CRC mismatch are treated.

  If set to `false`, CRC mismatch errors in collection datafiles will lead
  to a collection not being loaded at all. If a collection needs to be loaded
  during WAL recovery, the WAL recovery will also abort (if not forced with
  `--wal.ignore-recovery-errors true`). Setting this flag to `false` protects
  users from unintentionally using a collection with corrupted datafiles, from
  which only a subset of the original data can be recovered.

  If set to `true`, CRC mismatch errors in collection datafiles will lead to
  the datafile being partially loaded. All data up to until the mismatch will
  be loaded. This will enable users to continue with a collection datafiles
  that are corrupted, but will result in only a partial load of the data.
  The WAL recovery will still abort when encountering a collection with a
  corrupted datafile, at least if `--wal.ignore-recovery-errors` is not set to
  `true`.

  The default value is *true*, so for collections with corrupted datafiles
  there might be partial data loads once the WAL recovery has finished. If
  the WAL recovery will need to load a collection with a corrupted datafile,
  it will still stop when using the default values.

* INCOMPATIBLE CHANGE:

  make the arangod server refuse to start if during startup it finds a non-readable
  `parameter.json` file for a database or a collection.

  Stopping the startup process in this case requires manual intervention (fixing
  the unreadable files), but prevents follow-up errors due to ignored databases or
  collections from happening.

* datafiles and `parameter.json` files written by arangod are now created with read and write
  privileges for the arangod process user, and with read and write privileges for the arangod
  process group.

  Previously, these files were created with user read and write permissions only.

* INCOMPATIBLE CHANGE:

  abort WAL recovery if one of the collection's datafiles cannot be opened

* INCOMPATIBLE CHANGE:

  never try to raise the privileges after dropping them, this can lead to a race condition while
  running the recovery

  If you require to run ArangoDB on a port lower than 1024, you must run ArangoDB as root.

* fixed inefficiencies in `remove` methods of general-graph module

* added option `--database.slow-query-threshold` for controlling the default AQL slow query
  threshold value on server start


v2.4.5 (2015-03-16)
-------------------

* added elapsed time to HTTP request logging output (`--log.requests-file`)

* added AQL current and slow query tracking, killing of AQL queries

  This change enables retrieving the list of currently running AQL queries inside the selected database.
  AQL queries with an execution time beyond a certain threshold can be moved to a "slow query" facility
  and retrieved from there. Queries can also be killed by specifying the query id.

  This change adds the following HTTP REST APIs:

  - `GET /_api/query/current`: for retrieving the list of currently running queries
  - `GET /_api/query/slow`: for retrieving the list of slow queries
  - `DELETE /_api/query/slow`: for clearing the list of slow queries
  - `GET /_api/query/properties`: for retrieving the properties for query tracking
  - `PUT /_api/query/properties`: for adjusting the properties for query tracking
  - `DELETE /_api/query/<id>`: for killing an AQL query

  The following JavaScript APIs have been added:

  - require("org/arangodb/aql/queries").current();
  - require("org/arangodb/aql/queries").slow();
  - require("org/arangodb/aql/queries").clearSlow();
  - require("org/arangodb/aql/queries").properties();
  - require("org/arangodb/aql/queries").kill();

* fixed issue #1265: arangod crashed with SIGSEGV

* fixed issue #1241: Wildcards in examples

* fixed comment parsing in Foxx controllers


v2.4.4 (2015-02-24)
-------------------

* fixed the generation template for foxx apps. It now does not create deprecated functions anymore

* add custom visitor functionality for `GRAPH_NEIGHBORS` function, too

* increased default value of traversal option *maxIterations* to 100 times of its previous
  default value


v2.4.3 (2015-02-06)
-------------------

* fix multi-threading with openssl when running under Windows

* fix timeout on socket operations when running under Windows

* Fixed an error in Foxx routing which caused some apps that worked in 2.4.1 to fail with status 500: `undefined is not a function` errors in 2.4.2
  This error was occurring due to seldom internal rerouting introduced by the malformed application handler.


v2.4.2 (2015-01-30)
-------------------

* added custom visitor functionality for AQL traversals

  This allows more complex result processing in traversals triggered by AQL. A few examples
  are shown in [this article](http://jsteemann.github.io/blog/2015/01/28/using-custom-visitors-in-aql-graph-traversals/).

* improved number of results estimated for nodes of type EnumerateListNode and SubqueryNode
  in AQL explain output

* added AQL explain helper to explain arbitrary AQL queries

  The helper function prints the query execution plan and the indexes to be used in the
  query. It can be invoked from the ArangoShell or the web interface as follows:

      require("org/arangodb/aql/explainer").explain(query);

* enable use of indexes for certain AQL conditions with non-equality predicates, in
  case the condition(s) also refer to indexed attributes

  The following queries will now be able to use indexes:

      FILTER a.indexed == ... && a.indexed != ...
      FILTER a.indexed == ... && a.nonIndexed != ...
      FILTER a.indexed == ... && ! (a.indexed == ...)
      FILTER a.indexed == ... && ! (a.nonIndexed == ...)
      FILTER a.indexed == ... && ! (a.indexed != ...)
      FILTER a.indexed == ... && ! (a.nonIndexed != ...)
      FILTER (a.indexed == ... && a.nonIndexed == ...) || (a.indexed == ... && a.nonIndexed == ...)
      FILTER (a.indexed == ... && a.nonIndexed != ...) || (a.indexed == ... && a.nonIndexed != ...)

* Fixed spuriously occurring "collection not found" errors when running queries on local
  collections on a cluster DB server

* Fixed upload of Foxx applications to the server for apps exceeding approx. 1 MB zipped.

* Malformed Foxx applications will now return a more useful error when any route is requested.

  In Production a Foxx app mounted on /app will display an html page on /app/* stating a 503 Service temporarily not available.
  It will not state any information about your Application.
  Before it was a 404 Not Found without any information and not distinguishable from a correct not found on your route.

  In Development Mode the html page also contains information about the error occurred.

* Unhandled errors thrown in Foxx routes are now handled by the Foxx framework itself.

  In Production the route will return a status 500 with a body {error: "Error statement"}.
  In Development the route will return a status 500 with a body {error: "Error statement", stack: "..."}

  Before, it was status 500 with a plain text stack including ArangoDB internal routing information.

* The Applications tab in web interface will now request development apps more often.
  So if you have a fixed a syntax error in your app it should always be visible after reload.


v2.4.1 (2015-01-19)
-------------------

* improved WAL recovery output

* fixed certain OR optimizations in AQL optimizer

* better diagnostics for arangoimp

* fixed invalid result of HTTP REST API method `/_admin/foxx/rescan`

* fixed possible segmentation fault when passing a Buffer object into a V8 function
  as a parameter

* updated AQB module to 1.8.0.


v2.4.0 (2015-01-13)
-------------------

* updated AQB module to 1.7.0.

* fixed V8 integration-related crashes

* make `fs.move(src, dest)` also fail when both `src` and `dest` are
  existing directories. This ensures the same behavior of the move operation
  on different platforms.

* fixed AQL insert operation for multi-shard collections in cluster

* added optional return value for AQL data-modification queries.
  This allows returning the documents inserted, removed or updated with the query, e.g.

      FOR doc IN docs REMOVE doc._key IN docs LET removed = OLD RETURN removed
      FOR doc IN docs INSERT { } IN docs LET inserted = NEW RETURN inserted
      FOR doc IN docs UPDATE doc._key WITH { } IN docs LET previous = OLD RETURN previous
      FOR doc IN docs UPDATE doc._key WITH { } IN docs LET updated = NEW RETURN updated

  The variables `OLD` and `NEW` are automatically available when a `REMOVE`, `INSERT`,
  `UPDATE` or `REPLACE` statement is immediately followed by a `LET` statement.
  Note that the `LET` and `RETURN` statements in data-modification queries are not as
  flexible as the general versions of `LET` and `RETURN`. When returning documents from
  data-modification operations, only a single variable can be assigned using `LET`, and
  the assignment can only be either `OLD` or `NEW`, but not an arbitrary expression. The
  `RETURN` statement also allows using the just-created variable only, and no arbitrary
  expressions.


v2.4.0-beta1 (2014-12-26)
--------------------------

* fixed superstates in FoxxGenerator

* fixed issue #1065: Aardvark: added creation of documents and edges with _key property

* fixed issue #1198: Aardvark: current AQL editor query is now cached

* Upgraded V8 version from 3.16.14 to 3.29.59

  The built-in version of V8 has been upgraded from 3.16.14 to 3.29.59.
  This activates several ES6 (also dubbed *Harmony* or *ES.next*) features in
  ArangoDB, both in the ArangoShell and the ArangoDB server. They can be
  used for scripting and in server-side actions such as Foxx routes, traversals
  etc.

  The following ES6 features are available in ArangoDB 2.4 by default:

  * iterators
  * the `of` operator
  * symbols
  * predefined collections types (Map, Set etc.)
  * typed arrays

  Many other ES6 features are disabled by default, but can be made available by
  starting arangod or arangosh with the appropriate options:

  * arrow functions
  * proxies
  * generators
  * String, Array, and Number enhancements
  * constants
  * enhanced object and numeric literals

  To activate all these ES6 features in arangod or arangosh, start it with
  the following options:

      arangosh --javascript.v8-options="--harmony --harmony_generators"

  More details on the available ES6 features can be found in
  [this blog](https://jsteemann.github.io/blog/2014/12/19/using-es6-features-in-arangodb/).

* Added Foxx generator for building Hypermedia APIs

  A more detailed description is [here](https://www.arangodb.com/2014/12/08/building-hypermedia-apis-foxxgenerator)

* New `Applications` tab in web interface:

  The `applications` tab got a complete redesign.
  It will now only show applications that are currently running on ArangoDB.
  For a selected application, a new detailed view has been created.
  This view provides a better overview of the app:
  * author
  * license
  * version
  * contributors
  * download links
  * API documentation

  To install a new application, a new dialog is now available.
  It provides the features already available in the console application `foxx-manager` plus some more:
  * install an application from Github
  * install an application from a zip file
  * install an application from ArangoDB's application store
  * create a new application from scratch: this feature uses a generator to
    create a Foxx application with pre-defined CRUD methods for a given list
    of collections. The generated Foxx app can either be downloaded as a zip file or
    be installed on the server. Starting with a new Foxx app has never been easier.

* fixed issue #1102: Aardvark: Layout bug in documents overview

  The documents overview was entirely destroyed in some situations on Firefox.
  We replaced the plugin we used there.

* fixed issue #1168: Aardvark: pagination buttons jumping

* fixed issue #1161: Aardvark: Click on Import JSON imports previously uploaded file

* removed configure options `--enable-all-in-one-v8`, `--enable-all-in-one-icu`,
  and `--enable-all-in-one-libev`.

* global internal rename to fix naming incompatibilities with JSON:

  Internal functions with names containing `array` have been renamed to `object`,
  internal functions with names containing `list` have been renamed to `array`.
  The renaming was mainly done in the C++ parts. The documentation has also been
  adjusted so that the correct JSON type names are used in most places.

  The change also led to the addition of a few function aliases in AQL:

  * `TO_LIST` now is an alias of the new `TO_ARRAY`
  * `IS_LIST` now is an alias of the new `IS_ARRAY`
  * `IS_DOCUMENT` now is an alias of the new `IS_OBJECT`

  The changed also renamed the option `mergeArrays` to `mergeObjects` for AQL
  data-modification query options and HTTP document modification API

* AQL: added optimizer rule "remove-filter-covered-by-index"

  This rule removes FilterNodes and CalculationNodes from an execution plan if the
  filter is already covered by a previous IndexRangeNode. Removing the CalculationNode
  and the FilterNode will speed up query execution because the query requires less
  computation.

* AQL: added optimizer rule "remove-sort-rand"

  This rule removes a `SORT RAND()` expression from a query and moves the random
  iteration into the appropriate `EnumerateCollectionNode`. This is more efficient
  than individually enumerating and then sorting randomly.

* AQL: range optimizations for IN and OR

  This change enables usage of indexes for several additional cases. Filters containing
  the `IN` operator can now make use of indexes, and multiple OR- or AND-combined filter
  conditions can now also use indexes if the filters are accessing the same indexed
  attribute.

  Here are a few examples of queries that can now use indexes but couldn't before:

    FOR doc IN collection
      FILTER doc.indexedAttribute == 1 || doc.indexedAttribute > 99
      RETURN doc

    FOR doc IN collection
      FILTER doc.indexedAttribute IN [ 3, 42 ] || doc.indexedAttribute > 99
      RETURN doc

    FOR doc IN collection
      FILTER (doc.indexedAttribute > 2 && doc.indexedAttribute < 10) ||
             (doc.indexedAttribute > 23 && doc.indexedAttribute < 42)
      RETURN doc

* fixed issue #500: AQL parentheses issue

  This change allows passing subqueries as AQL function parameters without using
  duplicate brackets (e.g. `FUNC(query)` instead of `FUNC((query))`

* added optional `COUNT` clause to AQL `COLLECT`

  This allows more efficient group count calculation queries, e.g.

      FOR doc IN collection
        COLLECT age = doc.age WITH COUNT INTO length
        RETURN { age: age, count: length }

  A count-only query is also possible:

      FOR doc IN collection
        COLLECT WITH COUNT INTO length
        RETURN length

* fixed missing makeDirectory when fetching a Foxx application from a zip file

* fixed issue #1134: Change the default endpoint to localhost

  This change will modify the IP address ArangoDB listens on to 127.0.0.1 by default.
  This will make new ArangoDB installations unaccessible from clients other than
  localhost unless changed. This is a security feature.

  To make ArangoDB accessible from any client, change the server's configuration
  (`--server.endpoint`) to either `tcp://0.0.0.0:8529` or the server's publicly
  visible IP address.

* deprecated `Repository#modelPrototype`. Use `Repository#model` instead.

* IMPORTANT CHANGE: by default, system collections are included in replication and all
  replication API return values. This will lead to user accounts and credentials
  data being replicated from master to slave servers. This may overwrite
  slave-specific database users.

  If this is undesired, the `_users` collection can be excluded from replication
  easily by setting the `includeSystem` attribute to `false` in the following commands:

  * replication.sync({ includeSystem: false });
  * replication.applier.properties({ includeSystem: false });

  This will exclude all system collections (including `_aqlfunctions`, `_graphs` etc.)
  from the initial synchronization and the continuous replication.

  If this is also undesired, it is also possible to specify a list of collections to
  exclude from the initial synchronization and the continuous replication using the
  `restrictCollections` attribute, e.g.:

      replication.applier.properties({
        includeSystem: true,
        restrictType: "exclude",
        restrictCollections: [ "_users", "_graphs", "foo" ]
      });

  The HTTP API methods for fetching the replication inventory and for dumping collections
  also support the `includeSystem` control flag via a URL parameter.

* removed DEPRECATED replication methods:
  * `replication.logger.start()`
  * `replication.logger.stop()`
  * `replication.logger.properties()`
  * HTTP PUT `/_api/replication/logger-start`
  * HTTP PUT `/_api/replication/logger-stop`
  * HTTP GET `/_api/replication/logger-config`
  * HTTP PUT `/_api/replication/logger-config`

* fixed issue #1174, which was due to locking problems in distributed
  AQL execution

* improved cluster locking for AQL avoiding deadlocks

* use DistributeNode for modifying queries with REPLACE and UPDATE, if
  possible


v2.3.6 (2015-XX-XX)
-------------------

* fixed AQL subquery optimization that produced wrong result when multiple subqueries
  directly followed each other and and a directly following `LET` statement did refer
  to any but the first subquery.


v2.3.5 (2015-01-16)
-------------------

* fixed intermittent 404 errors in Foxx apps after mounting or unmounting apps

* fixed issue #1200: Expansion operator results in "Cannot call method 'forEach' of null"

* fixed issue #1199: Cannot unlink root node of plan


v2.3.4 (2014-12-23)
-------------------

* fixed cerberus path for MyArangoDB


v2.3.3 (2014-12-17)
-------------------

* fixed error handling in instantiation of distributed AQL queries, this
  also fixes a bug in cluster startup with many servers

* issue #1185: parse non-fractional JSON numbers with exponent (e.g. `4e-261`)

* issue #1159: allow --server.request-timeout and --server.connect-timeout of 0


v2.3.2 (2014-12-09)
-------------------

* fixed issue #1177: Fix bug in the user app's storage

* fixed issue #1173: AQL Editor "Save current query" resets user password

* fixed missing makeDirectory when fetching a Foxx application from a zip file

* put in warning about default changed: fixed issue #1134: Change the default endpoint to localhost

* fixed issue #1163: invalid fullCount value returned from AQL

* fixed range operator precedence

* limit default maximum number of plans created by AQL optimizer to 256 (from 1024)

* make AQL optimizer not generate an extra plan if an index can be used, but modify
  existing plans in place

* fixed AQL cursor ttl (time-to-live) issue

  Any user-specified cursor ttl value was not honored since 2.3.0.

* fixed segfault in AQL query hash index setup with unknown shapes

* fixed memleaks

* added AQL optimizer rule for removing `INTO` from a `COLLECT` statement if not needed

* fixed issue #1131

  This change provides the `KEEP` clause for `COLLECT ... INTO`. The `KEEP` clause
  allows controlling which variables will be kept in the variable created by `INTO`.

* fixed issue #1147, must protect dispatcher ID for etcd

v2.3.1 (2014-11-28)
-------------------

* recreate password if missing during upgrade

* fixed issue #1126

* fixed non-working subquery index optimizations

* do not restrict summary of Foxx applications to 60 characters

* fixed display of "required" path parameters in Foxx application documentation

* added more optimizations of constants values in AQL FILTER conditions

* fixed invalid or-to-in optimization for FILTERs containing comparisons
  with boolean values

* fixed replication of `_graphs` collection

* added AQL list functions `PUSH`, `POP`, `UNSHIFT`, `SHIFT`, `REMOVE_VALUES`,
  `REMOVE_VALUE`, `REMOVE_NTH` and `APPEND`

* added AQL functions `CALL` and `APPLY` to dynamically call other functions

* fixed AQL optimizer cost estimation for LIMIT node

* prevent Foxx queues from permanently writing to the journal even when
  server is idle

* fixed AQL COLLECT statement with INTO clause, which copied more variables
  than v2.2 and thus lead to too much memory consumption.
  This deals with #1107.

* fixed AQL COLLECT statement, this concerned every COLLECT statement,
  only the first group had access to the values of the variables before
  the COLLECT statement. This deals with #1127.

* fixed some AQL internals, where sometimes too many items were
  fetched from upstream in the presence of a LIMIT clause. This should
  generally improve performance.


v2.3.0 (2014-11-18)
-------------------

* fixed syslog flags. `--log.syslog` is deprecated and setting it has no effect,
  `--log.facility` now works as described. Application name has been changed from
  `triagens` to `arangod`. It can be changed using `--log.application`. The syslog
  will only contain the actual log message. The datetime prefix is omitted.

* fixed deflate in SimpleHttpClient

* fixed issue #1104: edgeExamples broken or changed

* fixed issue #1103: Error while importing user queries

* fixed issue #1100: AQL: HAS() fails on doc[attribute_name]

* fixed issue #1098: runtime error when creating graph vertex

* hide system applications in **Applications** tab by default

  Display of system applications can be toggled by using the *system applications*
  toggle in the UI.

* added HTTP REST API for managing tasks (`/_api/tasks`)

* allow passing character lists as optional parameter to AQL functions `TRIM`,
  `LTRIM` and `RTRIM`

  These functions now support trimming using custom character lists. If no character
  lists are specified, all whitespace characters will be removed as previously:

      TRIM("  foobar\t \r\n ")         // "foobar"
      TRIM(";foo;bar;baz, ", "; ")     // "foo;bar;baz"

* added AQL string functions `LTRIM`, `RTRIM`, `FIND_FIRST`, `FIND_LAST`, `SPLIT`,
  `SUBSTITUTE`

* added AQL functions `ZIP`, `VALUES` and `PERCENTILE`

* made AQL functions `CONCAT` and `CONCAT_SEPARATOR` work with list arguments

* dynamically create extra dispatcher threads if required

* fixed issue #1097: schemas in the API docs no longer show required properties as optional


v2.3.0-beta2 (2014-11-08)
-------------------------

* front-end: new icons for uploading and downloading JSON documents into a collection

* front-end: fixed documents pagination css display error

* front-end: fixed flickering of the progress view

* front-end: fixed missing event for documents filter function

* front-end: jsoneditor: added CMD+Return (Mac) CTRL+Return (Linux/Win) shortkey for
  saving a document

* front-end: added information tooltip for uploading json documents.

* front-end: added database management view to the collapsed navigation menu

* front-end: added collection truncation feature

* fixed issue #1086: arangoimp: Odd errors if arguments are not given properly

* performance improvements for AQL queries that use JavaScript-based expressions
  internally

* added AQL geo functions `WITHIN_RECTANGLE` and `IS_IN_POLYGON`

* fixed non-working query results download in AQL editor of web interface

* removed debug print message in AQL editor query export routine

* fixed issue #1075: Aardvark: user name required even if auth is off #1075

  The fix for this prefills the username input field with the current user's
  account name if any and `root` (the default username) otherwise. Additionally,
  the tooltip text has been slightly adjusted.

* fixed issue #1069: Add 'raw' link to swagger ui so that the raw swagger
  json can easily be retrieved

  This adds a link to the Swagger API docs to an application's detail view in
  the **Applications** tab of the web interface. The link produces the Swagger
  JSON directly. If authentication is turned on, the link requires authentication,
  too.

* documentation updates


v2.3.0-beta1 (2014-11-01)
-------------------------

* added dedicated `NOT IN` operator for AQL

  Previously, a `NOT IN` was only achievable by writing a negated `IN` condition:

      FOR i IN ... FILTER ! (i IN [ 23, 42 ]) ...

  This can now alternatively be expressed more intuitively as follows:

      FOR i IN ... FILTER i NOT IN [ 23, 42 ] ...

* added alternative logical operator syntax for AQL

  Previously, the logical operators in AQL could only be written as:
  - `&&`: logical and
  - `||`: logical or
  - `!`: negation

  ArangoDB 2.3 introduces the alternative variants for these operators:
  - `AND`: logical and
  - `OR`: logical or
  - `NOT`: negation

  The new syntax is just an alternative to the old syntax, allowing easier
  migration from SQL. The old syntax is still fully supported and will be.

* improved output of `ArangoStatement.parse()` and POST `/_api/query`

  If an AQL query can be parsed without problems, The return value of
  `ArangoStatement.parse()` now contains an attribute `ast` with the abstract
  syntax tree of the query (before optimizations). Though this is an internal
  representation of the query and is subject to change, it can be used to inspect
  how ArangoDB interprets a given query.

* improved `ArangoStatement.explain()` and POST `/_api/explain`

  The commands for explaining AQL queries have been improved.

* added command-line option `--javascript.v8-contexts` to control the number of
  V8 contexts created in arangod.

  Previously, the number of V8 contexts was equal to the number of server threads
  (as specified by option `--server.threads`).

  However, it may be sensible to create different amounts of threads and V8
  contexts. If the option is not specified, the number of V8 contexts created
  will be equal to the number of server threads. Thus no change in configuration
  is required to keep the old behavior.

  If you are using the default config files or merge them with your local config
  files, please review if the default number of server threads is okay in your
  environment. Additionally you should verify that the number of V8 contexts
  created (as specified in option `--javascript.v8-contexts`) is okay.

* the number of server.threads specified is now the minimum of threads
  started. There are situation in which threads are waiting for results of
  distributed database servers. In this case the number of threads is
  dynamically increased.

* removed index type "bitarray"

  Bitarray indexes were only half-way documented and integrated in previous versions
  of ArangoDB so their benefit was limited. The support for bitarray indexes has
  thus been removed in ArangoDB 2.3. It is not possible to create indexes of type
  "bitarray" with ArangoDB 2.3.

  When a collection is opened that contains a bitarray index definition created
  with a previous version of ArangoDB, ArangoDB will ignore it and log the following
  warning:

      index type 'bitarray' is not supported in this version of ArangoDB and is ignored

  Future versions of ArangoDB may automatically remove such index definitions so the
  warnings will eventually disappear.

* removed internal "_admin/modules/flush" in order to fix requireApp

* added basic support for handling binary data in Foxx

  Requests with binary payload can be processed in Foxx applications by
  using the new method `res.rawBodyBuffer()`. This will return the unparsed request
  body as a Buffer object.

  There is now also the method `req.requestParts()` available in Foxx to retrieve
  the individual components of a multipart HTTP request.

  Buffer objects can now be used when setting the response body of any Foxx action.
  Additionally, `res.send()` has been added as a convenience method for returning
  strings, JSON objects or buffers from a Foxx action:

      res.send("<p>some HTML</p>");
      res.send({ success: true });
      res.send(new Buffer("some binary data"));

  The convenience method `res.sendFile()` can now be used to easily return the
  contents of a file from a Foxx action:

      res.sendFile(applicationContext.foxxFilename("image.png"));

  `fs.write` now accepts not only strings but also Buffer objects as second parameter:

      fs.write(filename, "some data");
      fs.write(filename, new Buffer("some binary data"));

  `fs.readBuffer` can be used to return the contents of a file in a Buffer object.

* improved performance of insertion into non-unique hash indexes significantly in case
  many duplicate keys are used in the index

* issue #1042: set time zone in log output

  the command-line option `--log.use-local-time` was added to print dates and times in
  the server-local timezone instead of UTC

* command-line options that require a boolean value now validate the
  value given on the command-line

  This prevents issues if no value is specified for an option that
  requires a boolean value. For example, the following command-line would
  have caused trouble in 2.2, because `--server.endpoint` would have been
  used as the value for the `--server.disable-authentication` options
  (which requires a boolean value):

      arangod --server.disable-authentication --server.endpoint tcp://127.0.0.1:8529 data

  In 2.3, running this command will fail with an error and requires to
  be modified to:

      arangod --server.disable-authentication true --server.endpoint tcp://127.0.0.1:8529 data

* improved performance of CSV import in arangoimp

* fixed issue #1027: Stack traces are off-by-one

* fixed issue #1026: Modules loaded in different files within the same app
  should refer to the same module

* fixed issue #1025: Traversal not as expected in undirected graph

* added a _relation function in the general-graph module.

  This deprecated _directedRelation and _undirectedRelation.
  ArangoDB does not offer any constraints for undirected edges
  which caused some confusion of users how undirected relations
  have to be handled. Relation now only supports directed relations
  and the user can actively simulate undirected relations.

* changed return value of Foxx.applicationContext#collectionName:

  Previously, the function could return invalid collection names because
  invalid characters were not replaced in the application name prefix, only
  in the collection name passed.

  Now, the function replaces invalid characters also in the application name
  prefix, which might to slightly different results for application names that
  contained any characters outside the ranges [a-z], [A-Z] and [0-9].

* prevent XSS in AQL editor and logs view

* integrated tutorial into ArangoShell and web interface

* added option `--backslash-escape` for arangoimp when running CSV file imports

* front-end: added download feature for (filtered) documents

* front-end: added download feature for the results of a user query

* front-end: added function to move documents to another collection

* front-end: added sort-by attribute to the documents filter

* front-end: added sorting feature to database, graph management and user management view.

* issue #989: front-end: Databases view not refreshing after deleting a database

* issue #991: front-end: Database search broken

* front-end: added infobox which shows more information about a document (_id, _rev, _key) or
  an edge (_id, _rev, _key, _from, _to). The from and to attributes are clickable and redirect
  to their document location.

* front-end: added edit-mode for deleting multiple documents at the same time.

* front-end: added delete button to the detailed document/edge view.

* front-end: added visual feedback for saving documents/edges inside the editor (error/success).

* front-end: added auto-focusing for the first input field in a modal.

* front-end: added validation for user input in a modal.

* front-end: user defined queries are now stored inside the database and are bound to the current
  user, instead of using the local storage functionality of the browsers. The outcome of this is
  that user defined queries are now independently usable from any device. Also queries can now be
  edited through the standard document editor of the front-end through the _users collection.

* front-end: added import and export functionality for user defined queries.

* front-end: added new keywords and functions to the aql-editor theme

* front-end: applied tile-style to the graph view

* front-end: now using the new graph api including multi-collection support

* front-end: foxx apps are now deletable

* front-end: foxx apps are now installable and updateable through github, if github is their
  origin.

* front-end: added foxx app version control. Multiple versions of a single foxx app are now
  installable and easy to manage and are also arranged in groups.

* front-end: the user-set filter of a collection is now stored until the user navigates to
  another collection.

* front-end: fetching and filtering of documents, statistics, and query operations are now
  handled with asynchronous ajax calls.

* front-end: added progress indicator if the front-end is waiting for a server operation.

* front-end: fixed wrong count of documents in the documents view of a collection.

* front-end: fixed unexpected styling of the manage db view and navigation.

* front-end: fixed wrong handling of select fields in a modal view.

* front-end: fixed wrong positioning of some tooltips.

* automatically call `toJSON` function of JavaScript objects (if present)
  when serializing them into database documents. This change allows
  storing JavaScript date objects in the database in a sensible manner.


v2.2.7 (2014-11-19)
-------------------

* fixed issue #998: Incorrect application URL for non-system Foxx apps

* fixed issue #1079: AQL editor: keyword WITH in UPDATE query is not highlighted

* fix memory leak in cluster nodes

* fixed registration of AQL user-defined functions in Web UI (JS shell)

* fixed error display in Web UI for certain errors
  (now error message is printed instead of 'undefined')

* fixed issue #1059: bug in js module console

* fixed issue #1056: "fs": zip functions fail with passwords

* fixed issue #1063: Docs: measuring unit of --wal.logfile-size?

* fixed issue #1062: Docs: typo in 14.2 Example data


v2.2.6 (2014-10-20)
-------------------

* fixed issue #972: Compilation Issue

* fixed issue #743: temporary directories are now unique and one can read
  off the tool that created them, if empty, they are removed atexit

* Highly improved performance of all AQL GRAPH_* functions.

* Orphan collections in general graphs can now be found via GRAPH_VERTICES
  if either "any" or no direction is defined

* Fixed documentation for AQL function GRAPH_NEIGHBORS.
  The option "vertexCollectionRestriction" is meant to filter the target
  vertices only, and should not filter the path.

* Fixed a bug in GRAPH_NEIGHBORS which enforced only empty results
  under certain conditions


v2.2.5 (2014-10-09)
-------------------

* fixed issue #961: allow non-JSON values in undocument request bodies

* fixed issue 1028: libicu is now statically linked

* fixed cached lookups of collections on the server, which may have caused spurious
  problems after collection rename operations


v2.2.4 (2014-10-01)
-------------------

* fixed accessing `_from` and `_to` attributes in `collection.byExample` and
  `collection.firstExample`

  These internal attributes were not handled properly in the mentioned functions, so
  searching for them did not always produce documents

* fixed issue #1030: arangoimp 2.2.3 crashing, not logging on large Windows CSV file

* fixed issue #1025: Traversal not as expected in undirected graph

* fixed issue #1020

  This requires re-introducing the startup option `--database.force-sync-properties`.

  This option can again be used to force fsyncs of collection, index and database properties
  stored as JSON strings on disk in files named `parameter.json`. Syncing these files after
  a write may be necessary if the underlying storage does not sync file contents by itself
  in a "sensible" amount of time after a file has been written and closed.

  The default value is `true` so collection, index and database properties will always be
  synced to disk immediately. This affects creating, renaming and dropping collections as
  well as creating and dropping databases and indexes. Each of these operations will perform
  an additional fsync on the `parameter.json` file if the option is set to `true`.

  It might be sensible to set this option to `false` for workloads that create and drop a
  lot of collections (e.g. test runs).

  Document operations such as creating, updating and dropping documents are not affected
  by this option.

* fixed issue #1016: AQL editor bug

* fixed issue #1014: WITHIN function returns wrong distance

* fixed AQL shortest path calculation in function `GRAPH_SHORTEST_PATH` to return
  complete vertex objects instead of just vertex ids

* allow changing of attributes of documents stored in server-side JavaScript variables

  Previously, the following did not work:

      var doc = db.collection.document(key);
      doc._key = "abc"; // overwriting internal attributes not supported
      doc.value = 123;  // overwriting existing attributes not supported

  Now, modifying documents stored in server-side variables (e.g. `doc` in the above case)
  is supported. Modifying the variables will not update the documents in the database,
  but will modify the JavaScript object (which can be written back to the database using
  `db.collection.update` or `db.collection.replace`)

* fixed issue #997: arangoimp apparently doesn't support files >2gig on Windows

  large file support (requires using `_stat64` instead of `stat`) is now supported on
  Windows


v2.2.3 (2014-09-02)
-------------------

* added `around` for Foxx controller

* added `type` option for HTTP API `GET /_api/document?collection=...`

  This allows controlling the type of results to be returned. By default, paths to
  documents will be returned, e.g.

      [
        `/_api/document/test/mykey1`,
        `/_api/document/test/mykey2`,
        ...
      ]

  To return a list of document ids instead of paths, the `type` URL parameter can be
  set to `id`:

      [
        `test/mykey1`,
        `test/mykey2`,
        ...
      ]

  To return a list of document keys only, the `type` URL parameter can be set to `key`:

      [
        `mykey1`,
        `mykey2`,
        ...
      ]


* properly capitalize HTTP response header field names in case the `x-arango-async`
  HTTP header was used in a request.

* fixed several documentation issues

* speedup for several general-graph functions, AQL functions starting with `GRAPH_`
  and traversals


v2.2.2 (2014-08-08)
-------------------

* allow storing non-reserved attribute names starting with an underscore

  Previous versions of ArangoDB parsed away all attribute names that started with an
  underscore (e.g. `_test', '_foo', `_bar`) on all levels of a document (root level
  and sub-attribute levels). While this behavior was documented, it was unintuitive and
  prevented storing documents inside other documents, e.g.:

      {
        "_key" : "foo",
        "_type" : "mydoc",
        "references" : [
          {
            "_key" : "something",
            "_rev" : "...",
            "value" : 1
          },
          {
            "_key" : "something else",
            "_rev" : "...",
            "value" : 2
          }
        ]
      }

  In the above example, previous versions of ArangoDB removed all attributes and
  sub-attributes that started with underscores, meaning the embedded documents would lose
  some of their attributes. 2.2.2 should preserve such attributes, and will also allow
  storing user-defined attribute names on the top-level even if they start with underscores
  (such as `_type` in the above example).

* fix conversion of JavaScript String, Number and Boolean objects to JSON.

  Objects created in JavaScript using `new Number(...)`, `new String(...)`, or
  `new Boolean(...)` were not converted to JSON correctly.

* fixed a race condition on task registration (i.e. `require("org/arangodb/tasks").register()`)

  this race condition led to undefined behavior when a just-created task with no offset and
  no period was instantly executed and deleted by the task scheduler, before the `register`
  function returned to the caller.

* changed run-tests.sh to execute all suitable tests.

* switch to new version of gyp

* fixed upgrade button


v2.2.1 (2014-07-24)
-------------------

* fixed hanging write-ahead log recovery for certain cases that involved dropping
  databases

* fixed issue with --check-version: when creating a new database the check failed

* issue #947 Foxx applicationContext missing some properties

* fixed issue with --check-version: when creating a new database the check failed

* added startup option `--wal.suppress-shape-information`

  Setting this option to `true` will reduce memory and disk space usage and require
  less CPU time when modifying documents or edges. It should therefore be turned on
  for standalone ArangoDB servers. However, for servers that are used as replication
  masters, setting this option to `true` will effectively disable the usage of the
  write-ahead log for replication, so it should be set to `false` for any replication
  master servers.

  The default value for this option is `false`.

* added optional `ttl` attribute to specify result cursor expiration for HTTP API method
  `POST /_api/cursor`

  The `ttl` attribute can be used to prevent cursor results from timing out too early.

* issue #947: Foxx applicationContext missing some properties

* (reported by Christian Neubauer):

  The problem was that in Google's V8, signed and unsigned chars are not always declared cleanly.
  so we need to force v8 to compile with forced signed chars which is done by the Flag:
    -fsigned-char
  at least it is enough to follow the instructions of compiling arango on rasperry
  and add "CFLAGS='-fsigned-char'" to the make command of V8 and remove the armv7=0

* Fixed a bug with the replication client. In the case of single document
  transactions the collection was not write locked.


v2.2.0 (2014-07-10)
-------------------

* The replication methods `logger.start`, `logger.stop` and `logger.properties` are
  no-ops in ArangoDB 2.2 as there is no separate replication logger anymore. Data changes
  are logged into the write-ahead log in ArangoDB 2.2, and not separately by the
  replication logger. The replication logger object is still there in ArangoDB 2.2 to
  ensure backwards-compatibility, however, logging cannot be started, stopped or
  configured anymore. Using any of these methods will do nothing.

  This also affects the following HTTP API methods:
  - `PUT /_api/replication/logger-start`
  - `PUT /_api/replication/logger-stop`
  - `GET /_api/replication/logger-config`
  - `PUT /_api/replication/logger-config`

  Using any of these methods is discouraged from now on as they will be removed in
  future versions of ArangoDB.

* INCOMPATIBLE CHANGE: replication of transactions has changed. Previously, transactions
  were logged on a master in one big block and shipped to a slave in one block, too.
  Now transactions will be logged and replicated as separate entries, allowing transactions
  to be bigger and also ensure replication progress.

  This change also affects the behavior of the `stop` method of the replication applier.
  If the replication applier is now stopped manually using the `stop` method and later
  restarted using the `start` method, any transactions that were unfinished at the
  point of stopping will be aborted on a slave, even if they later commit on the master.

  In ArangoDB 2.2, stopping the replication applier manually should be avoided unless the
  goal is to stop replication permanently or to do a full resync with the master anyway.
  If the replication applier still must be stopped, it should be made sure that the
  slave has fetched and applied all pending operations from a master, and that no
  extra transactions are started on the master before the `stop` command on the slave
  is executed.

  Replication of transactions in ArangoDB 2.2 might also lock the involved collections on
  the slave while a transaction is either committed or aborted on the master and the
  change has been replicated to the slave. This change in behavior may be important for
  slave servers that are used for read-scaling. In order to avoid long lasting collection
  locks on the slave, transactions should be kept small.

  The `_replication` system collection is not used anymore in ArangoDB 2.2 and its usage is
  discouraged.

* INCOMPATIBLE CHANGE: the figures reported by the `collection.figures` method
  now only reflect documents and data contained in the journals and datafiles of
  collections. Documents or deletions contained only in the write-ahead log will
  not influence collection figures until the write-ahead log garbage collection
  kicks in. The figures for a collection might therefore underreport the total
  resource usage of a collection.

  Additionally, the attributes `lastTick` and `uncollectedLogfileEntries` have been
  added to the result of the `figures` operation and the HTTP API method
  `PUT /_api/collection/figures`

* added `insert` method as an alias for `save`. Documents can now be inserted into
  a collection using either method:

      db.test.save({ foo: "bar" });
      db.test.insert({ foo: "bar" });

* added support for data-modification AQL queries

* added AQL keywords `INSERT`, `UPDATE`, `REPLACE` and `REMOVE` (and `WITH`) to
  support data-modification AQL queries.

  Unquoted usage of these keywords for attribute names in AQL queries will likely
  fail in ArangoDB 2.2. If any such attribute name needs to be used in a query, it
  should be enclosed in backticks to indicate the usage of a literal attribute
  name.

  For example, the following query will fail in ArangoDB 2.2 with a parse error:

      FOR i IN foo RETURN i.remove

  and needs to be rewritten like this:

      FOR i IN foo RETURN i.`remove`

* disallow storing of JavaScript objects that contain JavaScript native objects
  of type `Date`, `Function`, `RegExp` or `External`, e.g.

      db.test.save({ foo: /bar/ });
      db.test.save({ foo: new Date() });

  will now print

      Error: <data> cannot be converted into JSON shape: could not shape document

  Previously, objects of these types were silently converted into an empty object
  (i.e. `{ }`).

  To store such objects in a collection, explicitly convert them into strings
  like this:

      db.test.save({ foo: String(/bar/) });
      db.test.save({ foo: String(new Date()) });

* The replication methods `logger.start`, `logger.stop` and `logger.properties` are
  no-ops in ArangoDB 2.2 as there is no separate replication logger anymore. Data changes
  are logged into the write-ahead log in ArangoDB 2.2, and not separately by the
  replication logger. The replication logger object is still there in ArangoDB 2.2 to
  ensure backwards-compatibility, however, logging cannot be started, stopped or
  configured anymore. Using any of these methods will do nothing.

  This also affects the following HTTP API methods:
  - `PUT /_api/replication/logger-start`
  - `PUT /_api/replication/logger-stop`
  - `GET /_api/replication/logger-config`
  - `PUT /_api/replication/logger-config`

  Using any of these methods is discouraged from now on as they will be removed in
  future versions of ArangoDB.

* INCOMPATIBLE CHANGE: replication of transactions has changed. Previously, transactions
  were logged on a master in one big block and shipped to a slave in one block, too.
  Now transactions will be logged and replicated as separate entries, allowing transactions
  to be bigger and also ensure replication progress.

  This change also affects the behavior of the `stop` method of the replication applier.
  If the replication applier is now stopped manually using the `stop` method and later
  restarted using the `start` method, any transactions that were unfinished at the
  point of stopping will be aborted on a slave, even if they later commit on the master.

  In ArangoDB 2.2, stopping the replication applier manually should be avoided unless the
  goal is to stop replication permanently or to do a full resync with the master anyway.
  If the replication applier still must be stopped, it should be made sure that the
  slave has fetched and applied all pending operations from a master, and that no
  extra transactions are started on the master before the `stop` command on the slave
  is executed.

  Replication of transactions in ArangoDB 2.2 might also lock the involved collections on
  the slave while a transaction is either committed or aborted on the master and the
  change has been replicated to the slave. This change in behavior may be important for
  slave servers that are used for read-scaling. In order to avoid long lasting collection
  locks on the slave, transactions should be kept small.

  The `_replication` system collection is not used anymore in ArangoDB 2.2 and its usage is
  discouraged.

* INCOMPATIBLE CHANGE: the figures reported by the `collection.figures` method
  now only reflect documents and data contained in the journals and datafiles of
  collections. Documents or deletions contained only in the write-ahead log will
  not influence collection figures until the write-ahead log garbage collection
  kicks in. The figures for a collection might therefore underreport the total
  resource usage of a collection.

  Additionally, the attributes `lastTick` and `uncollectedLogfileEntries` have been
  added to the result of the `figures` operation and the HTTP API method
  `PUT /_api/collection/figures`

* added `insert` method as an alias for `save`. Documents can now be inserted into
  a collection using either method:

      db.test.save({ foo: "bar" });
      db.test.insert({ foo: "bar" });

* added support for data-modification AQL queries

* added AQL keywords `INSERT`, `UPDATE`, `REPLACE` and `REMOVE` (and `WITH`) to
  support data-modification AQL queries.

  Unquoted usage of these keywords for attribute names in AQL queries will likely
  fail in ArangoDB 2.2. If any such attribute name needs to be used in a query, it
  should be enclosed in backticks to indicate the usage of a literal attribute
  name.

  For example, the following query will fail in ArangoDB 2.2 with a parse error:

      FOR i IN foo RETURN i.remove

  and needs to be rewritten like this:

      FOR i IN foo RETURN i.`remove`

* disallow storing of JavaScript objects that contain JavaScript native objects
  of type `Date`, `Function`, `RegExp` or `External`, e.g.

      db.test.save({ foo: /bar/ });
      db.test.save({ foo: new Date() });

  will now print

      Error: <data> cannot be converted into JSON shape: could not shape document

  Previously, objects of these types were silently converted into an empty object
  (i.e. `{ }`).

  To store such objects in a collection, explicitly convert them into strings
  like this:

      db.test.save({ foo: String(/bar/) });
      db.test.save({ foo: String(new Date()) });

* honor startup option `--server.disable-statistics` when deciding whether or not
  to start periodic statistics collection jobs

  Previously, the statistics collection jobs were started even if the server was
  started with the `--server.disable-statistics` flag being set to `true`

* removed startup option `--random.no-seed`

  This option had no effect in previous versions of ArangoDB and was thus removed.

* removed startup option `--database.remove-on-drop`

  This option was used for debugging only.

* removed startup option `--database.force-sync-properties`

  This option is now superfluous as collection properties are now stored in the
  write-ahead log.

* introduced write-ahead log

  All write operations in an ArangoDB server instance are automatically logged
  to the server's write-ahead log. The write-ahead log is a set of append-only
  logfiles, and it is used in case of a crash recovery and for replication.
  Data from the write-ahead log will eventually be moved into the journals or
  datafiles of collections, allowing the server to remove older write-ahead log
  logfiles. Figures of collections will be updated when data are moved from the
  write-ahead log into the journals or datafiles of collections.

  Cross-collection transactions in ArangoDB should benefit considerably by this
  change, as less writes than in previous versions are required to ensure the data
  of multiple collections are atomically and durably committed. All data-modifying
  operations inside transactions (insert, update, remove) will write their
  operations into the write-ahead log directly, making transactions with multiple
  operations also require less physical memory than in previous versions of ArangoDB,
  that required all transaction data to fit into RAM.

  The `_trx` system collection is not used anymore in ArangoDB 2.2 and its usage is
  discouraged.

  The data in the write-ahead log can also be used in the replication context.
  The `_replication` collection that was used in previous versions of ArangoDB to
  store all changes on the server is not used anymore in ArangoDB 2.2. Instead,
  slaves can read from a master's write-ahead log to get informed about most
  recent changes. This removes the need to store data-modifying operations in
  both the actual place and the `_replication` collection.

* removed startup option `--server.disable-replication-logger`

  This option is superfluous in ArangoDB 2.2. There is no dedicated replication
  logger in ArangoDB 2.2. There is now always the write-ahead log, and it is also
  used as the server's replication log. Specifying the startup option
  `--server.disable-replication-logger` will do nothing in ArangoDB 2.2, but the
  option should not be used anymore as it might be removed in a future version.

* changed behavior of replication logger

  There is no dedicated replication logger in ArangoDB 2.2 as there is the
  write-ahead log now. The existing APIs for starting and stopping the replication
  logger still exist in ArangoDB 2.2 for downwards-compatibility, but calling
  the start or stop operations are no-ops in ArangoDB 2.2. When querying the
  replication logger status via the API, the server will always report that the
  replication logger is running. Configuring the replication logger is a no-op
  in ArangoDB 2.2, too. Changing the replication logger configuration has no
  effect. Instead, the write-ahead log configuration can be changed.

* removed MRuby integration for arangod

  ArangoDB had an experimental MRuby integration in some of the publish builds.
  This wasn't continuously developed, and so it has been removed in ArangoDB 2.2.

  This change has led to the following startup options being superfluous:

  - `--ruby.gc-interval`
  - `--ruby.action-directory`
  - `--ruby.modules-path`
  - `--ruby.startup-directory`

  Specifying these startup options will do nothing in ArangoDB 2.2, but the
  options should be avoided from now on as they might be removed in future versions.

* reclaim index memory when last document in collection is deleted

  Previously, deleting documents from a collection did not lead to index sizes being
  reduced. Instead, the already allocated index memory was re-used when a collection
  was refilled.

  Now, index memory for primary indexes and hash indexes is reclaimed instantly when
  the last document from a collection is removed.

* inlined and optimized functions in hash indexes

* added AQL TRANSLATE function

  This function can be used to perform lookups from static lists, e.g.

      LET countryNames = { US: "United States", UK: "United Kingdom", FR: "France" }
      RETURN TRANSLATE("FR", countryNames)

* fixed datafile debugger

* fixed check-version for empty directory

* moved try/catch block to the top of routing chain

* added mountedApp function for foxx-manager

* fixed issue #883: arango 2.1 - when starting multi-machine cluster, UI web
  does not change to cluster overview

* fixed dfdb: should not start any other V8 threads

* cleanup of version-check, added module org/arangodb/database-version,
  added --check-version option

* fixed issue #881: [2.1.0] Bombarded (every 10 sec or so) with
  "WARNING format string is corrupt" when in non-system DB Dashboard

* specialized primary index implementation to allow faster hash table
  rebuilding and reduce lookups in datafiles for the actual value of `_key`.

* issue #862: added `--overwrite` option to arangoimp

* removed number of property lookups for documents during AQL queries that
  access documents

* prevent buffering of long print results in arangosh's and arangod's print
  command

  this change will emit buffered intermediate print results and discard the
  output buffer to quickly deliver print results to the user, and to prevent
  constructing very large buffers for large results

* removed sorting of attribute names for use in a collection's shaper

  sorting attribute names was done on document insert to keep attributes
  of a collection in sorted order for faster comparisons. The sort order
  of attributes was only used in one particular and unlikely case, so it
  was removed. Collections with many different attribute names should
  benefit from this change by faster inserts and slightly less memory usage.

* fixed a bug in arangodump which got the collection name in _from and _to
  attributes of edges wrong (all were "_unknown")

* fixed a bug in arangorestore which did not recognize wrong _from and _to
  attributes of edges

* improved error detection and reporting in arangorestore


v2.1.1 (2014-06-06)
-------------------

* fixed dfdb: should not start any other V8 threads

* signature for collection functions was modified

  The basic change was the substitution of the input parameter of the
  function by an generic options object which can contain multiple
  option parameter of the function.
  Following functions were modified
  remove
  removeBySample
  replace
  replaceBySample
  update
  updateBySample

  Old signature is yet supported but it will be removed in future versions

v2.1.0 (2014-05-29)
-------------------

* implemented upgrade procedure for clusters

* fixed communication issue with agency which prevented reconnect
  after an agent failure

* fixed cluster dashboard in the case that one but not all servers
  in the cluster are down

* fixed a bug with coordinators creating local database objects
  in the wrong order (_system needs to be done first)

* improved cluster dashboard


v2.1.0-rc2 (2014-05-25)
-----------------------

* fixed issue #864: Inconsistent behavior of AQL REVERSE(list) function


v2.1.0-rc1 (XXXX-XX-XX)
-----------------------

* added server-side periodic task management functions:

  - require("org/arangodb/tasks").register(): registers a periodic task
  - require("org/arangodb/tasks").unregister(): unregisters and removes a
    periodic task
  - require("org/arangodb/tasks").get(): retrieves a specific tasks or all
    existing tasks

  the previous undocumented function `internal.definePeriodic` is now
  deprecated and will be removed in a future release.

* decrease the size of some seldom used system collections on creation.

  This will make these collections use less disk space and mapped memory.

* added AQL date functions

* added AQL FLATTEN() list function

* added index memory statistics to `db.<collection>.figures()` function

  The `figures` function will now return a sub-document `indexes`, which lists
  the number of indexes in the `count` sub-attribute, and the total memory
  usage of the indexes in bytes in the `size` sub-attribute.

* added AQL CURRENT_DATABASE() function

  This function returns the current database's name.

* added AQL CURRENT_USER() function

  This function returns the current user from an AQL query. The current user is the
  username that was specified in the `Authorization` HTTP header of the request. If
  authentication is turned off or the query was executed outside a request context,
  the function will return `null`.

* fixed issue #796: Searching with newline chars broken?

  fixed slightly different handling of backslash escape characters in a few
  AQL functions. Now handling of escape sequences should be consistent, and
  searching for newline characters should work the same everywhere

* added OpenSSL version check for configure

  It will report all OpenSSL versions < 1.0.1g as being too old.
  `configure` will only complain about an outdated OpenSSL version but not stop.

* require C++ compiler support (requires g++ 4.8, clang++ 3.4 or Visual Studio 13)

* less string copying returning JSONified documents from ArangoDB, e.g. via
  HTTP GET `/_api/document/<collection>/<document>`

* issue #798: Lower case http headers from arango

  This change allows returning capitalized HTTP headers, e.g.
  `Content-Length` instead of `content-length`.
  The HTTP spec says that headers are case-insensitive, but
  in fact several clients rely on a specific case in response
  headers.
  This change will capitalize HTTP headers if the `X-Arango-Version`
  request header is sent by the client and contains a value of at
  least `20100` (for version 2.1). The default value for the
  compatibility can also be set at server start, using the
  `--server.default-api-compatibility` option.

* simplified usage of `db._createStatement()`

  Previously, the function could not be called with a query string parameter as
  follows:

      db._createStatement(queryString);

  Calling it as above resulted in an error because the function expected an
  object as its parameter. From now on, it's possible to call the function with
  just the query string.

* make ArangoDB not send back a `WWW-Authenticate` header to a client in case the
  client sends the `X-Omit-WWW-Authenticate` HTTP header.

  This is done to prevent browsers from showing their built-in HTTP authentication
  dialog for AJAX requests that require authentication.
  ArangoDB will still return an HTTP 401 (Unauthorized) if the request doesn't
  contain valid credentials, but it will omit the `WWW-Authenticate` header,
  allowing clients to bypass the browser's authentication dialog.

* added REST API method HTTP GET `/_api/job/job-id` to query the status of an
  async job without potentially fetching it from the list of done jobs

* fixed non-intuitive behavior in jobs API: previously, querying the status
  of an async job via the API HTTP PUT `/_api/job/job-id` removed a currently
  executing async job from the list of queryable jobs on the server.
  Now, when querying the result of an async job that is still executing,
  the job is kept in the list of queryable jobs so its result can be fetched
  by a subsequent request.

* use a new data structure for the edge index of an edge collection. This
  improves the performance for the creation of the edge index and in
  particular speeds up removal of edges in graphs. Note however that
  this change might change the order in which edges starting at
  or ending in a vertex are returned. However, this order was never
  guaranteed anyway and it is not sensible to guarantee any particular
  order.

* provide a size hint to edge and hash indexes when initially filling them
  this will lead to less re-allocations when populating these indexes

  this may speed up building indexes when opening an existing collection

* don't requeue identical context methods in V8 threads in case a method is
  already registered

* removed arangod command line option `--database.remove-on-compacted`

* export the sort attribute for graph traversals to the HTTP interface

* add support for arangodump/arangorestore for clusters


v2.0.8 (XXXX-XX-XX)
-------------------

* fixed too-busy iteration over skiplists

  Even when a skiplist query was restricted by a limit clause, the skiplist
  index was queried without the limit. this led to slower-than-necessary
  execution times.

* fixed timeout overflows on 32 bit systems

  this bug has led to problems when select was called with a high timeout
  value (2000+ seconds) on 32bit systems that don't have a forgiving select
  implementation. when the call was made on these systems, select failed
  so no data would be read or sent over the connection

  this might have affected some cluster-internal operations.

* fixed ETCD issues on 32 bit systems

  ETCD was non-functional on 32 bit systems at all. The first call to the
  watch API crashed it. This was because atomic operations worked on data
  structures that were not properly aligned on 32 bit systems.

* fixed issue #848: db.someEdgeCollection.inEdge does not return correct
  value when called the 2nd time after a .save to the edge collection


v2.0.7 (2014-05-05)
-------------------

* issue #839: Foxx Manager missing "unfetch"

* fixed a race condition at startup

  this fixes undefined behavior in case the logger was involved directly at
  startup, before the logger initialization code was called. This should have
  occurred only for code that was executed before the invocation of main(),
  e.g. during ctor calls of statically defined objects.


v2.0.6 (2014-04-22)
-------------------

* fixed issue #835: arangosh doesn't show correct database name



v2.0.5 (2014-04-21)
-------------------

* Fixed a caching problem in IE JS Shell

* added cancelation for async jobs

* upgraded to new gyp for V8

* new Windows installer


v2.0.4 (2014-04-14)
-------------------

* fixed cluster authentication front-end issues for Firefox and IE, there are
  still problems with Chrome


v2.0.3 (2014-04-14)
-------------------

* fixed AQL optimizer bug

* fixed front-end issues

* added password change dialog


v2.0.2 (2014-04-06)
-------------------

* during cluster startup, do not log (somewhat expected) connection errors with
  log level error, but with log level info

* fixed dashboard modals

* fixed connection check for cluster planning front end: firefox does
  not support async:false

* document how to persist a cluster plan in order to relaunch an existing
  cluster later


v2.0.1 (2014-03-31)
-------------------

* make ArangoDB not send back a `WWW-Authenticate` header to a client in case the
  client sends the `X-Omit-WWW-Authenticate` HTTP header.

  This is done to prevent browsers from showing their built-in HTTP authentication
  dialog for AJAX requests that require authentication.
  ArangoDB will still return an HTTP 401 (Unauthorized) if the request doesn't
  contain valid credentials, but it will omit the `WWW-Authenticate` header,
  allowing clients to bypass the browser's authentication dialog.

* fixed isses in arango-dfdb:

  the dfdb was not able to unload certain system collections, so these couldn't be
  inspected with the dfdb sometimes. Additionally, it did not truncate corrupt
  markers from datafiles under some circumstances

* added `changePassword` attribute for users

* fixed non-working "save" button in collection edit view of web interface
  clicking the save button did nothing. one had to press enter in one of the input
  fields to send modified form data

* fixed V8 compile error on MacOS X

* prevent `body length: -9223372036854775808` being logged in development mode for
  some Foxx HTTP responses

* fixed several bugs in web interface dashboard

* fixed issue #783: coffee script not working in manifest file

* fixed issue #783: coffee script not working in manifest file

* fixed issue #781: Cant save current query from AQL editor ui

* bumped version in `X-Arango-Version` compatibility header sent by arangosh and other
  client tools from `1.5` to `2.0`.

* fixed startup options for arango-dfdb, added details option for arango-dfdb

* fixed display of missing error messages and codes in arangosh

* when creating a collection via the web interface, the collection type was always
  "document", regardless of the user's choice


v2.0.0 (2014-03-10)
-------------------

* first 2.0 release


v2.0.0-rc2 (2014-03-07)
-----------------------

* fixed cluster authorization


v2.0.0-rc1 (2014-02-28)
-----------------------

* added sharding :-)

* added collection._dbName attribute to query the name of the database from a collection

  more detailed documentation on the sharding and cluster features can be found in the user
  manual, section **Sharding**

* INCOMPATIBLE CHANGE: using complex values in AQL filter conditions with operators other
  than equality (e.g. >=, >, <=, <) will disable usage of skiplist indexes for filter
  evaluation.

  For example, the following queries will be affected by change:

      FOR doc IN docs FILTER doc.value < { foo: "bar" } RETURN doc
      FOR doc IN docs FILTER doc.value >= [ 1, 2, 3 ] RETURN doc

  The following queries will not be affected by the change:

      FOR doc IN docs FILTER doc.value == 1 RETURN doc
      FOR doc IN docs FILTER doc.value == "foo" RETURN doc
      FOR doc IN docs FILTER doc.value == [ 1, 2, 3 ] RETURN doc
      FOR doc IN docs FILTER doc.value == { foo: "bar" } RETURN doc

* INCOMPATIBLE CHANGE: removed undocumented method `collection.saveOrReplace`

  this feature was never advertised nor documented nor tested.

* INCOMPATIBLE CHANGE: removed undocumented REST API method `/_api/simple/BY-EXAMPLE-HASH`

  this feature was never advertised nor documented nor tested.

* added explicit startup parameter `--server.reuse-address`

  This flag can be used to control whether sockets should be acquired with the SO_REUSEADDR
  flag.

  Regardless of this setting, sockets on Windows are always acquired using the
  SO_EXCLUSIVEADDRUSE flag.

* removed undocumented REST API method GET `/_admin/database-name`

* added user validation API at POST `/_api/user/<username>`

* slightly improved users management API in `/_api/user`:

  Previously, when creating a new user via HTTP POST, the username needed to be
  passed in an attribute `username`. When users were returned via this API,
  the usernames were returned in an attribute named `user`. This was slightly
  confusing and was changed in 2.0 as follows:

  - when adding a user via HTTP POST, the username can be specified in an attribute
  `user`. If this attribute is not used, the API will look into the attribute `username`
  as before and use that value.
  - when users are returned via HTTP GET, the usernames are still returned in an
    attribute `user`.

  This change should be fully downwards-compatible with the previous version of the API.

* added AQL SLICE function to extract slices from lists

* made module loader more node compatible

* the startup option `--javascript.package-path` for arangosh is now deprecated and does
  nothing. Using it will not cause an error, but the option is ignored.

* added coffee script support

* Several UI improvements.

* Exchanged icons in the graphviewer toolbar

* always start networking and HTTP listeners when starting the server (even in
  console mode)

* allow vertex and edge filtering with user-defined functions in TRAVERSAL,
  TRAVERSAL_TREE and SHORTEST_PATH AQL functions:

      // using user-defined AQL functions for edge and vertex filtering
      RETURN TRAVERSAL(friends, friendrelations, "friends/john", "outbound", {
        followEdges: "myfunctions::checkedge",
        filterVertices: "myfunctions::checkvertex"
      })

      // using the following custom filter functions
      var aqlfunctions = require("org/arangodb/aql/functions");
      aqlfunctions.register("myfunctions::checkedge", function (config, vertex, edge, path) {
        return (edge.type !== 'dislikes'); // don't follow these edges
      }, false);

      aqlfunctions.register("myfunctions::checkvertex", function (config, vertex, path) {
        if (vertex.isDeleted || ! vertex.isActive) {
          return [ "prune", "exclude" ]; // exclude these and don't follow them
        }
        return [ ]; // include everything else
      }, false);

* fail if invalid `strategy`, `order` or `itemOrder` attribute values
  are passed to the AQL TRAVERSAL function. Omitting these attributes
  is not considered an error, but specifying an invalid value for any
  of these attributes will make an AQL query fail.

* issue #751: Create database through API should return HTTP status code 201

  By default, the server now returns HTTP 201 (created) when creating a new
  database successfully. To keep compatibility with older ArangoDB versions, the
  startup parameter `--server.default-api-compatibility` can be set to a value
  of `10400` to indicate API compatibility with ArangoDB 1.4. The compatibility
  can also be enforced by setting the `X-Arango-Version` HTTP header in a
  client request to this API on a per-request basis.

* allow direct access from the `db` object to collections whose names start
  with an underscore (e.g. db._users).

  Previously, access to such collections via the `db` object was possible from
  arangosh, but not from arangod (and thus Foxx and actions). The only way
  to access such collections from these places was via the `db._collection(<name>)`
  workaround.

* allow `\n` (as well as `\r\n`) as line terminator in batch requests sent to
  `/_api/batch` HTTP API.

* use `--data-binary` instead of `--data` parameter in generated cURL examples

* issue #703: Also show path of logfile for fm.config()

* issue #675: Dropping a collection used in "graph" module breaks the graph

* added "static" Graph.drop() method for graphs API

* fixed issue #695: arangosh server.password error

* use pretty-printing in `--console` mode by default

* simplified ArangoDB startup options

  Some startup options are now superfluous or their usage is simplified. The
  following options have been changed:

  * `--javascript.modules-path`: this option has been removed. The modules paths
    are determined by arangod and arangosh automatically based on the value of
    `--javascript.startup-directory`.

    If the option is set on startup, it is ignored so startup will not abort with
    an error `unrecognized option`.

  * `--javascript.action-directory`: this option has been removed. The actions
    directory is determined by arangod automatically based on the value of
    `--javascript.startup-directory`.

    If the option is set on startup, it is ignored so startup will not abort with
    an error `unrecognized option`.

  * `--javascript.package-path`: this option is still available but it is not
    required anymore to set the standard package paths (e.g. `js/npm`). arangod
    will automatically use this standard package path regardless of whether it
    was specified via the options.

    It is possible to use this option to add additional package paths to the
    standard value.

  Configuration files included with arangod are adjusted accordingly.

* layout of the graphs tab adapted to better fit with the other tabs

* database selection is moved to the bottom right corner of the web interface

* removed priority queue index type

  this feature was never advertised nor documented nor tested.

* display internal attributes in document source view of web interface

* removed separate shape collections

  When upgrading to ArangoDB 2.0, existing collections will be converted to include
  shapes and attribute markers in the datafiles instead of using separate files for
  shapes.

  When a collection is converted, existing shapes from the SHAPES directory will
  be written to a new datafile in the collection directory, and the SHAPES directory
  will be removed afterwards.

  This saves up to 2 MB of memory and disk space for each collection
  (savings are higher, the less different shapes there are in a collection).
  Additionally, one less file descriptor per opened collection will be used.

  When creating a new collection, the amount of sync calls may be reduced. The same
  may be true for documents with yet-unknown shapes. This may help performance
  in these cases.

* added AQL functions `NTH` and `POSITION`

* added signal handler for arangosh to save last command in more cases

* added extra prompt placeholders for arangosh:
  - `%e`: current endpoint
  - `%u`: current user

* added arangosh option `--javascript.gc-interval` to control amount of
  garbage collection performed by arangosh

* fixed issue #651: Allow addEdge() to take vertex ids in the JS library

* removed command-line option `--log.format`

  In previous versions, this option did not have an effect for most log messages, so
  it got removed.

* removed C++ logger implementation

  Logging inside ArangoDB is now done using the LOG_XXX() macros. The LOGGER_XXX()
  macros are gone.

* added collection status "loading"


v1.4.16 (XXXX-XX-XX)
--------------------

* fixed too eager datafile deletion

  this issue could have caused a crash when the compaction had marked datafiles as obsolete
  and they were removed while "old" temporary query results still pointed to the old datafile
  positions

* fixed issue #826: Replication fails when a collection's configuration changes


v1.4.15 (2014-04-19)
--------------------

* bugfix for AQL query optimizer

  the following type of query was too eagerly optimized, leading to errors in code-generation:

      LET a = (FOR i IN [] RETURN i) LET b = (FOR i IN [] RETURN i) RETURN 1

  the problem occurred when both lists in the subqueries were empty. In this case invalid code
  was generated and the query couldn't be executed.


v1.4.14 (2014-04-05)
--------------------

* fixed race conditions during shape / attribute insertion

  A race condition could have led to spurious `cannot find attribute #xx` or
  `cannot find shape #xx` (where xx is a number) warning messages being logged
  by the server. This happened when a new attribute was inserted and at the same
  time was queried by another thread.

  Also fixed a race condition that may have occurred when a thread tried to
  access the shapes / attributes hash tables while they were resized. In this
  cases, the shape / attribute may have been hashed to a wrong slot.

* fixed a memory barrier / cpu synchronization problem with libev, affecting
  Windows with Visual Studio 2013 (probably earlier versions are affected, too)

  The issue is described in detail here:
  http://lists.schmorp.de/pipermail/libev/2014q1/002318.html


v1.4.13 (2014-03-14)
--------------------

* added diagnostic output for Foxx application upload

* allow dump & restore from ArangoDB 1.4 with an ArangoDB 2.0 server

* allow startup options `temp-path` and `default-language` to be specified from the arangod
  configuration file and not only from the command line

* fixed too eager compaction

  The compaction will now wait for several seconds before trying to re-compact the same
  collection. Additionally, some other limits have been introduced for the compaction.


v1.4.12 (2014-03-05)
--------------------

* fixed display bug in web interface which caused the following problems:
  - documents were displayed in web interface as being empty
  - document attributes view displayed many attributes with content "undefined"
  - document source view displayed many attributes with name "TYPEOF" and value "undefined"
  - an alert popping up in the browser with message "Datatables warning..."

* re-introduced old-style read-write locks to supports Windows versions older than
  Windows 2008R2 and Windows 7. This should re-enable support for Windows Vista and
  Windows 2008.


v1.4.11 (2014-02-27)
--------------------

* added SHORTEST_PATH AQL function

  this calculates the shortest paths between two vertices, using the Dijkstra
  algorithm, employing a min-heap

  By default, ArangoDB does not know the distance between any two vertices and
  will use a default distance of 1. A custom distance function can be registered
  as an AQL user function to make the distance calculation use any document
  attributes or custom logic:

      RETURN SHORTEST_PATH(cities, motorways, "cities/CGN", "cities/MUC", "outbound", {
        paths: true,
        distance: "myfunctions::citydistance"
      })

      // using the following custom distance function
      var aqlfunctions = require("org/arangodb/aql/functions");
      aqlfunctions.register("myfunctions::distance", function (config, vertex1, vertex2, edge) {
        return Math.sqrt(Math.pow(vertex1.x - vertex2.x) + Math.pow(vertex1.y - vertex2.y));
      }, false);

* fixed bug in Graph.pathTo function

* fixed small memleak in AQL optimizer

* fixed access to potentially uninitialized variable when collection had a cap constraint


v1.4.10 (2014-02-21)
--------------------

* fixed graph constructor to allow graph with some parameter to be used

* added node.js "events" and "stream"

* updated npm packages

* added loading of .json file

* Fixed http return code in graph api with waitForSync parameter.

* Fixed documentation in graph, simple and index api.

* removed 2 tests due to change in ruby library.

* issue #756: set access-control-expose-headers on CORS response

  the following headers are now whitelisted by ArangoDB in CORS responses:
  - etag
  - content-encoding
  - content-length
  - location
  - server
  - x-arango-errors
  - x-arango-async-id


v1.4.9 (2014-02-07)
-------------------

* return a document's current etag in response header for HTTP HEAD requests on
  documents that return an HTTP 412 (precondition failed) error. This allows
  retrieving the document's current revision easily.

* added AQL function `SKIPLIST` to directly access skiplist indexes from AQL

  This is a shortcut method to use a skiplist index for retrieving specific documents in
  indexed order. The function capability is rather limited, but it may be used
  for several cases to speed up queries. The documents are returned in index order if
  only one condition is used.

      /* return all documents with mycollection.created > 12345678 */
      FOR doc IN SKIPLIST(mycollection, { created: [[ '>', 12345678 ]] })
        RETURN doc

      /* return first document with mycollection.created > 12345678 */
      FOR doc IN SKIPLIST(mycollection, { created: [[ '>', 12345678 ]] }, 0, 1)
        RETURN doc

      /* return all documents with mycollection.created between 12345678 and 123456790 */
      FOR doc IN SKIPLIST(mycollection, { created: [[ '>', 12345678 ], [ '<=', 123456790 ]] })
        RETURN doc

      /* return all documents with mycollection.a equal 1 and .b equal 2 */
      FOR doc IN SKIPLIST(mycollection, { a: [[ '==', 1 ]], b: [[ '==', 2 ]] })
        RETURN doc

  The function requires a skiplist index with the exact same attributes to
  be present on the specified collection. All attributes present in the skiplist
  index must be specified in the conditions specified for the `SKIPLIST` function.
  Attribute declaration order is important, too: attributes must be specified in the
  same order in the condition as they have been declared in the skiplist index.

* added command-line option `--server.disable-authentication-unix-sockets`

  with this option, authentication can be disabled for all requests coming
  in via UNIX domain sockets, enabling clients located on the same host as
  the ArangoDB server to connect without authentication.
  Other connections (e.g. TCP/IP) are not affected by this option.

  The default value for this option is `false`.
  Note: this option is only supported on platforms that support Unix domain
  sockets.

* call global arangod instance destructor on shutdown

* issue #755: TRAVERSAL does not use strategy, order and itemOrder options

  these options were not honored when configuring a traversal via the AQL
  TRAVERSAL function. Now, these options are used if specified.

* allow vertex and edge filtering with user-defined functions in TRAVERSAL,
  TRAVERSAL_TREE and SHORTEST_PATH AQL functions:

      // using user-defined AQL functions for edge and vertex filtering
      RETURN TRAVERSAL(friends, friendrelations, "friends/john", "outbound", {
        followEdges: "myfunctions::checkedge",
        filterVertices: "myfunctions::checkvertex"
      })

      // using the following custom filter functions
      var aqlfunctions = require("org/arangodb/aql/functions");
      aqlfunctions.register("myfunctions::checkedge", function (config, vertex, edge, path) {
        return (edge.type !== 'dislikes'); // don't follow these edges
      }, false);

      aqlfunctions.register("myfunctions::checkvertex", function (config, vertex, path) {
        if (vertex.isDeleted || ! vertex.isActive) {
          return [ "prune", "exclude" ]; // exclude these and don't follow them
        }
        return [ ]; // include everything else
      }, false);

* issue #748: add vertex filtering to AQL's TRAVERSAL[_TREE]() function


v1.4.8 (2014-01-31)
-------------------

* install foxx apps in the web interface

* fixed a segfault in the import API


v1.4.7 (2014-01-23)
-------------------

* issue #744: Add usage example arangoimp from Command line

* issue #738: added __dirname, __filename pseudo-globals. Fixes #733. (@by pluma)

* mount all Foxx applications in system apps directory on startup


v1.4.6 (2014-01-20)
-------------------

* issue #736: AQL function to parse collection and key from document handle

* added fm.rescan() method for Foxx-Manager

* fixed issue #734: foxx cookie and route problem

* added method `fm.configJson` for arangosh

* include `startupPath` in result of API `/_api/foxx/config`


v1.4.5 (2014-01-15)
-------------------

* fixed issue #726: Alternate Windows Install Method

* fixed issue #716: dpkg -P doesn't remove everything

* fixed bugs in description of HTTP API `_api/index`

* fixed issue #732: Rest API GET revision number

* added missing documentation for several methods in HTTP API `/_api/edge/...`

* fixed typos in description of HTTP API `_api/document`

* defer evaluation of AQL subqueries and logical operators (lazy evaluation)

* Updated font in WebFrontend, it now contains a version that renders properly on Windows

* generally allow function return values as call parameters to AQL functions

* fixed potential deadlock in global context method execution

* added override file "arangod.conf.local" (and co)


v1.4.4 (2013-12-24)
-------------------

* uid and gid are now set in the scripts, there is no longer a separate config file for
  arangod when started from a script

* foxx-manager is now an alias for arangosh

* arango-dfdb is now an alias for arangod, moved from bin to sbin

* changed from readline to linenoise for Windows

* added --install-service and --uninstall-service for Windows

* removed --daemon and --supervisor for Windows

* arangosh and arangod now uses the config-file which maps the binary name, i. e. if you
  rename arangosh to foxx-manager it will use the config file foxx-manager.conf

* fixed lock file for Windows

* fixed issue #711, #687: foxx-manager throws internal errors

* added `--server.ssl-protocol` option for client tools
  this allows connecting from arangosh, arangoimp, arangoimp etc. to an ArangoDB
  server that uses a non-default value for `--server.ssl-protocol`. The default
  value for the SSL protocol is 4 (TLSv1). If the server is configured to use a
  different protocol, it was not possible to connect to it with the client tools.

* added more detailed request statistics

  This adds the number of async-executed HTTP requests plus the number of HTTP
  requests per individual HTTP method type.

* added `--force` option for arangorestore
  this option allows continuing a restore operation even if the server reports errors
  in the middle of the restore operation

* better error reporting for arangorestore
  in case the server returned an HTTP error, arangorestore previously reported this
  error as `internal error` without any details only. Now server-side errors are
  reported by arangorestore with the server's error message

* include more system collections in dumps produced by arangodump
  previously some system collections were intentionally excluded from dumps, even if the
  dump was run with `--include-system-collections`. for example, the collections `_aal`,
  `_modules`, `_routing`, and `_users` were excluded. This makes sense in a replication
  context but not always in a dump context.
  When specifying `--include-system-collections`, arangodump will now include the above-
  mentioned collections in the dump, too. Some other system collections are still excluded
  even when the dump is run with `--include-system-collections`, for example `_replication`
  and `_trx`.

* fixed issue #701: ArangoStatement undefined in arangosh

* fixed typos in configuration files


v1.4.3 (2013-11-25)
-------------------

* fixed a segfault in the AQL optimizer, occurring when a constant non-list value was
  used on the right-hand side of an IN operator that had a collection attribute on the
  left-hand side

* issue #662:

  Fixed access violation errors (crashes) in the Windows version, occurring under some
  circumstances when accessing databases with multiple clients in parallel

* fixed issue #681: Problem with ArchLinux PKGBUILD configuration


v1.4.2 (2013-11-20)
-------------------

* fixed issue #669: Tiny documentation update

* ported Windows version to use native Windows API SRWLocks (slim read-write locks)
  and condition variables instead of homemade versions

  MSDN states the following about the compatibility of SRWLocks and Condition Variables:

      Minimum supported client:
      Windows Server 2008 [desktop apps | Windows Store apps]

      Minimum supported server:
      Windows Vista [desktop apps | Windows Store apps]

* fixed issue #662: ArangoDB on Windows hanging

  This fixes a deadlock issue that occurred on Windows when documents were written to
  a collection at the same time when some other thread tried to drop the collection.

* fixed file-based logging in Windows

  the logger complained on startup if the specified log file already existed

* fixed startup of server in daemon mode (`--daemon` startup option)

* fixed a segfault in the AQL optimizer

* issue #671: Method graph.measurement does not exist

* changed Windows condition variable implementation to use Windows native
  condition variables

  This is an attempt to fix spurious Windows hangs as described in issue #662.

* added documentation for JavaScript traversals

* added --code-page command-line option for Windows version of arangosh

* fixed a problem when creating edges via the web interface.

  The problem only occurred if a collection was created with type "document
  collection" via the web interface, and afterwards was dropped and re-created
  with type "edge collection". If the web interface page was not reloaded,
  the old collection type (document) was cached, making the subsequent creation
  of edges into the (seeming-to-be-document) collection fail.

  The fix is to not cache the collection type in the web interface. Users of
  an older version of the web interface can reload the collections page if they
  are affected.

* fixed a caching problem in arangosh: if a collection was created using the web
  interface, and then removed via arangosh, arangosh did not actually drop the
  collection due to caching.

  Because the `drop` operation was not carried out, this caused misleading error
  messages when trying to re-create the collection (e.g. `cannot create collection:
  duplicate name`).

* fixed ALT-introduced characters for arangosh console input on Windows

  The Windows readline port was not able to handle characters that are built
  using CTRL or ALT keys. Regular characters entered using the CTRL or ALT keys
  were silently swallowed and not passed to the terminal input handler.

  This did not seem to cause problems for the US keyboard layout, but was a
  severe issue for keyboard layouts that require the ALT (or ALT-GR) key to
  construct characters. For example, entering the character `{` with a German
  keyboard layout requires pressing ALT-GR + 9.

* fixed issue #665: Hash/skiplist combo madness bit my ass

  this fixes a problem with missing/non-deterministic rollbacks of inserts in
  case of a unique constraint violation into a collection with multiple secondary
  indexes (with at least one of them unique)

* fixed issue #664: ArangoDB installer on Windows requires drive c:

* partly fixed issue #662: ArangoDB on Windows hanging

  This fixes dropping databases on Windows. In previous 1.4 versions on Windows,
  one shape collection file was not unloaded and removed when dropping a database,
  leaving one directory and one shape collection file in the otherwise-dropped
  database directory.

* fixed issue #660: updated documentation on indexes


v1.4.1 (2013-11-08)
-------------------

* performance improvements for skip-list deletes


v1.4.1-rc1 (2013-11-07)
-----------------------

* fixed issue #635: Web-Interface should have a "Databases" Menu for Management

* fixed issue #624: Web-Interface is missing a Database selector

* fixed segfault in bitarray query

* fixed issue #656: Cannot create unique index through web interface

* fixed issue #654: bitarray index makes server down

* fixed issue #653: Slow query

* fixed issue #650: Randomness of any() should be improved

* made AQL `DOCUMENT()` function polymorphic and work with just one parameter.

  This allows using the `DOCUMENT` function like this:

      DOCUMENT('users/john')
      DOCUMENT([ 'users/john', 'users/amy' ])

  in addition to the existing use cases:

      DOCUMENT(users, 'users/john')
      DOCUMENT(users, 'john')
      DOCUMENT(users, [ 'users/john' ])
      DOCUMENT(users, [ 'users/john', 'users/amy' ])
      DOCUMENT(users, [ 'john', 'amy' ])

* simplified usage of ArangoDB batch API

  It is not necessary anymore to send the batch boundary in the HTTP `Content-Type`
  header. Previously, the batch API expected the client to send a Content-Type header
  of`multipart/form-data; boundary=<some boundary value>`. This is still supported in
  ArangoDB 2.0, but clients can now also omit this header. If the header is not
  present in a client request, ArangoDB will ignore the request content type and
  read the MIME boundary from the beginning of the request body.

  This also allows using the batch API with the Swagger "Try it out" feature (which is
  not too good at sending a different or even dynamic content-type request header).

* added API method GET `/_api/database/user`

  This returns the list of databases a specific user can see without changing the
  username/passwd.

* issue #424: Documentation about IDs needs to be upgraded


v1.4.0 (2013-10-29)
-------------------

* fixed issue #648: /batch API is missing from Web Interface API Documentation (Swagger)

* fixed issue #647: Icon tooltips missing

* fixed issue #646: index creation in web interface

* fixed issue #645: Allow jumping from edge to linked vertices

* merged PR for issue #643: Some minor corrections and a link to "Downloads"

* fixed issue #642: Completion of error handling

* fixed issue #639: compiling v1.4 on maverick produces warnings on -Wstrict-null-sentinel

* fixed issue #634: Web interface bug: Escape does not always propagate

* fixed issue #620: added startup option `--server.default-api-compatibility`

  This adds the following changes to the ArangoDB server and clients:
  - the server provides a new startup option `--server.default-api-compatibility`.
    This option can be used to determine the compatibility of (some) server API
    return values. The value for this parameter is a server version number,
    calculated as follows: `10000 * major + 100 * minor` (e.g. `10400` for ArangoDB
    1.3). The default value is `10400` (1.4), the minimum allowed value is `10300`
    (1.3).

    When setting this option to a value lower than the current server version,
    the server might respond with old-style results to "old" clients, increasing
    compatibility with "old" (non-up-to-date) clients.

  - the server will on each incoming request check for an HTTP header
    `x-arango-version`. Clients can optionally set this header to the API
    version number they support. For example, if a client sends the HTTP header
    `x-arango-version: 10300`, the server will pick this up and might send ArangoDB
    1.3-style responses in some situations.

    Setting either the startup parameter or using the HTTP header (or both) allows
    running "old" clients with newer versions of ArangoDB, without having to adjust
    the clients too much.

  - the `location` headers returned by the server for the APIs `/_api/document/...`
    and `/_api/collection/...` will have different values depending on the used API
    version. If the API compatibility is `10300`, the `location` headers returned
    will look like this:

        location: /_api/document/....

    whereas when an API compatibility of `10400` or higher is used, the `location`
    headers will look like this:

        location: /_db/<database name>/_api/document/...

  Please note that even in the presence of this, old API versions still may not
  be supported forever by the server.

* fixed issue #643: Some minor corrections and a link to "Downloads" by @frankmayer

* started issue #642: Completion of error handling

* fixed issue #639: compiling v1.4 on maverick produces warnings on
  -Wstrict-null-sentinel

* fixed issue #621: Standard Config needs to be fixed

* added function to manage indexes (web interface)

* improved server shutdown time by signaling shutdown to applicationserver,
  logging, cleanup and compactor threads

* added foxx-manager `replace` command

* added foxx-manager `installed` command (a more intuitive alias for `list`)

* fixed issue #617: Swagger API is missing '/_api/version'

* fixed issue #615: Swagger API: Some commands have no parameter entry forms

* fixed issue #614: API : Typo in : Request URL /_api/database/current

* fixed issue #609: Graph viz tool - different background color

* fixed issue #608: arangosh config files - eventually missing in the manual

* fixed issue #607: Admin interface: no core documentation

* fixed issue #603: Aardvark Foxx App Manager

* fixed a bug in type-mapping between AQL user functions and the AQL layer

  The bug caused errors like the following when working with collection documents
  in an AQL user function:

      TypeError: Cannot assign to read only property '_id' of #<ShapedJson>

* create less system collections when creating a new database

  This is achieved by deferring collection creation until the collections are actually
  needed by ArangoDB. The following collections are affected by the change:
  - `_fishbowl`
  - `_structures`


v1.4.0-beta2 (2013-10-14)
-------------------------

* fixed compaction on Windows

  The compaction on Windows did not ftruncate the cleaned datafiles to a smaller size.
  This has been fixed so not only the content of the files is cleaned but also files
  are re-created with potentially smaller sizes.

* only the following system collections will be excluded from replication from now on:
  - `_replication`
  - `_trx`
  - `_users`
  - `_aal`
  - `_fishbowl`
  - `_modules`
  - `_routing`

  Especially the following system collections will now be included in replication:
  - `_aqlfunctions`
  - `_graphs`

  In previous versions of ArangoDB, all system collections were excluded from the
  replication.

  The change also caused a change in the replication logger and applier:
  in previous versions of ArangoDB, only a collection's id was logged for an operation.
  This has not caused problems for non-system collections but for system collections
  there ids might differ. In addition to a collection id ArangoDB will now also log the
  name of a collection for each replication event.

  The replication applier will now look for the collection name attribute in logged
  events preferably.

* added database selection to arango-dfdb

* provide foxx-manager, arangodump, and arangorestore in Windows build

* ArangoDB 1.4 will refuse to start if option `--javascript.app-path` is not set.

* added startup option `--server.allow-method-override`

  This option can be set to allow overriding the HTTP request method in a request using
  one of the following custom headers:

  - x-http-method-override
  - x-http-method
  - x-method-override

  This allows bypassing proxies and tools that would otherwise just let certain types of
  requests pass. Enabling this option may impose a security risk, so it should only be
  used in very controlled environments.

  The default value for this option is `false` (no method overriding allowed).

* added "details" URL parameter for bulk import API

  Setting the `details` URL parameter to `true` in a call to POST `/_api/import` will make
  the import return details about non-imported documents in the `details` attribute. If
  `details` is `false` or omitted, no `details` attribute will be present in the response.
  This is the same behavior that previous ArangoDB versions exposed.

* added "complete" option for bulk import API

  Setting the `complete` URL parameter to `true` in a call to POST `/_api/import` will make
  the import completely fail if at least one of documents cannot be imported successfully.

  It defaults to `false`, which will make ArangoDB continue importing the other documents
  from the import even if some documents cannot be imported. This is the same behavior that
  previous ArangoDB versions exposed.

* added missing swagger documentation for `/_api/log`

* calling `/_api/logs` (or `/_admin/logs`) is only permitted from the `_system` database now.

  Calling this API method for/from other database will result in an HTTP 400.

' ported fix from https://github.com/novus/nvd3/commit/0894152def263b8dee60192f75f66700cea532cc

  This prevents JavaScript errors from occurring in Chrome when in the admin interface,
  section "Dashboard".

* show current database name in web interface (bottom right corner)

* added missing documentation for /_api/import in swagger API docs

* allow specification of database name for replication sync command replication applier

  This allows syncing from a master database with a different name than the slave database.

* issue #601: Show DB in prompt

  arangosh now displays the database name as part of the prompt by default.

  Can change the prompt by using the `--prompt` option, e.g.

      > arangosh --prompt "my db is named \"%d\"> "


v1.4.0-beta1 (2013-10-01)
-------------------------

* make the Foxx manager use per-database app directories

  Each database now has its own subdirectory for Foxx applications. Each database
  can thus use different Foxx applications if required. A Foxx app for a specific
  database resides in `<app-path>/databases/<database-name>/<app-name>`.

  System apps are shared between all databases. They reside in `<app-path>/system/<app-name>`.

* only trigger an engine reset in development mode for URLs starting with `/dev/`

  This prevents ArangoDB from reloading all Foxx applications when it is not
  actually necessary.

* changed error code from 10 (bad parameter) to 1232 (invalid key generator) for
  errors that are due to an invalid key generator specification when creating a new
  collection

* automatic detection of content-type / mime-type for Foxx assets based on filenames,
  added possibility to override auto detection

* added endpoint management API at `/_api/endpoint`

* changed HTTP return code of PUT `/_api/cursor` from 400 to 404 in case a
  non-existing cursor is referred to

* issue #360: added support for asynchronous requests

  Incoming HTTP requests with the headers `x-arango-async: true` or
  `x-arango-async: store` will be answered by the server instantly with a generic
  HTTP 202 (Accepted) response.

  The actual requests will be queued and processed by the server asynchronously,
  allowing the client to continue sending other requests without waiting for the
  server to process the actually requested operation.

  The exact point in time when a queued request is executed is undefined. If an
  error occurs during execution of an asynchronous request, the client will not
  be notified by the server.

  The maximum size of the asynchronous task queue can be controlled using the new
  option `--scheduler.maximal-queue-size`. If the queue contains this many number of
  tasks and a new asynchronous request comes in, the server will reject it with an
  HTTP 500 (internal server error) response.

  Results of incoming requests marked with header `x-arango-async: true` will be
  discarded by the server immediately. Clients have no way of accessing the result
  of such asynchronously executed request. This is just _fire and forget_.

  To later retrieve the result of an asynchronously executed request, clients can
  mark a request with the header `x-arango-async: keep`. This makes the server
  store the result of the request in memory until explicitly fetched by a client
  via the `/_api/job` API. The `/_api/job` API also provides methods for basic
  inspection of which pending or already finished requests there are on the server,
  plus ways for garbage collecting unneeded results.

* Added new option `--scheduler.maximal-queue-size`.

* issue #590: Manifest Lint

* added data dump and restore tools, arangodump and arangorestore.

  arangodump can be used to create a logical dump of an ArangoDB database, or
  just dedicated collections. It can be used to dump both a collection's structure
  (properties and indexes) and data (documents).

  arangorestore can be used to restore data from a dump created with arangodump.
  arangorestore currently does not re-create any indexes, and doesn't yet handle
  referenced documents in edges properly when doing just partial restores.
  This will be fixed until 1.4 stable.

* introduced `--server.database` option for arangosh, arangoimp, and arangob.

  The option allows these client tools to use a certain database for their actions.
  In arangosh, the current database can be switched at any time using the command

      db._useDatabase(<name>);

  When no database is specified, all client tools will assume they should use the
  default database `_system`. This is done for downwards-compatibility reasons.

* added basic multi database support (alpha)

  New databases can be created using the REST API POST `/_api/database` and the
  shell command `db._createDatabase(<name>)`.

  The default database in ArangoDB is called `_system`. This database is always
  present and cannot be deleted by the user. When an older version of ArangoDB is
  upgraded to 1.4, the previously only database will automatically become the
  `_system` database.

  New databases can be created with the above commands, and can be deleted with the
  REST API DELETE `/_api/database/<name>` or the shell command `db._dropDatabase(<name>);`.

  Deleting databases is still unstable in ArangoDB 1.4 alpha and might crash the
  server. This will be fixed until 1.4 stable.

  To access a specific database via the HTTP REST API, the `/_db/<name>/` prefix
  can be used in all URLs. ArangoDB will check if an incoming request starts with
  this prefix, and will automatically pick the database name from it. If the prefix
  is not there, ArangoDB will assume the request is made for the default database
  (`_system`). This is done for downwards-compatibility reasons.

  That means, the following URL pathnames are logically identical:

      /_api/document/mycollection/1234
      /_db/_system/document/mycollection/1234

  To access a different database (e.g. `test`), the URL pathname would look like this:

      /_db/test/document/mycollection/1234

  New databases can also be created and existing databases can only be dropped from
  within the default database (`_system`). It is not possible to drop the `_system`
  database itself.

  Cross-database operations are unintended and unsupported. The intention of the
  multi-database feature is to have the possibility to have a few databases managed
  by ArangoDB in parallel, but to only access one database at a time from a connection
  or a request.

  When accessing the web interface via the URL pathname `/_admin/html/` or `/_admin/aardvark`,
  the web interface for the default database (`_system`) will be displayed.
  To access the web interface for a different database, the database name can be
  put into the URLs as a prefix, e.g. `/_db/test/_admin/html` or
  `/_db/test/_admin/aardvark`.

  All internal request handlers and also all user-defined request handlers and actions
  (including Foxx) will only get to see the unprefixed URL pathnames (i.e. excluding
  any database name prefix). This is to ensure downwards-compatibility.

  To access the name of the requested database from any action (including Foxx), use
  use `req.database`.

  For example, when calling the URL `/myapp/myaction`, the content of `req.database`
  will be `_system` (the default database because no database got specified) and the
  content of `req.url` will be `/myapp/myaction`.

  When calling the URL `/_db/test/myapp/myaction`, the content of `req.database` will be
  `test`, and the content of `req.url` will still be `/myapp/myaction`.

* Foxx now excludes files starting with . (dot) when bundling assets

  This mitigates problems with editor swap files etc.

* made the web interface a Foxx application

  This change caused the files for the web interface to be moved from `html/admin` to
  `js/apps/aardvark` in the file system.

  The base URL for the admin interface changed from `_admin/html/index.html` to
  `_admin/aardvark/index.html`.

  The "old" redirection to `_admin/html/index.html` will now produce a 404 error.

  When starting ArangoDB with the `--upgrade` option, this will automatically be remedied
  by putting in a redirection from `/` to `/_admin/aardvark/index.html`, and from
  `/_admin/html/index.html` to `/_admin/aardvark/index.html`.

  This also obsoletes the following configuration (command-line) options:
  - `--server.admin-directory`
  - `--server.disable-admin-interface`

  when using these now obsolete options when the server is started, no error is produced
  for downwards-compatibility.

* changed User-Agent value sent by arangoimp, arangosh, and arangod from "VOC-Agent" to
  "ArangoDB"

* changed journal file creation behavior as follows:

  Previously, a journal file for a collection was always created when a collection was
  created. When a journal filled up and became full, the current journal was made a
  datafile, and a new (empty) journal was created automatically. There weren't many
  intended situations when a collection did not have at least one journal.

  This is changed now as follows:
  - when a collection is created, no journal file will be created automatically
  - when there is a write into a collection without a journal, the journal will be
    created lazily
  - when there is a write into a collection with a full journal, a new journal will
    be created automatically

  From the end user perspective, nothing should have changed, except that there is now
  less disk usage for empty collections. Disk usage of infrequently updated collections
  might also be reduced significantly by running the `rotate()` method of a collection,
  and not writing into a collection subsequently.

* added method `collection.rotate()`

  This allows premature rotation of a collection's current journal file into a (read-only)
  datafile. The purpose of using `rotate()` is to prematurely allow compaction (which is
  performed on datafiles only) on data, even if the journal was not filled up completely.

  Using `rotate()` may make sense in the following scenario:

      c = db._create("test");
      for (i = 0; i < 1000; ++i) {
        c.save(...); // insert lots of data here
      }

      ...
      c.truncate(); // collection is now empty
      // only data in datafiles will be compacted by following compaction runs
      // all data in the current journal would not be compacted

      // calling rotate will make the current journal a datafile, and thus make it
      // eligible for compaction
      c.rotate();

  Using `rotate()` may also be useful when data in a collection is known to not change
  in the immediate future. After having completed all write operations on a collection,
  performing a `rotate()` will reduce the size of the current journal to the actually
  required size (remember that journals are pre-allocated with a specific size) before
  making the journal a datafile. Thus `rotate()` may cause disk space savings, even if
  the datafiles does not qualify for compaction after rotation.

  Note: rotating the journal is asynchronous, so that the actual rotation may be executed
  after `rotate()` returns to the caller.

* changed compaction to merge small datafiles together (up to 3 datafiles are merged in
  a compaction run)

  In the regular case, this should leave less small datafiles stay around on disk and allow
  using less file descriptors in total.

* added AQL MINUS function

* added AQL UNION_DISTINCT function (more efficient than combination of `UNIQUE(UNION())`)

* updated mruby to 2013-08-22

* issue #587: Add db._create() in help for startup arangosh

* issue #586: Share a link on installation instructions in the User Manual

* issue #585: Bison 2.4 missing on Mac for custom build

* issue #584: Web interface images broken in devel

* issue #583: Small documentation update

* issue #581: Parameter binding for attributes

* issue #580: Small improvements (by @guidoreina)

* issue #577: Missing documentation for collection figures in implementor manual

* issue #576: Get disk usage for collections and graphs

  This extends the result of the REST API for /_api/collection/figures with
  the attributes `compactors.count`, `compactors.fileSize`, `shapefiles.count`,
  and `shapefiles.fileSize`.

* issue #575: installing devel version on mac (low prio)

* issue #574: Documentation (POST /_admin/routing/reload)

* issue #558: HTTP cursors, allow count to ignore LIMIT


v1.4.0-alpha1 (2013-08-02)
--------------------------

* added replication. check online manual for details.

* added server startup options `--server.disable-replication-logger` and
  `--server.disable-replication-applier`

* removed action deployment tool, this now handled with Foxx and its manager or
  by kaerus node utility

* fixed a server crash when using byExample / firstExample inside a transaction
  and the collection contained a usable hash/skiplist index for the example

* defineHttp now only expects a single context

* added collection detail dialog (web interface)

  Shows collection properties, figures (datafiles, journals, attributes, etc.)
  and indexes.

* added documents filter (web interface)

  Allows searching for documents based on attribute values. One or many filter
  conditions can be defined, using comparison operators such as '==', '<=', etc.

* improved AQL editor (web interface)

  Editor supports keyboard shortcuts (Submit, Undo, Redo, Select).
  Editor allows saving and reusing of user-defined queries.
  Added example queries to AQL editor.
  Added comment button.

* added document import (web interface)

  Allows upload of JSON-data from files. Files must have an extension of .json.

* added dashboard (web interface)

  Shows the status of replication and multiple system charts, e.g.
  Virtual Memory Size, Request Time, HTTP Connections etc.

* added API method `/_api/graph` to query all graphs with all properties.

* added example queries in web interface AQL editor

* added arango.reconnect(<host>) method for arangosh to dynamically switch server or
  user name

* added AQL range operator `..`

  The `..` operator can be used to easily iterate over a sequence of numeric
  values. It will produce a list of values in the defined range, with both bounding
  values included.

  Example:

      2010..2013

  will produce the following result:

      [ 2010, 2011, 2012, 2013 ]

* added AQL RANGE function

* added collection.first(count) and collection.last(count) document access functions

  These functions allow accessing the first or last n documents in a collection. The order
  is determined by document insertion/update time.

* added AQL INTERSECTION function

* INCOMPATIBLE CHANGE: changed AQL user function namespace resolution operator from `:` to `::`

  AQL user-defined functions were introduced in ArangoDB 1.3, and the namespace resolution
  operator for them was the single colon (`:`). A function call looked like this:

      RETURN mygroup:myfunc()

  The single colon caused an ambiguity in the AQL grammar, making it indistinguishable from
  named attributes or the ternary operator in some cases, e.g.

      { mygroup:myfunc ? mygroup:myfunc }

  The change of the namespace resolution operator from `:` to `::` fixes this ambiguity.

  Existing user functions in the database will be automatically fixed when starting ArangoDB
  1.4 with the `--upgrade` option. However, queries using user-defined functions need to be
  adjusted on the client side to use the new operator.

* allow multiple AQL LET declarations separated by comma, e.g.
  LET a = 1, b = 2, c = 3

* more useful AQL error messages

  The error position (line/column) is more clearly indicated for parse errors.
  Additionally, if a query references a collection that cannot be found, the error
  message will give a hint on the collection name

* changed return value for AQL `DOCUMENT` function in case document is not found

  Previously, when the AQL `DOCUMENT` function was called with the id of a document and
  the document could not be found, it returned `undefined`. This value is not part of the
  JSON type system and this has caused some problems.
  Starting with ArangoDB 1.4, the `DOCUMENT` function will return `null` if the document
  looked for cannot be found.

  In case the function is called with a list of documents, it will continue to return all
  found documents, and will not return `null` for non-found documents. This has not changed.

* added single line comments for AQL

  Single line comments can be started with a double forward slash: `//`.
  They end at the end of the line, or the end of the query string, whichever is first.

* fixed documentation issues #567, #568, #571.

* added collection.checksum(<withData>) method to calculate CRC checksums for
  collections

  This can be used to
  - check if data in a collection has changed
  - compare the contents of two collections on different ArangoDB instances

* issue #565: add description line to aal.listAvailable()

* fixed several out-of-memory situations when double freeing or invalid memory
  accesses could happen

* less msyncing during the creation of collections

  This is achieved by not syncing the initial (standard) markers in shapes collections.
  After all standard markers are written, the shapes collection will get synced.

* renamed command-line option `--log.filter` to `--log.source-filter` to avoid
  misunderstandings

* introduced new command-line option `--log.content-filter` to optionally restrict
  logging to just specific log messages (containing the filter string, case-sensitive).

  For example, to filter on just log entries which contain `ArangoDB`, use:

      --log.content-filter "ArangoDB"

* added optional command-line option `--log.requests-file` to log incoming HTTP
  requests to a file.

  When used, all HTTP requests will be logged to the specified file, containing the
  client IP address, HTTP method, requests URL, HTTP response code, and size of the
  response body.

* added a signal handler for SIGUSR1 signal:

  when ArangoDB receives this signal, it will respond all further incoming requests
  with an HTTP 503 (Service Unavailable) error. This will be the case until another
  SIGUSR1 signal is caught. This will make ArangoDB start serving requests regularly
  again. Note: this is not implemented on Windows.

* limited maximum request URI length to 16384 bytes:

  Incoming requests with longer request URIs will be responded to with an HTTP
  414 (Request-URI Too Long) error.

* require version 1.0 or 1.1 in HTTP version signature of requests sent by clients:

  Clients sending requests with a non-HTTP 1.0 or non-HTTP 1.1 version number will
  be served with an HTTP 505 (HTTP Version Not Supported) error.

* updated manual on indexes:

  using system attributes such as `_id`, `_key`, `_from`, `_to`, `_rev` in indexes is
  disallowed and will be rejected by the server. This was the case since ArangoDB 1.3,
  but was not properly documented.

* issue #563: can aal become a default object?

  aal is now a prefab object in arangosh

* prevent certain system collections from being renamed, dropped, or even unloaded.

  Which restrictions there are for which system collections may vary from release to
  release, but users should in general not try to modify system collections directly
  anyway.

  Note: there are no such restrictions for user-created collections.

* issue #559: added Foxx documentation to user manual

* added server startup option `--server.authenticate-system-only`. This option can be
  used to restrict the need for HTTP authentication to internal functionality and APIs,
  such as `/_api/*` and `/_admin/*`.
  Setting this option to `true` will thus force authentication for the ArangoDB APIs
  and the web interface, but allow unauthenticated requests for other URLs (including
  user defined actions and Foxx applications).
  The default value of this option is `false`, meaning that if authentication is turned
  on, authentication is still required for *all* incoming requests. Only by setting the
  option to `true` this restriction is lifted and authentication becomes required for
  URLs starting with `/_` only.

  Please note that authentication still needs to be enabled regularly by setting the
  `--server.disable-authentication` parameter to `false`. Otherwise no authentication
  will be required for any URLs as before.

* protect collections against unloading when there are still document barriers around.

* extended cap constraints to optionally limit the active data size in a collection to
  a specific number of bytes.

  The arguments for creating a cap constraint are now:
  `collection.ensureCapConstraint(<count>, <byteSize>);`

  It is supported to specify just a count as in ArangoDB 1.3 and before, to specify
  just a fileSize, or both. The first met constraint will trigger the automated
  document removal.

* added `db._exists(doc)` and `collection.exists(doc)` for easy document existence checks

* added API `/_api/current-database` to retrieve information about the database the
  client is currently connected to (note: the API `/_api/current-database` has been
  removed in the meantime. The functionality is accessible via `/_api/database/current`
  now).

* ensure a proper order of tick values in datafiles/journals/compactors.
  any new files written will have the _tick values of their markers in order. for
  older files, there are edge cases at the beginning and end of the datafiles when
  _tick values are not properly in order.

* prevent caching of static pages in PathHandler.
  whenever a static page is requested that is served by the general PathHandler, the
  server will respond to HTTP GET requests with a "Cache-Control: max-age=86400" header.

* added "doCompact" attribute when creating collections and to collection.properties().
  The attribute controls whether collection datafiles are compacted.

* changed the HTTP return code from 400 to 404 for some cases when there is a referral
  to a non-existing collection or document.

* introduced error code 1909 `too many iterations` that is thrown when graph traversals
  hit the `maxIterations` threshold.

* optionally limit traversals to a certain number of iterations
  the limitation can be achieved via the traversal API by setting the `maxIterations`
  attribute, and also via the AQL `TRAVERSAL` and `TRAVERSAL_TREE` functions by setting
  the same attribute. If traversals are not limited by the end user, a server-defined
  limit for `maxIterations` may be used to prevent server-side traversals from running
  endlessly.

* added graph traversal API at `/_api/traversal`

* added "API" link in web interface, pointing to REST API generated with Swagger

* moved "About" link in web interface into "links" menu

* allow incremental access to the documents in a collection from out of AQL
  this allows reading documents from a collection chunks when a full collection scan
  is required. memory usage might be must lower in this case and queries might finish
  earlier if there is an additional LIMIT statement

* changed AQL COLLECT to use a stable sort, so any previous SORT order is preserved

* issue #547: Javascript error in the web interface

* issue #550: Make AQL graph functions support key in addition to id

* issue #526: Unable to escape when an errorneous command is entered into the js shell

* issue #523: Graph and vertex methods for the javascript api

* issue #517: Foxx: Route parameters with capital letters fail

* issue #512: Binded Parameters for LIMIT


v1.3.3 (2013-08-01)
-------------------

* issue #570: updateFishbowl() fails once

* updated and fixed generated examples

* issue #559: added Foxx documentation to user manual

* added missing error reporting for errors that happened during import of edges


v1.3.2 (2013-06-21)
-------------------

* fixed memleak in internal.download()

* made the shape-collection journal size adaptive:
  if too big shapes come in, a shape journal will be created with a big-enough size
  automatically. the maximum size of a shape journal is still restricted, but to a
  very big value that should never be reached in practice.

* fixed a segfault that occurred when inserting documents with a shape size bigger
  than the default shape journal size (2MB)

* fixed a locking issue in collection.truncate()

* fixed value overflow in accumulated filesizes reported by collection.figures()

* issue #545: AQL FILTER unnecessary (?) loop

* issue #549: wrong return code with --daemon


v1.3.1 (2013-05-24)
-------------------

* removed currently unused _ids collection

* fixed usage of --temp-path in aranogd and arangosh

* issue #540: suppress return of temporary internal variables in AQL

* issue #530: ReferenceError: ArangoError is not a constructor

* issue #535: Problem with AQL user functions javascript API

* set --javascript.app-path for test execution to prevent startup error

* issue #532: Graph _edgesCache returns invalid data?

* issue #531: Arangod errors

* issue #529: Really weird transaction issue

* fixed usage of --temp-path in aranogd and arangosh


v1.3.0 (2013-05-10)
-------------------

* fixed problem on restart ("datafile-xxx is not sealed") when server was killed
  during a compaction run

* fixed leak when using cursors with very small batchSize

* issue #508: `unregistergroup` function not mentioned in http interface docs

* issue #507: GET /_api/aqlfunction returns code inside parentheses

* fixed issue #489: Bug in aal.install

* fixed issue 505: statistics not populated on MacOS


v1.3.0-rc1 (2013-04-24)
-----------------------

* updated documentation for 1.3.0

* added node modules and npm packages

* changed compaction to only compact datafiles with more at least 10% of dead
  documents (byte size-wise)

* issue #498: fixed reload of authentication info when using
  `require("org/arangodb/users").reload()`

* issue #495: Passing an empty array to create a document results in a
  "phantom" document

* added more precision for requests statistics figures

* added "sum" attribute for individual statistics results in statistics API
  at /_admin/statistics

* made "limit" an optional parameter in AQL function NEAR().
  limit can now be either omitted completely, or set to 0. If so, an internal
  default value (currently 100) will be applied for the limit.

* issue #481

* added "attributes.count" to output of `collection.figures()`
  this also affects the REST API /_api/collection/<name>/figures

* added IndexedPropertyGetter for ShapedJson objects

* added API for user-defined AQL functions

* issue #475: A better error message for deleting a non-existent graph

* issue #474: Web interface problems with the JS Shell

* added missing documentation for AQL UNION function

* added transaction support.
  This provides ACID transactions for ArangoDB. Transactions can be invoked
  using the `db._executeTransaction()` function, or the `/_api/transaction`
  REST API.

* switched to semantic versioning (at least for alpha & alpha naming)

* added saveOrReplace() for server-side JS

v1.3.alpha1 (2013-04-05)
------------------------

* cleanup of Module, Package, ArangoApp and modules "internal", "fs", "console"

* use Error instead of string in throw to allow stack-trace

* issue #454: error while creation of Collection

* make `collection.count()` not recalculate the number of documents on the fly, but
  use some internal document counters.

* issue #457: invalid string value in web interface

* make datafile id (datafile->_fid) identical to the numeric part of the filename.
  E.g. the datafile `journal-123456.db` will now have a datafile marker with the same
  fid (i.e. `123456`) instead of a different value. This change will only affect
  datafiles that are created with 1.3 and not any older files.
  The intention behind this change is to make datafile debugging easier.

* consistently discard document attributes with reserved names (system attributes)
  but without any known meaning, for example `_test`, `_foo`, ...

  Previously, these attributes were saved with the document regularly in some cases,
  but were discarded in other cases.
  Now these attributes are discarded consistently. "Real" system attributes such as
  `_key`, `_from`, `_to` are not affected and will work as before.

  Additionally, attributes with an empty name (``) are discarded when documents are
  saved.

  Though using reserved or empty attribute names in documents was not really and
  consistently supported in previous versions of ArangoDB, this change might cause
  an incompatibility for clients that rely on this feature.

* added server startup flag `--database.force-sync-properties` to force syncing of
  collection properties on collection creation, deletion and on property update.
  The default value is true to mimic the behavior of previous versions of ArangoDB.
  If set to false, collection properties are written to disk but no call to sync()
  is made.

* added detailed output of server version and components for REST APIs
  `/_admin/version` and `/_api/version`. To retrieve this extended information,
  call the REST APIs with URL parameter `details=true`.

* issue #443: For git-based builds include commit hash in version

* adjust startup log output to be more compact, less verbose

* set the required minimum number of file descriptors to 256.
  On server start, this number is enforced on systems that have rlimit. If the limit
  cannot be enforced, starting the server will fail.
  Note: 256 is considered to be the absolute minimum value. Depending on the use case
  for ArangoDB, a much higher number of file descriptors should be used.

  To avoid checking & potentially changing the number of maximum open files, use the
  startup option `--server.descriptors-minimum 0`

* fixed shapedjson to json conversion for special numeric values (NaN, +inf, -inf).
  Before, "NaN", "inf", or "-inf" were written into the JSONified output, but these
  values are not allowed in JSON. Now, "null" is written to the JSONified output as
  required.

* added AQL functions VARIANCE_POPULATION(), VARIANCE_SAMPLE(), STDDEV_POPULATION(),
  STDDEV_SAMPLE(), AVERAGE(), MEDIAN() to calculate statistical values for lists

* added AQL SQRT() function

* added AQL TRIM(), LEFT() and RIGHT() string functions

* fixed issue #436: GET /_api/document on edge

* make AQL REVERSE() and LENGTH() functions work on strings, too

* disabled DOT generation in `make doxygen`. this speeds up docs generation

* renamed startup option `--dispatcher.report-intervall` to `--dispatcher.report-interval`

* renamed startup option `--scheduler.report-intervall` to `--scheduler.report-interval`

* slightly changed output of REST API method /_admin/log.
  Previously, the log messages returned also contained the date and log level, now
  they will only contain the log message, and no date and log level information.
  This information can be re-created by API users from the `timestamp` and `level`
  attributes of the result.

* removed configure option `--enable-zone-debug`
  memory zone debugging is now automatically turned on when compiling with ArangoDB
  `--enable-maintainer-mode`

* removed configure option `--enable-arangob`
  arangob is now always included in the build


v1.2.3 (XXXX-XX-XX)
-------------------

* added optional parameter `edgexamples` for AQL function EDGES() and NEIGHBORS()

* added AQL function NEIGHBORS()

* added freebsd support

* fixed firstExample() query with `_id` and `_key` attributes

* issue triAGENS/ArangoDB-PHP#55: AQL optimizer may have mis-optimized duplicate
  filter statements with limit


v1.2.2 (2013-03-26)
-------------------

* fixed save of objects with common sub-objects

* issue #459: fulltext internal memory allocation didn't scale well
  This fix improves loading times for collections with fulltext indexes that have
  lots of equal words indexed.

* issue #212: auto-increment support

  The feature can be used by creating a collection with the extra `keyOptions`
  attribute as follows:

      db._create("mycollection", { keyOptions: { type: "autoincrement", offset: 1, increment: 10, allowUserKeys: true } });

  The `type` attribute will make sure the keys will be auto-generated if no
  `_key` attribute is specified for a document.

  The `allowUserKeys` attribute determines whether users might still supply own
  `_key` values with documents or if this is considered an error.

  The `increment` value determines the actual increment value, whereas the `offset`
  value can be used to seed to value sequence with a specific starting value.
  This will be useful later in a multi-master setup, when multiple servers can use
  different auto-increment seed values and thus generate non-conflicting auto-increment values.

  The default values currently are:

  - `allowUserKeys`: `true`
  - `offset`: `0`
  - `increment`: `1`

  The only other available key generator type currently is `traditional`.
  The `traditional` key generator will auto-generate keys in a fashion as ArangoDB
  always did (some increasing integer value, with a more or less unpredictable
  increment value).

  Note that for the `traditional` key generator there is only the option to disallow
  user-supplied keys and give the server the sole responsibility for key generation.
  This can be achieved by setting the `allowUserKeys` property to `false`.

  This change also introduces the following errors that API implementors may want to check
  the return values for:

  - 1222: `document key unexpected`: will be raised when a document is created with
    a `_key` attribute, but the underlying collection was set up with the `keyOptions`
    attribute `allowUserKeys: false`.

  - 1225: `out of keys`: will be raised when the auto-increment key generator runs
    out of keys. This may happen when the next key to be generated is 2^64 or higher.
    In practice, this will only happen if the values for `increment` or `offset` are
    not set appropriately, or if users are allowed to supply own keys, those keys
    are near the 2^64 threshold, and later the auto-increment feature kicks in and
    generates keys that cross that threshold.

    In practice it should not occur with proper configuration and proper usage of the
    collections.

  This change may also affect the following REST APIs:
  - POST `/_api/collection`: the server does now accept the optional `keyOptions`
    attribute in the second parameter
  - GET `/_api/collection/properties`: will return the `keyOptions` attribute as part
    of the collection's properties. The previous optional attribute `createOptions`
    is now gone.

* fixed `ArangoStatement.explain()` method with bind variables

* fixed misleading "cursor not found" error message in arangosh that occurred when
  `count()` was called for client-side cursors

* fixed handling of empty attribute names, which may have crashed the server under
  certain circumstances before

* fixed usage of invalid pointer in error message output when index description could
  not be opened


v1.2.1 (2013-03-14)
-------------------

* issue #444: please darken light color in arangosh

* issue #442: pls update post install info on osx

* fixed conversion of special double values (NaN, -inf, +inf) when converting from
  shapedjson to JSON

* fixed compaction of markers (location of _key was not updated correctly in memory,
  leading to _keys pointing to undefined memory after datafile rotation)

* fixed edge index key pointers to use document master pointer plus offset instead
  of direct _key address

* fixed case when server could not create any more journal or compactor files.
  Previously a wrong status code may have been returned, and not being able to create
  a new compactor file may have led to an infinite loop with error message
  "could not create compactor".

* fixed value truncation for numeric filename parts when renaming datafiles/journals


v1.2.0 (2013-03-01)
-------------------

* by default statistics are now switch off; in order to enable comment out
  the "disable-statistics = yes" line in "arangod.conf"

* fixed issue #435: csv parser skips data at buffer border

* added server startup option `--server.disable-statistics` to turn off statistics
  gathering without recompilation of ArangoDB.
  This partly addresses issue #432.

* fixed dropping of indexes without collection name, e.g.
  `db.xxx.dropIndex("123456");`
  Dropping an index like this failed with an assertion error.

* fixed issue #426: arangoimp should be able to import edges into edge collections

* fixed issue #425: In case of conflict ArangoDB returns HTTP 400 Bad request
  (with 1207 Error) instead of HTTP 409 Conflict

* fixed too greedy token consumption in AQL for negative values:
  e.g. in the statement `RETURN { a: 1 -2 }` the minus token was consumed as part
  of the value `-2`, and not interpreted as the binary arithmetic operator


v1.2.beta3 (2013-02-22)
-----------------------

* issue #427: ArangoDB Importer Manual has no navigation links (previous|home|next)

* issue #319: Documentation missing for Emergency console and incomplete for datafile debugger.

* issue #370: add documentation for reloadRouting and flushServerModules

* issue #393: added REST API for user management at /_api/user

* issue #393, #128: added simple cryptographic functions for user actions in module "crypto":
  * require("org/arangodb/crypto").md5()
  * require("org/arangodb/crypto").sha256()
  * require("org/arangodb/crypto").rand()

* added replaceByExample() Javascript and REST API method

* added updateByExample() Javascript and REST API method

* added optional "limit" parameter for removeByExample() Javascript and REST API method

* fixed issue #413

* updated bundled V8 version from 3.9.4 to 3.16.14.1
  Note: the Windows version used a more recent version (3.14.0.1) and was not updated.

* fixed issue #404: keep original request url in request object


v1.2.beta2 (2013-02-15)
-----------------------

* fixed issue #405: 1.2 compile warnings

* fixed issue #333: [debian] Group "arangodb" is not used when starting vie init.d script

* added optional parameter 'excludeSystem' to GET /_api/collection
  This parameter can be used to disable returning system collections in the list
  of all collections.

* added AQL functions KEEP() and UNSET()

* fixed issue #348: "HTTP Interface for Administration and Monitoring"
  documentation errors.

* fix stringification of specific positive int64 values. Stringification of int64
  values with the upper 32 bits cleared and the 33rd bit set were broken.

* issue #395:  Collection properties() function should return 'isSystem' for
  Javascript and REST API

* make server stop after upgrade procedure when invoked with `--upgrade option`.
  When started with the `--upgrade` option, the server will perfom
  the upgrade, and then exit with a status code indicating the result of the
  upgrade (0 = success, 1 = failure). To start the server regularly in either
  daemon or console mode, the `--upgrade` option must not be specified.
  This change was introduced to allow init.d scripts check the result of
  the upgrade procedure, even in case an upgrade was successful.
  this was introduced as part of issue #391.

* added AQL function EDGES()

* added more crash-protection when reading corrupted collections at startup

* added documentation for AQL function CONTAINS()

* added AQL function LIKE()

* replaced redundant error return code 1520 (Unable to open collection) with error code
  1203 (Collection not found). These error codes have the same meanings, but one of
  them was returned from AQL queries only, the other got thrown by other parts of
  ArangoDB. Now, error 1203 (Collection not found) is used in AQL too in case a
  non-existing collection is used.

v1.2.beta1 (2013-02-01)
-----------------------

* fixed issue #382: [Documentation error] Maschine... should be Machine...

* unified history file locations for arangod, arangosh, and arangoirb.
  - The readline history for arangod (emergency console) is now stored in file
    $HOME/.arangod. It was stored in $HOME/.arango before.
  - The readline history for arangosh is still stored in $HOME/.arangosh.
  - The readline history for arangoirb is now stored in $HOME/.arangoirb. It was
    stored in $HOME/.arango-mrb before.

* fixed issue #381: _users user should have a unique constraint

* allow negative list indexes in AQL to access elements from the end of a list,
  e.g. ```RETURN values[-1]``` will return the last element of the `values` list.

* collection ids, index ids, cursor ids, and document revision ids created and
  returned by ArangoDB are now returned as strings with numeric content inside.
  This is done to prevent some value overrun/truncation in any part of the
  complete client/server workflow.
  In ArangoDB 1.1 and before, these values were previously returned as
  (potentially very big) integer values. This may cause problems (clipping, overrun,
  precision loss) for clients that do not support big integers natively and store
  such values in IEEE754 doubles internally. This type loses precision after about
  52 bits and is thus not safe to hold an id.
  Javascript and 32 bit-PHP are examples for clients that may cause such problems.
  Therefore, ids are now returned by ArangoDB as strings, with the string
  content being the integer value as before.

  Example for documents ("_rev" attribute):
  - Document returned by ArangoDB 1.1: { "_rev": 1234, ... }
  - Document returned by ArangoDB 1.2: { "_rev": "1234", ... }

  Example for collections ("id" attribute / "_id" property):
  - Collection returned by ArangoDB 1.1: { "id": 9327643, "name": "test", ... }
  - Collection returned by ArangoDB 1.2: { "id": "9327643", "name": "test", ... }

  Example for cursors ("id" attribute):
  - Collection returned by ArangoDB 1.1: { "id": 11734292, "hasMore": true, ... }
  - Collection returned by ArangoDB 1.2: { "id": "11734292", "hasMore": true, ... }

* global variables are not automatically available anymore when starting the
  arangod Javascript emergency console (i.e. ```arangod --console```).

  Especially, the variables `db`, `edges`, and `internal` are not available
  anymore. `db` and `internal` can be made available in 1.2 by
  ```var db = require("org/arangodb").db;``` and
  ```var internal = require("internal");```, respectively.
  The reason for this change is to get rid of global variables in the server
  because this will allow more specific inclusion of functionality.

  For convenience, the global variable `db` is still available by default in
  arangosh. The global variable `edges`, which since ArangoDB 1.1 was kind of
  a redundant wrapper of `db`, has been removed in 1.2 completely.
  Please use `db` instead, and if creating an edge collection, use the explicit
  ```db._createEdgeCollection()``` command.

* issue #374: prevent endless redirects when calling admin interface with
  unexpected URLs

* issue #373: TRAVERSAL() `trackPaths` option does not work. Instead `paths` does work

* issue #358: added support for CORS

* honor optional waitForSync property for document removal, replace, update, and
  save operations in arangosh. The waitForSync parameter for these operations
  was previously honored by the REST API and on the server-side, but not when
  the waitForSync parameter was specified for a document operation in arangosh.

* calls to db.collection.figures() and /_api/collection/<collection>/figures now
  additionally return the number of shapes used in the collection in the
  extra attribute "shapes.count"

* added AQL TRAVERSAL_TREE() function to return a hierarchical result from a traversal

* added AQL TRAVERSAL() function to return the results from a traversal

* added AQL function ATTRIBUTES() to return the attribute names of a document

* removed internal server-side AQL functions from global scope.

  Now the AQL internal functions can only be accessed via the exports of the
  ahuacatl module, which can be included via ```require("org/arangodb/ahuacatl")```.
  It shouldn't be necessary for clients to access this module at all, but
  internal code may use this module.

  The previously global AQL-related server-side functions were moved to the
  internal namespace. This produced the following function name changes on
  the server:

     old name              new name
     ------------------------------------------------------
     AHUACATL_RUN       => require("internal").AQL_QUERY
     AHUACATL_EXPLAIN   => require("internal").AQL_EXPLAIN
     AHUACATL_PARSE     => require("internal").AQL_PARSE

  Again, clients shouldn't have used these functions at all as there is the
  ArangoStatement object to execute AQL queries.

* fixed issue #366: Edges index returns strange description

* added AQL function MATCHES() to check a document against a list of examples

* added documentation and tests for db.collection.removeByExample

* added --progress option for arangoimp. This will show the percentage of the input
  file that has been processed by arangoimp while the import is still running. It can
  be used as a rough indicator of progress for the entire import.

* make the server log documents that cannot be imported via /_api/import into the
  logfile using the warning log level. This may help finding illegal documents in big
  import runs.

* check on server startup whether the database directory and all collection directories
  are writable. if not, the server startup will be aborted. this prevents serious
  problems with collections being non-writable and this being detected at some pointer
  after the server has been started

* allow the following AQL constructs: FUNC(...)[...], FUNC(...).attribute

* fixed issue #361: Bug in Admin Interface. Header disappears when clicking new collection

* Added in-memory only collections

  Added collection creation parameter "isVolatile":
  if set to true, the collection is created as an in-memory only collection,
  meaning that all document data of that collection will reside in memory only,
  and will not be stored permanently to disk.
  This means that all collection data will be lost when the collection is unloaded
  or the server is shut down.
  As this collection type does not have datafile disk overhead for the regular
  document operations, it may be faster than normal disk-backed collections. The
  actual performance gains strongly depend on the underlying OS, filesystem, and
  settings though.
  This collection type should be used for caches only and not for any sensible data
  that cannot be re-created otherwise.
  Some platforms, namely Windows, currently do not support this collection type.
  When creating an in-memory collection on such platform, an error message will be
  returned by ArangoDB telling the user the platform does not support it.

  Note: in-memory collections are an experimental feature. The feature might
  change drastically or even be removed altogether in a future version of ArangoDB.

* fixed issue #353: Please include "pretty print" in Emergency Console

* fixed issue #352: "pretty print" console.log
  This was achieved by adding the dump() function for the "internal" object

* reduced insertion time for edges index
  Inserting into the edges index now avoids costly comparisons in case of a hash
  collision, reducing the prefilling/loading timer for bigger edge collections

* added fulltext queries to AQL via FULLTEXT() function. This allows search
  fulltext indexes from an AQL query to find matching documents

* added fulltext index type. This index type allows indexing words and prefixes of
  words from a specific document attribute. The index can be queries using a
  SimpleQueryFull object, the HTTP REST API at /_api/simple/fulltext, or via AQL

* added collection.revision() method to determine whether a collection has changed.
  The revision method returns a revision string that can be used by client programs
  for equality/inequality comparisons. The value returned by the revision method
  should be treated by clients as an opaque string and clients should not try to
  figure out the sense of the revision id. This is still useful enough to check
  whether data in a collection has changed.

* issue #346: adaptively determine NUMBER_HEADERS_PER_BLOCK

* issue #338: arangosh cursor positioning problems

* issue #326: use limit optimization with filters

* issue #325: use index to avoid sorting

* issue #324: add limit optimization to AQL

* removed arango-password script and added Javascript functionality to add/delete
  users instead. The functionality is contained in module `users` and can be invoked
  as follows from arangosh and arangod:
  * require("users").save("name", "passwd");
  * require("users").replace("name", "newPasswd");
  * require("users").remove("name");
  * require("users").reload();
  These functions are intentionally not offered via the web interface.
  This also addresses issue #313

* changed print output in arangosh and the web interface for JSON objects.
  Previously, printing a JSON object in arangosh resulted in the attribute values
  being printed as proper JSON, but attribute names were printed unquoted and
  unescaped. This was fine for the purpose of arangosh, but lead to invalid
  JSON being produced. Now, arangosh will produce valid JSON that can be used
  to send it back to ArangoDB or use it with arangoimp etc.

* fixed issue #300: allow importing documents via the REST /_api/import API
  from a JSON list, too.
  So far, the API only supported importing from a format that had one JSON object
  on each line. This is sometimes inconvenient, e.g. when the result of an AQL
  query or any other list is to be imported. This list is a JSON list and does not
  necessary have a document per line if pretty-printed.
  arangoimp now supports the JSON list format, too. However, the format requires
  arangoimp and the server to read the entire dataset at once. If the dataset is
  too big (bigger than --max-upload-size) then the import will be rejected. Even if
  increased, the entire list must fit in memory on both the client and the server,
  and this may be more resource-intensive than importing individual lines in chunks.

* removed unused parameter --reuse-ids for arangoimp. This parameter did not have
  any effect in 1.2, was never publicly announced and did evil (TM) things.

* fixed issue #297 (partly): added whitespace between command line and
  command result in arangosh, added shell colors for better usability

* fixed issue #296: system collections not usable from AQL

* fixed issue #295: deadlock on shutdown

* fixed issue #293: AQL queries should exploit edges index

* fixed issue #292: use index when filtering on _key in AQL

* allow user-definable document keys
  users can now define their own document keys by using the _key attribute
  when creating new documents or edges. Once specified, the value of _key is
  immutable.
  The restrictions for user-defined key values are:
  * the key must be at most 254 bytes long
  * it must consist of the letters a-z (lower or upper case), the digits 0-9,
    the underscore (_) or dash (-) characters only
  * any other characters, especially multi-byte sequences, whitespace or
    punctuation characters cannot be used inside key values

  Specifying a document key is optional when creating new documents. If no
  document key is specified, ArangoDB will create a document key itself.
  There are no guarantees about the format and pattern of auto-generated document
  keys other than the above restrictions.
  Clients should therefore treat auto-generated document keys as opaque values.
  Keys can be used to look up and reference documents, e.g.:
  * saving a document: `db.users.save({ "_key": "fred", ... })`
  * looking up a document: `db.users.document("fred")`
  * referencing other documents: `edges.relations.save("users/fred", "users/john", ...)`

  This change is downwards-compatible to ArangoDB 1.1 because in ArangoDB 1.1
  users were not able to define their own keys. If the user does not supply a _key
  attribute when creating a document, ArangoDB 1.2 will still generate a key of
  its own as ArangoDB 1.1 did. However, all documents returned by ArangoDB 1.2 will
  include a _key attribute and clients should be able to handle that (e.g. by
  ignoring it if not needed). Documents returned will still include the _id attribute
  as in ArangoDB 1.1.

* require collection names everywhere where a collection id was allowed in
  ArangoDB 1.1 & 1.0
  This change requires clients to use a collection name in place of a collection id
  at all places the client deals with collections.
  Examples:
  * creating edges: the _from and _to attributes must now contain collection names instead
    of collection ids: `edges.relations.save("test/my-key1", "test/my-key2", ...)`
  * retrieving edges: the returned _from and _to attributes now will contain collection
    names instead of ids, too: _from: `test/fred` instead of `1234/3455`
  * looking up documents: db.users.document("fred") or db._document("users/fred")

  Collection names must be used in REST API calls instead of collection ids, too.
  This change is thus not completely downwards-compatible to ArangoDB 1.1. ArangoDB 1.1
  required users to use collection ids in many places instead of collection names.
  This was unintuitive and caused overhead in cases when just the collection name was
  known on client-side but not its id. This overhead can now be avoided so clients can
  work with the collection names directly. There is no need to work with collection ids
  on the client side anymore.
  This change will likely require adjustments to API calls issued by clients, and also
  requires a change in how clients handle the _id value of returned documents. Previously,
  the _id value of returned documents contained the collection id, a slash separator and
  the document number. Since 1.2, _id will contain the collection name, a slash separator
  and the document key. The same applies to the _from and _to attribute values of edges
  that are returned by ArangoDB.

  Also removed (now unnecessary) location header in responses of the collections REST API.
  The location header was previously returned because it was necessary for clients.
  When clients created a collection, they specified the collection name. The collection
  id was generated on the server, but the client needed to use the server-generated
  collection id for further API calls, e.g. when creating edges etc. Therefore, the
  full collection URL, also containing the collection id, was returned by the server in
  responses to the collection API, in the HTTP location header.
  Returning the location header has become unnecessary in ArangoDB 1.2 because users
  can access collections by name and do not need to care about collection ids.


v1.1.3 (2013-XX-XX)
-------------------

* fix case when an error message was looked up for an error code but no error
  message was found. In this case a NULL ptr was returned and not checked everywhere.
  The place this error popped up was when inserting into a non-unique hash index
  failed with a specific, invalid error code.

* fixed issue #381:  db._collection("_users").getIndexes();

* fixed issue #379: arango-password fatal issue javscript.startup-directory

* fixed issue #372: Command-Line Options for the Authentication and Authorization


v1.1.2 (2013-01-20)
-------------------

* upgraded to mruby 2013-01-20 583983385b81c21f82704b116eab52d606a609f4

* fixed issue #357: Some spelling and grammar errors

* fixed issue #355: fix quotes in pdf manual

* fixed issue #351: Strange arangosh error message for long running query

* fixed randomly hanging connections in arangosh on MacOS

* added "any" query method: this returns a random document from a collection. It
  is also available via REST HTTP at /_api/simple/any.

* added deployment tool

* added getPeerVertex

* small fix for logging of long messages: the last character of log messages longer
  than 256 bytes was not logged.

* fixed truncation of human-readable log messages for web interface: the trailing \0
  byte was not appended for messages longer than 256 bytes

* fixed issue #341: ArangoDB crashes when stressed with Batch jobs
  Contrary to the issue title, this did not have anything to do with batch jobs but
  with too high memory usage. The memory usage of ArangoDB is now reduced for cases
   when there are lots of small collections with few documents each

* started with issue #317: Feature Request (from Google Groups): DATE handling

* backported issue #300: Extend arangoImp to Allow importing result set-like
  (list of documents) formatted files

* fixed issue #337: "WaitForSync" on new collection does not work on Win/X64

* fixed issue #336: Collections REST API docs

* fixed issue #335: mmap errors due to wrong memory address calculation

* fixed issue #332: arangoimp --use-ids parameter seems to have no impact

* added option '--server.disable-authentication' for arangosh as well. No more passwd
  prompts if not needed

* fixed issue #330: session logging for arangosh

* fixed issue #329: Allow passing script file(s) as parameters for arangosh to run

* fixed issue #328: 1.1 compile warnings

* fixed issue #327: Javascript parse errors in front end


v1.1.1 (2012-12-18)
-------------------

* fixed issue #339: DELETE /_api/cursor/cursor-identifier return incollect errorNum

  The fix for this has led to a signature change of the function actions.resultNotFound().
  The meaning of parameter #3 for This function has changed from the error message string
  to the error code. The error message string is now parameter #4.
  Any client code that uses this function in custom actions must be adjusted.

* fixed issue #321: Problem upgrading arangodb 1.0.4 to 1.1.0 with Homebrew (OSX 10.8.2)

* fixed issue #230: add navigation and search for online documentation

* fixed issue #315: Strange result in PATH

* fixed issue #323: Wrong function returned in error message of AQL CHAR_LENGTH()

* fixed some log errors on startup / shutdown due to pid file handling and changing
  of directories


v1.1.0 (2012-12-05)
-------------------

* WARNING:
  arangod now performs a database version check at startup. It will look for a file
  named "VERSION" in its database directory. If the file is not present, arangod will
  perform an automatic upgrade of the database directory. This should be the normal
  case when upgrading from ArangoDB 1.0 to ArangoDB 1.1.

  If the VERSION file is present but is from an older version of ArangoDB, arangod
  will refuse to start and ask the user to run a manual upgrade first. A manual upgrade
  can be performed by starting arangod with the option `--upgrade`.

  This upgrade procedure shall ensure that users have full control over when they
  perform any updates/upgrades of their data, and can plan backups accordingly. The
  procedure also guarantees that the server is not run without any required system
  collections or with in incompatible data state.

* added AQL function DOCUMENT() to retrieve a document by its _id value

* fixed issue #311: fixed segfault on unload

* fixed issue #309: renamed stub "import" button from web interface

* fixed issue #307: added WaitForSync column in collections list in in web interface

* fixed issue #306: naming in web interface

* fixed issue #304: do not clear AQL query text input when switching tabs in
  web interface

* fixed issue #303: added documentation about usage of var keyword in web interface

* fixed issue #301: PATCH does not work in web interface

# fixed issue #269: fix make distclean & clean

* fixed issue #296: system collections not usable from AQL

* fixed issue #295: deadlock on shutdown

* added collection type label to web interface

* fixed issue #290: the web interface now disallows creating non-edges in edge collections
  when creating collections via the web interface, the collection type must also be
  specified (default is document collection)

* fixed issue #289: tab-completion does not insert any spaces

* fixed issue #282: fix escaping in web interface

* made AQL function NOT_NULL take any number of arguments. Will now return its
  first argument that is not null, or null if all arguments are null. This is downwards
  compatible.

* changed misleading AQL function name NOT_LIST() to FIRST_LIST() and slightly changed
  the behavior. The function will now return its first argument that is a list, or null
  if none of the arguments are lists.
  This is mostly downwards-compatible. The only change to the previous implementation in
  1.1-beta will happen if two arguments were passed and the 1st and 2nd arguments were
  both no lists. In previous 1.1, the 2nd argument was returned as is, but now null
  will be returned.

* add AQL function FIRST_DOCUMENT(), with same behavior as FIRST_LIST(), but working
  with documents instead of lists.

* added UPGRADING help text

* fixed issue #284: fixed Javascript errors when adding edges/vertices without own
  attributes

* fixed issue #283: AQL LENGTH() now works on documents, too

* fixed issue #281: documentation for skip lists shows wrong example

* fixed AQL optimizer bug, related to OR-combined conditions that filtered on the
  same attribute but with different conditions

* fixed issue #277: allow usage of collection names when creating edges
  the fix of this issue also implies validation of collection names / ids passed to
  the REST edge create method. edges with invalid collection ids or names in the
  "from" or "to" values will be rejected and not saved


v1.1.beta2 (2012-11-13)
-----------------------

* fixed arangoirb compilation

* fixed doxygen


v1.1.beta1 (2012-10-24)
-----------------------

* fixed AQL optimizer bug

* WARNING:
  - the user has changed from "arango" to "arangodb", the start script has changed from
    "arangod" to "arangodb", the database directory has changed from "/var/arangodb" to
    "/var/lib/arangodb" to be compliant with various Linux policies

  - In 1.1, we have introduced types for collections: regular documents go into document
    collections, and edges go into edge collections. The prefixing (db.xxx vs. edges.xxx)
    works slightly different in 1.1: edges.xxx can still be used to access collections,
    however, it will not determine the type of existing collections anymore. To create an
    edge collection 1.1, you can use db._createEdgeCollection() or edges._create().
    And there's of course also db._createDocumentCollection().
    db._create() is also still there and will create a document collection by default,
    whereas edges._create() will create an edge collection.

  - the admin web interface that was previously available via the simple URL suffix /
    is now available via a dedicated URL suffix only: /_admin/html
    The reason for this is that routing and URLs are now subject to changes by the end user,
    and only URLs parts prefixed with underscores (e.g. /_admin or /_api) are reserved
    for ArangoDB's internal usage.

* the server now handles requests with invalid Content-Length header values as follows:
  - if Content-Length is negative, the server will respond instantly with HTTP 411
    (length required)

  - if Content-Length is positive but shorter than the supplied body, the server will
    respond with HTTP 400 (bad request)

  - if Content-Length is positive but longer than the supplied body, the server will
    wait for the client to send the missing bytes. The server allows 90 seconds for this
    and will close the connection if the client does not send the remaining data

  - if Content-Length is bigger than the maximum allowed size (512 MB), the server will
    fail with HTTP 413 (request entity too large).

  - if the length of the HTTP headers is greater than the maximum allowed size (1 MB),
    the server will fail with HTTP 431 (request header fields too large)

* issue #265: allow optional base64 encoding/decoding of action response data

* issue #252: create _modules collection using arango-upgrade (note: arango-upgrade was
  finally replaced by the `--upgrade` option for arangod)

* issue #251: allow passing arbitrary options to V8 engine using new command line option:
  --javascript.v8-options. Using this option, the Harmony features or other settings in
  v8 can be enabled if the end user requires them

* issue #248: allow AQL optimizer to pull out completely uncorrelated subqueries to the
  top level, resulting in less repeated evaluation of the subquery

* upgraded to Doxygen 1.8.0

* issue #247: added AQL function MERGE_RECURSIVE

* issue #246: added clear() function in arangosh

* issue #245: Documentation: Central place for naming rules/limits inside ArangoDB

* reduced size of hash index elements by 50 %, allowing more index elements to fit in
  memory

* issue #235: GUI Shell throws Error:ReferenceError: db is not defined

* issue #229: methods marked as "under construction"

* issue #228: remove unfinished APIs (/_admin/config/*)

* having the OpenSSL library installed is now a prerequisite to compiling ArangoDB
  Also removed the --enable-ssl configure option because ssl is always required.

* added AQL functions TO_LIST, NOT_LIST

* issue #224: add optional Content-Id for batch requests

* issue #221: more documentation on AQL explain functionality. Also added
  ArangoStatement.explain() client method

* added db._createStatement() method on server as well (was previously available
  on the client only)

* issue #219: continue in case of "document not found" error in PATHS() function

* issue #213: make waitForSync overridable on specific actions

* changed AQL optimizer to use indexes in more cases. Previously, indexes might
  not have been used when in a reference expression the inner collection was
  specified last. Example: FOR u1 IN users FOR u2 IN users FILTER u1._id == u2._id
  Previously, this only checked whether an index could be used for u2._id (not
  possible). It was not checked whether an index on u1._id could be used (possible).
  Now, for expressions that have references/attribute names on both sides of the
  above as above, indexes are checked for both sides.

* issue #204: extend the CSV import by TSV and by user configurable
  separator character(s)

* issue #180: added support for batch operations

* added startup option --server.backlog-size
  this allows setting the value of the backlog for the listen() system call.
  the default value is 10, the maximum value is platform-dependent

* introduced new configure option "--enable-maintainer-mode" for
  ArangoDB maintainers. this option replaces the previous compile switches
  --with-boost-test, --enable-bison, --enable-flex and --enable-errors-dependency
  the individual configure options have been removed. --enable-maintainer-mode
  turns them all on.

* removed potentially unused configure option --enable-memfail

* fixed issue #197: HTML web interface calls /_admin/user-manager/session

* fixed issue #195: VERSION file in database directory

* fixed issue #193: REST API HEAD request returns a message body on 404

* fixed issue #188: intermittent issues with 1.0.0
  (server-side cursors not cleaned up in all cases, pthreads deadlock issue)

* issue #189: key store should use ISO datetime format bug

* issue #187: run arango-upgrade on server start (note: arango-upgrade was finally
  replaced by the `--upgrade` option for arangod)n

* fixed issue #183: strange unittest error

* fixed issue #182: manual pages

* fixed issue #181: use getaddrinfo

* moved default database directory to "/var/lib/arangodb" in accordance with
  http://www.pathname.com/fhs/pub/fhs-2.3.html

* fixed issue #179: strange text in import manual

* fixed issue #178: test for aragoimp is missing

* fixed issue #177: a misleading error message was returned if unknown variables
  were used in certain positions in an AQL query.

* fixed issue #176: explain how to use AQL from the arangosh

* issue #175: re-added hidden (and deprecated) option --server.http-port. This
  option is only there to be downwards-compatible to Arango 1.0.

* fixed issue #174: missing Documentation for `within`

* fixed issue #170: add db.<coll_name>.all().toArray() to arangosh help screen

* fixed issue #169: missing argument in Simple Queries

* added program arango-upgrade. This program must be run after installing ArangoDB
  and after upgrading from a previous version of ArangoDB. The arango-upgrade script
  will ensure all system collections are created and present in the correct state.
  It will also perform any necessary data updates.
  Note: arango-upgrade was finally replaced by the `--upgrade` option for arangod.

* issue #153: edge collection should be a flag for a collection
  collections now have a type so that the distinction between document and edge
  collections can now be done at runtime using a collection's type value.
  A collection's type can be queried in Javascript using the <collection>.type() method.

  When new collections are created using db._create(), they will be document
  collections by default. When edge._create() is called, an edge collection will be created.
  To explicitly create a collection of a specific/different type, use the methods
  _createDocumentCollection() or _createEdgeCollection(), which are available for
  both the db and the edges object.
  The Javascript objects ArangoEdges and ArangoEdgesCollection have been removed
  completely.
  All internal and test code has been adjusted for this, and client code
  that uses edges.* should also still work because edges is still there and creates
  edge collections when _create() is called.

  INCOMPATIBLE CHANGE: Client code might still need to be changed in the following aspect:
  Previously, collections did not have a type so documents and edges could be inserted
  in the same collection. This is now disallowed. Edges can only be inserted into
  edge collections now. As there were no collection types in 1.0, ArangoDB will perform
  an automatic upgrade when migrating from 1.0 to 1.1.
  The automatic upgrade will check every collection and determine its type as follows:
  - if among the first 50 documents in the collection there are documents with
    attributes "_from" and "_to", the collection is typed as an edge collection
  - if among the first 50 documents in the collection there are no documents with
    attributes "_from" and "_to", the collection is made as a document collection

* issue #150: call V8 garbage collection on server periodically

* issue #110: added support for partial updates

  The REST API for documents now offers an HTTP PATCH method to partially update
  documents. Overwriting/replacing documents is still available via the HTTP PUT method
  as before. The Javascript API in the shell also offers a new update() method in extension to
  the previously existing replace() method.


v1.0.4 (2012-11-12)
-------------------

* issue #275: strange error message in arangosh 1.0.3 at startup


v1.0.3 (2012-11-08)
-------------------

* fixed AQL optimizer bug

* issue #273: fixed segfault in arangosh on HTTP 40x

* issue #265: allow optional base64 encoding/decoding of action response data

* issue #252: _modules collection not created automatically


v1.0.2 (2012-10-22)
-------------------

* repository CentOS-X.Y moved to CentOS-X, same for Debian

* bugfix for rollback from edges

* bugfix for hash indexes

* bugfix for StringBuffer::erase_front

* added autoload for modules

* added AQL function TO_LIST


v1.0.1 (2012-09-30)
-------------------

* draft for issue #165: front-end application howto

* updated mruby to cf8fdea4a6598aa470e698e8cbc9b9b492319d

* fix for issue #190: install doesn't create log directory

* fix for issue #194: potential race condition between creating and dropping collections

* fix for issue #193: REST API HEAD request returns a message body on 404

* fix for issue #188: intermittent issues with 1.0.0

* fix for issue #163: server cannot create collection because of abandoned files

* fix for issue #150: call V8 garbage collection on server periodically


v1.0.0 (2012-08-17)
-------------------

* fix for issue #157: check for readline and ncurses headers, not only libraries


v1.0.beta4 (2012-08-15)
-----------------------

* fix for issue #152: fix memleak for barriers


v1.0.beta3 (2012-08-10)
-----------------------

* fix for issue #151: Memleak, collection data not removed

* fix for issue #149: Inconsistent port for admin interface

* fix for issue #163: server cannot create collection because of abandoned files

* fix for issue #157: check for readline and ncurses headers, not only libraries

* fix for issue #108: db.<collection>.truncate() inefficient

* fix for issue #109: added startup note about cached collection names and how to
  refresh them

* fix for issue #156: fixed memleaks in /_api/import

* fix for issue #59: added tests for /_api/import

* modified return value for calls to /_api/import: now, the attribute "empty" is
  returned as well, stating the number of empty lines in the input. Also changed the
  return value of the error code attribute ("errorNum") from 1100 ("corrupted datafile")
  to 400 ("bad request") in case invalid/unexpected JSON data was sent to the server.
  This error code is more appropriate as no datafile is broken but just input data is
  incorrect.

* fix for issue #152: Memleak for barriers

* fix for issue #151: Memleak, collection data not removed

* value of --database.maximal-journal-size parameter is now validated on startup. If
  value is smaller than the minimum value (currently 1048576), an error is thrown and
  the server will not start. Before this change, the global value of maximal journal
  size was not validated at server start, but only on collection level

* increased sleep value in statistics creation loop from 10 to 500 microseconds. This
  reduces accuracy of statistics values somewhere after the decimal points but saves
  CPU time.

* avoid additional sync() calls when writing partial shape data (attribute name data)
  to disk. sync() will still be called when the shape marker (will be written after
  the attributes) is written to disk

* issue #147: added flag --database.force-sync-shapes to force synching of shape data
  to disk. The default value is true so it is the same behavior as in version 1.0.
  if set to false, shape data is synched to disk if waitForSync for the collection is
  set to true, otherwise, shape data is not synched.

* fix for issue #145: strange issue on Travis: added epsilon for numeric comparison in
  geo index

* fix for issue #136: adjusted message during indexing

* issue #131: added timeout for HTTP keep-alive connections. The default value is 300
  seconds. There is a startup parameter server.keep-alive-timeout to configure the value.
  Setting it to 0 will disable keep-alive entirely on the server.

* fix for issue #137: AQL optimizer should use indexes for ref accesses with
  2 named attributes


v1.0.beta2 (2012-08-03)
-----------------------

* fix for issue #134: improvements for centos RPM

* fixed problem with disable-admin-interface in config file


v1.0.beta1 (2012-07-29)
-----------------------

* fixed issue #118: We need a collection "debugger"

* fixed issue #126: Access-Shaper must be cached

* INCOMPATIBLE CHANGE: renamed parameters "connect-timeout" and "request-timeout"
  for arangosh and arangoimp to "--server.connect-timeout" and "--server.request-timeout"

* INCOMPATIBLE CHANGE: authorization is now required on the server side
  Clients sending requests without HTTP authorization will be rejected with HTTP 401
  To allow backwards compatibility, the server can be started with the option
  "--server.disable-authentication"

* added options "--server.username" and "--server.password" for arangosh and arangoimp
  These parameters must be used to specify the user and password to be used when
  connecting to the server. If no password is given on the command line, arangosh/
  arangoimp will interactively prompt for a password.
  If no user name is specified on the command line, the default user "root" will be
  used.

* added startup option "--server.ssl-cipher-list" to determine which ciphers to
  use in SSL context. also added SSL_OP_CIPHER_SERVER_PREFERENCE to SSL default
  options so ciphers are tried in server and not in client order

* changed default SSL protocol to TLSv1 instead of SSLv2

* changed log-level of SSL-related messages

* added SSL connections if server is compiled with OpenSSL support. Use --help-ssl

* INCOMPATIBLE CHANGE: removed startup option "--server.admin-port".
  The new endpoints feature (see --server.endpoint) allows opening multiple endpoints
  anyway, and the distinction between admin and "other" endpoints can be emulated
  later using privileges.

* INCOMPATIBLE CHANGE: removed startup options "--port", "--server.port", and
  "--server.http-port" for arangod.
  These options have been replaced by the new "--server.endpoint" parameter

* INCOMPATIBLE CHANGE: removed startup option "--server" for arangosh and arangoimp.
  These options have been replaced by the new "--server.endpoint" parameter

* Added "--server.endpoint" option to arangod, arangosh, and arangoimp.
  For arangod, this option allows specifying the bind endpoints for the server
  The server can be bound to one or multiple endpoints at once. For arangosh
  and arangoimp, the option specifies the server endpoint to connect to.
  The following endpoint syntax is currently supported:
  - tcp://host:port or http@tcp://host:port (HTTP over IPv4)
  - tcp://[host]:port or http@tcp://[host]:port (HTTP over IPv6)
  - ssl://host:port or http@tcp://host:port (HTTP over SSL-encrypted IPv4)
  - ssl://[host]:port or http@tcp://[host]:port (HTTP over SSL-encrypted IPv6)
  - unix:///path/to/socket or http@unix:///path/to/socket (HTTP over UNIX socket)

  If no port is specified, the default port of 8529 will be used.

* INCOMPATIBLE CHANGE: removed startup options "--server.require-keep-alive" and
  "--server.secure-require-keep-alive".
  The server will now behave as follows which should be more conforming to the
  HTTP standard:
  * if a client sends a "Connection: close" header, the server will close the
    connection
  * if a client sends a "Connection: keep-alive" header, the server will not
    close the connection
  * if a client does not send any "Connection" header, the server will assume
    "keep-alive" if the request was an HTTP/1.1 request, and "close" if the
    request was an HTTP/1.0 request

* (minimal) internal optimizations for HTTP request parsing and response header
  handling

* fixed Unicode unescaping bugs for \f and surrogate pairs in BasicsC/strings.c

* changed implementation of TRI_BlockCrc32 algorithm to use 8 bytes at a time

* fixed issue #122: arangod doesn't start if <log.file> cannot be created

* fixed issue #121: wrong collection size reported

* fixed issue #98: Unable to change journalSize

* fixed issue #88: fds not closed

* fixed escaping of document data in HTML admin front end

* added HTTP basic authentication, this is always turned on

* added server startup option --server.disable-admin-interface to turn off the
  HTML admin interface

* honor server startup option --database.maximal-journal-size when creating new
  collections without specific journalsize setting. Previously, these
  collections were always created with journal file sizes of 32 MB and the
  --database.maximal-journal-size setting was ignored

* added server startup option --database.wait-for-sync to control the default
  behavior

* renamed "--unit-tests" to "--javascript.unit-tests"


v1.0.alpha3 (2012-06-30)
------------------------

* fixed issue #116: createCollection=create option doesn't work

* fixed issue #115: Compilation issue under OSX 10.7 Lion & 10.8 Mountain Lion
  (homebrew)

* fixed issue #114: image not found

* fixed issue #111: crash during "make unittests"

* fixed issue #104: client.js -> ARANGO_QUIET is not defined


v1.0.alpha2 (2012-06-24)
------------------------

* fixed issue #112: do not accept document with duplicate attribute names

* fixed issue #103: Should we cleanup the directory structure

* fixed issue #100: "count" attribute exists in cursor response with "count:
  false"

* fixed issue #84 explain command

* added new MRuby version (2012-06-02)

* added --log.filter

* cleanup of command line options:
** --startup.directory => --javascript.startup-directory
** --quite => --quiet
** --gc.interval => --javascript.gc-interval
** --startup.modules-path => --javascript.modules-path
** --action.system-directory => --javascript.action-directory
** --javascript.action-threads => removed (is now the same pool as --server.threads)

* various bug-fixes

* support for import

* added option SKIP_RANGES=1 for make unittests

* fixed several range-related assertion failures in the AQL query optimizer

* fixed AQL query optimizations for some edge cases (e.g. nested subqueries with
  invalid constant filter expressions)


v1.0.alpha1 (2012-05-28)
------------------------

Alpha Release of ArangoDB 1.0