This file describes changes in recent versions of Slurm. It primarily documents those changes that are of interest to users and administrators. * Changes in Slurm 24.08.0rc1 ============================= * Changes in Slurm 23.11.1 ========================== -- Fix scontrol update job=... TimeLimit+=/-= when used with a raw JobId of job array element. -- Reject TimeLimit increment/decrement when called on job with TimeLimit=UNLIMITED. -- Fix slurmctld segfault when reconfiguring after a job resize. -- Fix compilation on FreeBSD. -- Fix issue with requesting a job with --licenses as well as --tres-per-task=license. -- slurmctld - Prevent segfault in getopt_long() with an invalid long option. -- Switch to man2html-base in Build-Depends for Debian package. -- slurmrestd - Added /meta/slurm/cluster field to responses. -- Adjust systemd service files to start daemons after remote-fs.target. -- Add "--with selinux" option to slurm.spec. -- Fix task/cgroup indexing tasks in cgroup plugins, which caused jobacct/gather to match the gathered stats with the wrong task id. -- select/linear - Fix regression in 23.11 in which jobs that requested --cpus-per-task were rejected. -- Fix crash in slurmstepd that can occur when launching tasks via mpi using the pmi2 plugin and using the route/topology plugin. -- Fix sgather not gathering from all nodes when using CR_PACK_NODES/--m pack. -- Fix mysql query syntax error when getting jobs with private data. -- Fix sanity check to prevent deleting default account of users. -- data_parser/v0.0.40 - Fix the parsing for /slurmdb/v0.0.40/jobs exit_code query parameter. -- Fix issue where TRES for energy wasn't always set before sending it to the jobcomp plugin. -- jobcomp/[kafka|elastisearch] Print raw TRES values along with the formatted versions as tres_[req|alloc]_raw. -- Fix inconsistencies with --cpu-bind/SLURM_CPU_BIND and --hint/SLURM_HINT. -- Fix ignoring invalid json in various subsystems. -- Remove shebang from bash completion script. -- Fix elapsed time in JobComp being set from invalid start and end times. -- Update service files to start slurmd, slurmctld, and slurmdbd after sssd. -- data_parser/v0.0.40 - Fix output of DefMemPerCpu, MaxMemPerCpu, and max_shares. -- When determining a jobs index in the database don't wait if there are more jobs waiting. -- If a job requests more shards which would allocate more than one sharing GRES (gpu) per node refuse it unless SelectTypeparameters has MULTIPLE_SHARING_GRES_PJ. -- Avoid refreshing the hwloc xml file when slurmd is reconfigured. This fixes an issue seen with CoreSpecCount used on nodes with Intel E-cores. -- Trigger fatal exit when Slurm API function is called before slurm_init() is called. -- slurmd - Fix issue with 'scontrol reconfigure' when started with '-c'. -- data_parser/v0.0.40 - Fix handling of negative job nice values. -- data_parser/v0.0.40 - Fill the "id" object for associations with the cluster, account, partition, and user in addition to the assoc id. -- data_parser/v0.0.40 - Remove unusable cpu_binding_flags enums from v00.0.40_job_desc_msg. -- Improve performance and resiliency of slurmscriptd shutdown on 'scontrol reconfigure'. -- slurmrestd - Job submissions that result in the following error codes will be considered as successfully submitted (with a warning), instead of returning an HTTP 500 error back: ESLURM_NODES_BUSY, ESLURM_RESERVATION_BUSY, ESLURM_JOB_HELD, ESLURM_NODE_NOT_AVAIL, ESLURM_QOS_THRES, ESLURM_ACCOUNTING_POLICY, ESLURM_RESERVATION_NOT_USABLE, ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE, ESLURM_BURST_BUFFER_WAIT, ESLURM_PARTITION_DOWN, ESLURM_LICENSES_UNAVAILABLE. -- Fix issue with node appearing to reboot on every "scontrol reconfigure" when slurmd was started with the '-b' flag. -- Fix a slurmctld fatal error when upgrading to 23.11 and changing from select/cons_res to select/cons_tres at the same time. -- slurmctld - Fix subsequent reconfigure hanging after a failed reconfigure. -- slurmctld - Reject arbitrary distribution jobs that have a minimum node count that differs from the number of unique nodes in the hostlist. -- Prevent slurmdbd errors when updating reservations with names containing apostrophes. * Changes in Slurm 23.11.0 ========================== -- For jobs that request --cpus-per-gpu, ensure that the --cpus-per-gpu request is honored on every node in the and not just for the job as a whole. -- Fix "srun -Z" for cred/munge and cred/none. -- Fix listing available data_parser plugins for json and yaml when giving no commands to scontrol or sacctmgr. -- data_parser/v0.0.40 - Fixed how deleted QOS and associations for jobs are dumped. -- data_parser/v0.0.40 - Fix how errors and warnings are dumped. -- Print consistent errors when serializer plugin fails to load. -- data_parser/v0.0.40 - Fix parsing of flag arrays to allow muliple flags to be set. -- slurmctld - Rework 'scontrol reconfigure' to avoid race conditions that can result in stray jobs. -- slurmctld - Shave ~1 second off average reconfigure time by terminating internal processing threads faster. -- Skip running slurmdbd -R if the connected cluster is 23.11 or newer. This operation is nolonger relevant for 23.11. -- Fix segfault when updating node instance id/type without accounting enabled. -- Ensure slurmscriptd shuts down before slurmctld is stopped / reconfigured. -- Improve error handling and error messages in slurmctld to slurmscriptd communications. This includes avoiding potential deadlock in slurmctld if slurmscript dies unexpectedly. -- Do not hold batch jobs whose extra constraints cannot be immediately satisfied, and set the state reason to "Constraints" instead of "BadConstraints". -- Fix verbose log message printing a hex number instead of a job id. -- Upgrade rate limit parameters message from debug to info. -- Fix missing symbols for those linking to libslurm. -- Fix memory leak when getting and forwarding addrs from client. -- Fix xassert when forwarding to non-addressable nodes. -- Fix regression in 23.11.0rc1 where assocs were created with an incorrect hierarchy for non-23.11 clusters. -- For SchedulerParameters=extra_constraints, prevent slurmctld segfault when starting a slurmd with --extra for a node that did not previously set this. This also ensures the extra constraints model works off the current node state, not the prior state. -- Fix regression in 23.11.0rc1 where data_t would not decode negative float values correctly, instead the absolute value was always returned. -- Fix issue where 'scontrol reconfigure' right as the controller was started or reconfigured could lead to it shutting down completely. -- Fix --tres-per-task assertion. -- Fix a few issues when creating reservations. -- Fix slurmctld segfault when packing a job step loaded from a < 23.11 state. -- Fix a 32-bit compile issue. -- Add SchedulerParameters=time_min_as_soft_limit option. * Changes in Slurm 23.11.0rc1 ============================= -- task/affinity - remove Power7 cpu-specific workarounds. -- Remove SLURM_WORKING_CLUSTER env from batch and srun environments. -- cli_filter/lua - return nil for unset time options rather than the string "2982616-04:14:00" (which is the internal macro "NO_VAL" represented as time string). -- Remove 'none' plugins for all but auth and cred. scontrol show config will report (null) now. -- Removed select/cons_res. Please update your configuration to select/cons_tres. -- mpi/pmix - When aborted with status 0, avoid marking job/step as failed. -- Fixed typo on "initialized" for the description of ESLURM_PLUGIN_NOT_LOADED. -- Added max_submit_line_size to SchedulerParameters. -- Change TreeWidth default from 50 to 16. -- cgroup.conf - Removed deprecated parameters AllowedKmemSpace, ConstrainKmemSpace, MaxKmemPercent, and MinKmemSpace. -- proctrack/cgroup - Add "SignalChildrenProcesses=" option to cgroup.conf. This allows signals for cancelling, suspending, resuming, etc. to be sent to children processes in a step/job rather than just the parent. -- Add PreemptParameters=suspend_grace_time parameter to control amount of time between SIGTSTP and SIGSTOP signals when suspending jobs. -- job_submit/throttle - improve reset of submitted job counts per user in order to better honor SchedulerParameters=jobs_per_user_per_hour=#. -- Load the user environment into a private pid namespace to avoid user scripts leaving background processes on a node. -- scontrol show assoc_mgr will display Lineage instead of Lft for associations. -- Add SlurmctldParameters=no_quick_restart to avoid a new slurmctld taking over the old slurmctld on accedent. -- Fix --cpus-per-gpu for step allocations, which was previously ignored for job steps. --cpus-per-gpu implies --exact. -- Fix mutual exclusivity of --cpus-per-gpu and --cpus-per-task: fatal if both options are requested in the commandline or both are requested in the environment. If one option is requested in the command line, it will override the other option in the environment. -- slurmrestd - openapi/dbv0.0.37 and openapi/v0.0.37 plugins have been removed. -- slurmrestd - openapi/dbv0.0.38 and openapi/v0.0.38 plugins have been tagged as deprecated. -- openapi/slurmctld - forked from openapi/v0.0.38. -- openapi/slurmdbd - forked from openapi/dbv0.0.38. -- data_parser/v0.0.40 - forked from data_parser/v0.0.39 plugin. -- data_parser/v0.0.40 - added OpenAPI schema generation of path parameters and OperationIds. -- slurmrestd - added auto population of info/version field. -- openapi/slurmctld - convert to using data_parser plugins for all input and output formatting. -- openapi/slurmdbd - convert to using data_parser plugins for all input and output formatting. -- data_parser/v0.0.39 - skip empty string when parsing QOS ids. -- data_parser/v0.0.40 - skip empty string when parsing QOS ids. -- data_parser/v0.0.40 - log errors on every level of parsing on failure. -- sdiag - add --yaml and --json arg support to specify data_parser plugin. -- sacct - add --yaml and --json arg support to specify data_parser plugin. -- scontrol - add --yaml and --json arg support to specify data_parser plugin. -- sinfo - add --yaml and --json arg support to specify data_parser plugin. -- squeue - add --yaml and --json arg support to specify data_parser plugin. -- data_parser/v0.0.40 - add warnings for unknown fields during parsing. -- data_parser/v0.0.40 - add FAST parameter to allow requestor to skip more time intensive warning checks. -- Changed the default SelectType to select/cons_tres (from select/linear). -- Allow SlurmUser/root to use reservations without specific permissions. -- Fix sending step signals to nodes not allocated by the step. -- Remove CgroupAutomount= option from cgroup.conf. -- Add TopologyRoute=RoutePart to route communications based on partition node lists. -- SPANK - added new spank_prepend_task_argv() function. -- slurmd - improve error logging at job startup during transition from slurmd to slurmstepd. -- Added ability for configless to push Prolog and Epilog scripts to slurmds. -- Prolog and Epilog do not have to be fully qualified pathnames. -- Changed default value of PriorityType from priority/basic to priority/multifactor. -- torque/mpiexec - Propogate exit code from launched process. -- slurmrestd - Add new rlimits fields for job submission. -- data_parser/v0.0.40 - convert job state field to flag array to provide enumeration of values. -- sbatch - removed --export-file option (used with defunct Moab integration). -- Define SPANK options environment variables when --export=[NIL|NONE] is specified. -- slurmrestd - Numeric input fields provided with a null formatted value will now convert to zero (0) where it can be a valid value. This is expected to be only be notable with job submission against v0.0.38 versioned endpoints with job requests with fields provided with null values. These fields were already rejected by v0.0.39+ endpoints, unless +complex parser value is provided to v0.0.40+ endpoints. -- slurmrestd - Improve parsing of integers and floating point numbers when handling incoming user provided numeric fields. Fields that would have not rejected a number for a numeric field followed by other non-numeric characters will now get rejected. This is expected to be only be notable with job submission against v0.0.38 versioned endpoints with malformed job requests. -- Reject reservation update if it will result in previously submitted jobs losing access to the reservation. -- data_parser/v0.0.40 - output partition state when dumping partitions. -- Allow for a shared suffix to be used with the hostlist format. E.g., "node[0001-0010]-int". -- Fix perlapi build when using non-default libdir. -- Replace SRUN_CPUS_PER_TASK with SLURM_CPUS_PER_TASK and get back the previous behavior before Slurm 22.05 since now we have the new external launcher step. -- data_parser/v0.0.40 - Change v0.0.40_job_info response to tag exit_code field as verbose job exit code object. -- data_parser/v0.0.40 - Change v0.0.40_job_info response to tag derived_exit_code field as verbose job exit code object. -- Avoid database upgrade failures with galera by enabling streaming replication for Galera 4 clusters during the upgrade process. -- job_container/tmpfs - disable plugin for nodes that are not listed in job_container.conf when there is no global BasePath set. -- job_container/tmpfs - Add "BasePath=none" option to disable plugin on node subsets when there is a global setting. -- Remove cloud_reg_addrs and make it default behavior. -- Remove NoAddrCache CommunicationParameter. -- Add QOS flag 'Relative'. If set the QOS limits will be treated as percentages of a cluster/partition instead of absolutes. -- Remove FIRST_CORES flag from reservations. -- scontrol/sview - Remove comma separated CoreCnt option from reservations. -- scontrol/sview - Remove comma separated NodeCnt option from reservations. -- Add cloud instance id and instance type to node records. Can be viewed/ updated with scontrol. -- slurmd - add "instance-id", "instance-type", and "extra" options to allow them to be set on startup. -- Add cloud instance accounting to database that can be viewed with 'sacctmgr show instance'. -- openapi/v0.0.40 - add /instance and /instances endpoints. -- SelectTypeParameters=cr_cpu - Fix a log error "CPU underflow" at step completion for steps that request --threads-per-core or --hint=nomultithread. -- select/linear - fix task launch failure that sometimes occurred when requesting --threads-per-core or --hint=nomultithread. This also fixes memory calculation with one of these options and --mem-per-cpu: Previously, memory = mem-per-cpu * all cpus including unusable threads. Now, memory = mem-per-cpu * only usuable threads. This behavior matches the documentation and select/cons_tres. -- gpu/nvml - Reduce chances of NVML_ERROR_INSUFFICIENT_SIZE error when getting gpu memory information. -- slurmrestd - Convert to generating OperationIDs based on path for all v0.0.40 tagged paths. -- slurmrestd - Reduce memory used while dumping a job's stdio paths. -- slurmrestd - Jobs queried from data_parser/v0.0.40 from slurmdb will have 'step/id' field given as a string to match CLI formatting instead of an object. -- sacct - Output in JSON or YAML output will will have the 'step/id' field given as a string instead of an object. -- scontrol/squeue - Step output in JSON or YAML output will will have the 'id' field given as a string instead of an object. -- slurmrestd - For 'GET /slurmdb/v0.0.40/jobs' mimick default behavior for handling of job start and end times as sacct when one or both fields are not provided as a query parameter. -- openapi/slurmctld - Add 'GET /slurm/v0.0.40/shares' endpoint to dump same output as sshare. -- sshare - add JSON/YAML support. -- data_parser/v0.0.40 - Remove "required/memory" output in json. It is replaced by "required/memory_per_cpu" and "required/memory_per_node". -- slurmrestd - Add numeric id to all association identifiers to allow unique identification where association has been deleted but is still referenced by accounting record. -- slurmrestd - Add accounting, id, and comment fields to association dumps. -- slurmrestd - Removed usage field from association dumps which was never populated. See accounting field for accounting usage records. -- slurmrestd - Default to not query assocations or coordinators with 'GET /slurmdb/v0.0.40/accounts'. -- slurmrestd - Default to not query assocations, wckeys or coordinators with 'GET /slurmdb/v0.0.40/user'. -- slurmrestd - Enforce user's default wckey on supplied list of user wckeys in 'POST /slurmdb/v0.0.40/user' queries to avoid conflicting or changed default wckey from being ignored. -- slurmrestd - Enforce user's default wckey on supplied list of user wckeys in 'POST /slurmdb/v0.0.40/user' queries to avoid conflicting or changed default wckey from being ignored. -- slurmrestd - 'POST /slurm/v0.0.40/job/submit' will return "step_id" as string to provide descriptive step names (batch, extern, interactive, TBD) for non-numeric steps. -- slurmrestd - Tagged "result" field from 'POST /slurm/v0.0.40/job/submit' as deprecated. -- slurmrestd - Warning will be added for rejected job submissions with submissions to 'POST /slurm/v0.0.40/job/submit'. -- slurmrestd - Tagged "job_id", "step_id", and "job_submit_user_msg" fields from 'POST /slurm/v0.0.40/job/{job_id}' response as deprecated due their only being valid for the first entry in the "result" field array. -- slurmrestd - Warning will be added for rejected job updates with queries to 'POST /slurm/v0.0.40/job/{job_id}'. -- slurmrestd - Add SLURMRESTD_JSON and SLURMRESTD_YAML input environment variables. -- slurmrestd - Correct issue where field and $ref description fields were not getting populated for OpenAPI specification generation for queries to 'GET /openapi/v3' for v0.0.40 endpoints. -- slurmdbd - Check for innodb_redo_log_capacity instead of innodb_log_file_size in MySQL 8.0.30+. -- slurmrestd - Fix log level requests SLURMRESTD_DEBUG and -v applying on top of each other and the default logging level (info). -v now applies on top of the default log level, and SLURMRESTD_DEBUG sets the log level if -v is not given. -- slurmrestd - Allow SLURMRESTD_DEBUG=quiet or 0, which was previously denied. Also deny negative values for SLURMRESTD_DEBUG, which previously set the debug level to debug5. -- The backup slurmctld now checks that the heartbeat file exists in StateSaveLocation before starting and attempting to assert control. This avoids issues with misconfiguration, or the shared filesystem being unavailable, that could previously have lead to all jobs being cancelled. -- The warning printed when using configure --without-PACKAGE has been changed to a notice. -- Fix --cpu-freq with userspace governor and frequency ranges behavior. -- Fix --cpu-freq parsing with incorrect frequencies. -- PMIx support is nolonger built by default. Passing --with-pmix option is now required to build with PMIx. -- Use memory.current in cgroup/v2 instead of manually calculating RSS. This makes accounting consistent with OOM Killer. -- Update slurmstepd processes with current SlurmctldHost settings, allowing for controller changes without draining all compute jobs. -- Add format_stderr to LogTimeFormat of slurm.conf and slurmdbd.conf. -- slurmrestd - add `GET /slurm/v0.0.40/reconfigure` endpoint to allow equivalent requests of `scontrol reconfigure`. -- sreport - cluster Utilization PlannedDown field now includes the time that all nodes were in the POWERED_DOWN state instead of just cloud nodes. -- scontrol update partition now allows Nodes+= and Nodes-= to add/delete nodes from the existing partition node list. Nodes=+host1,-host2 is also allowed. -- sacctmgr - add --yaml and --json arg support to specify data_parser plugin. -- slurmrestd - Add last_update and last_backfill fields to response to `GET /slurm/v0.0.40/job` and `GET /slurm/v0.0.40/jobs` queries. -- slurmrestd - Add last_update fields to response to `GET /slurm/v0.0.40/node` and `GET /slurm/v0.0.40/nodes` queries. -- slurmrestd - Add last_update fields to response to `GET /slurm/v0.0.40/partition` and `GET /slurm/v0.0.40/partitions` queries. -- slurmrestd - Add last_update fields to response to `GET /slurm/v0.0.40/licenses` query. -- auth/jwt - fatal when jwt or jwks key files are writable by other. -- sacctmgr can now modify QOS's RawUsage to zero or a positive value. -- sdiag - Added statistics on why the main and backfill schedulers have stopped evaluation on each scheduling cycle. -- openapi/v0.0.40 - add /{accounts,users}_association endpoints. -- slurm.spec - Add `--with yaml` argument to require YAML support. -- Add new rl_log_freq option to SlurmctldParameters to allow sites to limit the number of 'RPC limit exceeded...' messages that are logged. -- Rename sbcast --fanout to --treewidth. -- Remove SLURM_NODE_ALIASES env variable. -- Enable fanout for dynamic and unaddresable cloud nodes. -- Fix how steps are dealloced in an allocation if the last step of an srun never completes due to a node failure. -- Remove redundant database indexes. -- Add database index to suspend table to speed up archive/purges. -- When requesting --tres-per-task alter incorrect request for TRES, it should be TRESType/TRESName not TRESType:TRESName. -- Make it so reservations can reserve GRES. -- switch/hpe_slingshot - Add disable_rdzv_get flag to disable rendezvous gets. -- slurmrestd - Avoid matching query URLs based on partial matches to an existing endpoint, e.g. POST request to /slurm/v0.0.40/job was routed to the handler of POST /slurm/v0.0.40/job/submit. -- Don't display old job_arrays/het_jobs in sacct if Job ID was reused. -- sbcast - use the specified --fanout value on all hops in message forwarding; previously the specified fanout was only used on the first hop, and additional hops used TreeWidth in slurm.conf. -- slurmrestd - remove logger prefix from '-s/-a list' options outputs. -- Fix fd socket name resolution which could flood log files at debug level. -- The rpmbuild "--with mysql" option has been removed. The rpm has long required sql development libraries to build and the existence of this option was confusing. The default behavior now is to always require one of the sql development libraries. -- Add support for Debian packaging. -- switch/hpe_slingshot - Add support for collectives. -- Nodes with suspended jobs can now be displayed as MIXED. -- sview - Fix search by node state returning incorrect node list. -- Fix inconsistent handling of using cli and/or environment options for tres_per_task=cpu:# and cpus_per_gpu. -- Requesting --cpus-per-task will now set SLURM_TRES_PER_TASK=cpu:# in the environment. -- For some tres related environment variables such as SLURM_TRES_PER_TASK, when srun requests a different value for that option, set these environment variables to the value requested by srun. Previously these environment variables were unchanged from the job allocation. This bug only affected the output environment variables, not the actual step resource allocation. -- RoutePlugin=route/topology has been replaced with TopologyParam=RouteTree. -- If ThreadsPerCore in slurm.conf is configured with less than the number of hardware threads, fix a bug where the task plugins used fewer cores instead of using fewer threads per core. -- Fix arbitrary distribution allowing it to be used with salloc and sbatch and fix how cpus are allocated to nodes. -- Allow nodes to reboot while node is drained or in a maintenance state. -- Allow scontrol reboot to use nodesets to filter nodes to reboot. -- Fix how the topology of typed gres gets updated. -- Changes to the Type option in gres.conf now can be applied with scontrol reconfig. -- Allow for jobs that request a newly configured gres type to be queued even when the needed slurmds have not yet registered. -- Kill recovered jobs that require unconfigured gres types. -- If keepalives are configured, enable them on all persistent connections. -- data_parser/v0.0.40 - add parsers for main/backfill cycle exit reasons. -- Configless - Also send Includes from configuration files not parsed by the controller (i.e. from plugstack.conf). -- Add gpu/nrt plugin for nodes using Trainium/Inferentia devices. -- data_parser/v0.0.40 - Add START_RECEIVED to job flags in dumped output. -- SPANK - Failures from most spank functions (not epilog or exit) will now cause the step to be marked as failed and the command (srun, salloc, sbatch --wait) to return 1. -- data_parser/v0.0.40 - Fix how the "INVALID" nodes state is dumped. -- Add SchedulerParameters=extra_constraints. This enables various node filtering options in the --extra flag of salloc, sbatch, and srun. -- Improve scontrol show node -d information of used shared gres (shard/mps) topology. * Changes in Slurm 23.02.7 ========================== -- libslurm_nss - Avoid causing glibc to assert due to an unexpected return from slurm_nss due to an error during lookup. -- Fix job requests with --tres-per-task sometimes resulting in bad allocations that cannot run subsequent job steps. -- Fix issue with slurmd where srun fails to be warned when a node prolog script runs beyond MsgTimeout set in slurm.conf. -- gres/shard - Fix plugin functions to have matching parameter orders. -- gpu/nvml - Fix issue that resulted in the wrong MIG devices being constrained to a job -- gpu/nvml - Fix linking issue with MIGs that prevented multiple MIGs being used in a single job for certain MIG configurations -- Add JobAcctGatherParams=DisableGPUAcct to disable gpu accounting. -- Fix file descriptor leak in slurmd when using acct_gather_energy/ipmi with DCMI devices. -- sview - avoid crash when job has a node list string > 49 characters. -- Prevent slurmctld crash during reconfigure when packing job start messages. -- Preserve reason uid on reconfig. -- Update node reason with updated INVAL state reason if different from last registration. -- acct_gather_energy/ipmi - Improve logging of DCMI issues. -- conmgr - Avoid NULL dereference when using auth/none. -- data_parser/v0.0.39 - Fixed how deleted QOS and associations for jobs are dumped. -- burst_buffer/lua - fix stage in counter not decrementing when a job is cancelled during stage in. This counter is used to enforce the limit of 128 scripts per stage. -- gpu/oneapi - Add support for new env vars ZE_FLAT_DEVICE_HIERARCHY and ZE_ENABLE_PCI_ID_DEVICE_ORDER. -- data_parser/v0.0.39 - Fix how the "INVALID" nodes state is dumped. -- data_parser/v0.0.39 - Fix parsing of flag arrays to allow muliple flags to be set. -- Avoid leaking sockets when an x11 application is closed in an allocation. -- Fix missing mutex unlock in group cache code which could cause slurmctld to freeze. -- Fix scrontab monthly jobs possibly skipping a month if added near the end of the month. -- Fix loading of the gpu account gather energy plugin. -- Fix slurmctld segfault when reconfiguring after a job resize. -- Fix crash in slurmstepd that can occur when launching tasks via mpi using the pmi2 plugin and using the route/topology plugin. -- data_parser/v0.0.39 - skip empty string when parsing QOS ids. -- Fix "qos doesn't exist" error message in assoc_mgr_update_assocs to print the attempted new default qos, rather than the current default qos. -- Remove error message from assoc_mgr_update_assocs when purposefully resetting the default qos. -- data_parser/v0.0.39 - Fix segfault when POSTing data with association usage. * Changes in Slurm 23.02.6 ========================== -- Fix CpusPerTres= not upgreadable with scontrol update -- Fix unintentional gres removal when validating the gres job state. -- Fix --without-hpe-slingshot configure option. -- Fix cgroup v2 memory calculations when transparent huge pages are used. -- Fix parsing of sgather --timeout option. -- Fix regression from 22.05.0 that caused srun --cpu-bind "=verbose" and "=v" options give different CPU bind masks. -- Fix "_find_node_record: lookup failure for node" error message appearing for all dynamic nodes during reconfigure. -- Avoid segfault if loading serializer plugin fails. -- slurmrestd - Correct OpenAPI format for 'GET /slurm/v0.0.39/licenses'. -- slurmrestd - Correct OpenAPI format for 'GET /slurm/v0.0.39/job/{job_id}'. -- slurmrestd - Change format to multiple fields in 'GET /slurmdb/v0.0.39/assocations' and 'GET /slurmdb/v0.0.39/qos' to handle infinite and unset states. -- When a node fails in a job with --no-kill, preserve the extern step on the remaining nodes to avoid breaking features that rely on the extern step such as pam_slurm_adopt, x11, and job_container/tmpfs. -- auth/jwt - Ignore 'x5c' field in JWKS files. -- auth/jwt - Treat 'alg' field as optional in JWKS files. -- Allow job_desc.selinux_context to be read from the job_submit.lua script. -- Skip check in slurmstepd that causes a large number of errors in the munge log: "Unauthorized credential for client UID=0 GID=0". This error will still appear on slurmd/slurmctld/slurmdbd start up and is not a cause for concern. -- slurmctld - Allow startup with zero partitions. -- Fix some mig profile names in slurm not matching nvidia mig profiles. -- Prevent slurmscriptd processing delays from blocking other threads in slurmctld while trying to launch {Prolog|Epilog}Slurmctld. -- Fix sacct printing ReqMem field when memory doesn't exist in requested TRES. -- Fix how heterogenous steps in an allocation with CR_PACK_NODE or -mpack are created. -- Fix slurmctld crash from race condition within job_submit_throttle plugin. -- Fix --with-systemdsystemunitdir when requesting a default location. -- Fix not being able to cancel an array task by the jobid (i.e. not _) through scancel, job launch failure or prolog failure. -- Fix cancelling the whole array job when the array task is the meta job and it fails job or prolog launch and is not requeable. Cancel only the specific task instead. -- Fix regression in 21.08.2 where MailProg did not run for mail-type=end for jobs with non-zero exit codes. -- Fix incorrect setting of memory.swap.max in cgroup/v2. -- Fix jobacctgather/cgroup collection of disk/io, gpumem, gpuutil TRES values. -- Fix -d singleton for heterogeneous jobs. -- Downgrade info logs about a job meeting a "maximum node limit" in the select plugin to DebugFlags=SelectType. These info logs could spam the slurmctld log file under certain circumstances. -- prep/script - Fix [Srun|Task] missing SLURM_JOB_NODELIST. -- gres - Rebuild GRES core bitmap for nodes at startup. This fixes error: "Core bitmaps size mismatch on node [HOSTNAME]", which causes jobs to enter state "Requested node configuration is not available". -- slurmctd - Allow startup with zero nodes. -- Fix filesystem handling race conditions that could lead to an attacker taking control of an arbitrary file, or removing entire directories' contents. CVE-2023-41914. -- Return error when updating ArrayTaskThrottle on a non-ArrayJobId. * Changes in Slurm 23.02.5 ========================== -- Add the JobId to debug() messages indicating when cpus_per_task/mem_per_cpu or pn_min_cpus are being automatically adjusted. -- Fix regression in 23.02.2 that caused slurmctld -R to crash on startup if a node features plugin is configured. -- Fix and prevent reoccurring reservations from overlapping. -- job_container/tmpfs - Avoid attempts to share BasePath between nodes. -- Change the log message warning for rate limited users from verbose to info. -- With CR_Cpu_Memory, fix node selection for jobs that request gres and --mem-per-cpu. -- Fix a regression from 22.05.7 in which some jobs were allocated too few nodes, thus overcommitting cpus to some tasks. -- Fix a job being stuck in the completing state if the job ends while the primary controller is down or unresponsive and the backup controller has not yet taken over. -- Fix slurmctld segfault when a node registers with a configured CpuSpecList while slurmctld configuration has the node without CpuSpecList. -- Fix cloud nodes getting stuck in POWERED_DOWN+NO_RESPOND state after not registering by ResumeTimeout. -- slurmstepd - Avoid cleanup of config.json-less containers spooldir getting skipped. -- slurmstepd - Cleanup per task generated environment for containers in spooldir. -- Fix scontrol segfault when 'completing' command requested repeatedly in interactive mode. -- Properly handle a race condition between bind() and listen() calls in the network stack when running with SrunPortRange set. -- Federation - Fix revoked jobs being returned regardless of the -a/--all option for privileged users. -- Federation - Fix canceling pending federated jobs from non-origin clusters which could leave federated jobs orphaned from the origin cluster. -- Fix sinfo segfault when printing multiple clusters with --noheader option. -- Federation - fix clusters not syncing if clusters are added to a federation before they have registered with the dbd. -- Change pmi2 plugin to honor the SrunPortRange option. This matches the new behavior of the pmix plugin in 23.02.0. Note that neither of these plugins makes use of the "MpiParams=ports=" option, and previously were only limited by the systems ephemeral port range. -- node_features/helpers - Fix node selection for jobs requesting changeable features with the '|' operator, which could prevent jobs from running on some valid nodes. -- node_features/helpers - Fix inconsistent handling of '&' and '|', where an AND'd feature was sometimes AND'd to all sets of features instead of just the current set. E.g. "foo|bar&baz" was interpreted as {foo,baz} or {bar,baz} instead of how it is documented: "{foo} or {bar,baz}". -- Fix job accounting so that when a job is requeued its allocated node count is cleared. After the requeue, sacct will correctly show that the job has 0 AllocNodes while it is pending or if it is canceled before restarting. -- sacct - AllocCPUS now correctly shows 0 if a job has not yet received an allocation or if the job was canceled before getting one. -- Fix intel oneapi autodetect: detect the /dev/dri/renderD[0-9]+ gpus, and do not detect /dev/dri/card[0-9]+. -- Format batch, extern, interactive, and pending step ids into strings that are human readable. -- Fix node selection for jobs that request --gpus and a number of tasks fewer than gpus, which resulted in incorrectly rejecting these jobs. -- Remove MYSQL_OPT_RECONNECT completely. -- Fix cloud nodes in POWERING_UP state disappearing (getting set to FUTURE) when an `scontrol reconfigure` happens. -- openapi/dbv0.0.39 - Avoid assert / segfault on missing coordinators list. -- slurmrestd - Correct memory leak while parsing OpenAPI specification templates with server overrides. -- slurmrestd - Reduce memory usage when printing out job CPU frequency. -- Fix overwriting user node reason with system message. -- Remove --uid / --gid options from salloc and srun commands. -- Prevent deadlock when rpc_queue is enabled. -- slurmrestd - Correct OpenAPI specification generation bug where fields with overlapping parent paths would not get generated. -- Fix memory leak as a result of a partition info query. -- Fix memory leak as a result of a job info query. -- slurmrestd - For 'GET /slurm/v0.0.39/node[s]', change format of node's energy field "current_watts" to a dictionary to account for unset value instead of dumping 4294967294. -- slurmrestd - For 'GET /slurm/v0.0.39/qos', change format of QOS's field "priority" to a dictionary to account for unset value instead of dumping 4294967294. -- slurmrestd - For 'GET /slurm/v0.0.39/job[s]', the 'return code' code field in v0.0.39_job_exit_code will be set to -127 instead of being left unset where job does not have a relevant return code. -- data_parser/v0.0.39 - Add required/memory_per_cpu and required/memory_per_node to `sacct --json` and `sacct --yaml` and 'GET /slurmdb/v0.0.39/jobs' from slurmrestd. -- For step allocations, fix --gres=none sometimes not ignoring gres from the job. -- Fix --exclusive jobs incorrectly gang-scheduling where they shouldn't. -- Fix allocations with CR_SOCKET, gres not assigned to a specific socket, and block core distribion potentially allocating more sockets than required. -- gpu/oneapi - Store cores correctly so CPU affinity is tracked. -- Revert a change in 23.02.3 where Slurm would kill a script's process group as soon as the script ended instead of waiting as long as any process in that process group held the stdout/stderr file descriptors open. That change broke some scripts that relied on the previous behavior. Setting time limits for scripts (such as PrologEpilogTimeout) is strongly encouraged to avoid Slurm waiting indefinitely for scripts to finish. -- Allow slurmdbd -R to work if the root assoc id is not 1. -- Fix slurmdbd -R not returning an error under certain conditions. -- slurmdbd - Avoid potential NULL pointer dereference in the mysql plugin. -- Revert a change in 23.02 where SLURM_NTASKS was no longer set in the job's environment when --ntasks-per-node was requested. -- Limit periodic node registrations to 50 instead of the full TreeWidth. Since unresolvable cloud/dynamic nodes must disable fanout by setting TreeWidth to a large number, this would cause all nodes to register at once. -- Fix regression in 23.02.3 which broken x11 forwarding for hosts when MUNGE sends a localhost address in the encode host field. This is caused when the node hostname is mapped to 127.0.0.1 (or similar) in /etc/hosts. -- openapi/[db]v0.0.39 - fix memory leak on parsing error. -- data_parser/v0.0.39 - fix updating qos for associations. -- openapi/dbv0.0.39 - fix updating values for associations with null users. -- Fix minor memory leak with --tres-per-task and licenses. -- Fix cyclic socket cpu distribution for tasks in a step where --cpus-per-task < usable threads per core. * Changes in Slurm 23.02.4 ========================== -- Fix sbatch return code when --wait is requested on a job array. -- switch/hpe_slingshot - avoid segfault when running with old libcxi. -- Avoid slurmctld segfault when specifying AccountingStorageExternalHost. -- Fix collected GPUUtilization values for acct_gather_profile plugins. -- Fix slurmrestd handling of job hold/release operations. -- Make spank S_JOB_ARGV item value hold the requested command argv instead of the srun --bcast value when --bcast requested (only in local context). -- Fix step running indefinitely when slurmctld takes more than MessageTimeout to respond. Now, slurmctld will cancel the step when detected, preventing following steps from getting stuck waiting for resources to be released. -- Fix regression to make job_desc.min_cpus accurate again in job_submit when requesting a job with --ntasks-per-node. -- scontrol - Permit changes to StdErr and StdIn for pending jobs. -- scontrol - Reset std{err,in,out} when set to empty string. -- slurmrestd - mark environment as a required field for job submission descriptions. -- slurmrestd - avoid dumping null in OpenAPI schema required fields. -- data_parser/v0.0.39 - avoid rejecting valid memory_per_node formatted as dictionary provided with a job description. -- data_parser/v0.0.39 - avoid rejecting valid memory_per_cpu formatted as dictionary provided with a job description. -- slurmrestd - Return HTTP error code 404 when job query fails. -- slurmrestd - Add return schema to error response to job and license query. -- Fix handling of ArrayTaskThrottle in backfill. -- Fix regression in 23.02.2 when checking gres state on slurmctld startup or reconfigure. Gres changes in the configuration were not updated on slurmctld startup. On startup or reconfigure, these messages were present in the log: "error: Attempt to change gres/gpu Count". -- Fix potential double count of gres when dealing with limits. -- switch/hpe_slingshot - support alternate traffic class names with "TC_" prefix. -- scrontab - Fix cutting off the final character of quoted variables. -- Fix slurmstepd segfault when ContainerPath is not set in oci.conf -- Change the log message warning for rate limited users from debug to verbose. -- Fixed an issue where jobs requesting licenses were incorrectly rejected. -- smail - Fix issues where e-mails at job completion were not being sent. -- scontrol/slurmctld - fix comma parsing when updating a reservation's nodes. -- cgroup/v2 - Avoid capturing log output for ebpf when constraining devices, as this can lead to inadvertent failure if the log buffer is too small. -- Fix --gpu-bind=single binding tasks to wrong gpus, leading to some gpus having more tasks than they should and other gpus being unused. -- Fix main scheduler loop not starting after failover to backup controller. -- Added error message when attempting to use sattach on batch or extern steps. -- Fix regression in 23.02 that causes slurmstepd to crash when srun requests more than TreeWidth nodes in a step and uses the pmi2 or pmix plugin. -- Reject job ArrayTaskThrottle update requests from unprivileged users. -- data_parser/v0.0.39 - populate description fields of property objects in generated OpenAPI specifications where defined. -- slurmstepd - Avoid segfault caused by ContainerPath not being terminated by '/' in oci.conf. -- data_parser/v0.0.39 - Change v0.0.39_job_info response to tag exit_code field as being complex instead of only an unsigned integer. -- job_container/tmpfs - Fix %h and %n substitution in BasePath where %h was substituted as the NodeName instead of the hostname, and %n was substituted as an empty string. -- Fix regression where --cpu-bind=verbose would override TaskPluginParam. -- scancel - Fix --clusters/-M for federations. Only filtered jobs (e.g. -A, -u, -p, etc.) from the specified clusters will be canceled, rather than all jobs in the federation. Specific jobids will still be routed to the origin cluster for cancellation. -- Add SelectTypeParameters=LL_SHARED_GRES. * Changes in Slurm 23.02.3 ========================== -- Fix regression in 23.02.2 that ignored the partition DefCpuPerGPU setting on the first pass of scheduling a job requesting --gpus --ntasks. -- openapi/dbv0.0.39/users - If a default account update failed, resulting in a no-op, the query returned success without any warning. Now a warning is sent back to the client that the default account wasn't modified. -- srun - fix issue creating regular and interactive steps because *_PACK_GROUP* environment variables were incorrectly set on non-HetSteps. -- Fix dynamic nodes getting stuck in allocated states when reconfiguring. -- Avoid job write lock when nodes are dynamically added/removed. -- burst_buffer/lua - allow jobs to get scheduled sooner after slurm_bb_data_in completes. -- mpi/pmix - fix regression introduced in 23.02.2 which caused PMIx shmem backed files permissions to be incorrect. -- api/submit - fix memory leaks when submission of batch regular jobs or batch HetJobs fails (response data is a return code). -- openapi/v0.0.39 - fix memory leak in _job_post_het_submit(). -- Fix regression in 23.02.2 that set the SLURM_NTASKS environment variable in sbatch jobs from --ntasks-per-node when --ntasks was not requested. -- Fix regression in 23.02 that caused sbatch jobs to set the wrong number of tasks when requesting --ntasks-per-node without --ntasks, and also requesting one of the following options: --sockets-per-node, --cores-per-socket, --threads-per-core (or --hint=nomultithread), or -B,--extra-node-info. -- Fix double counting suspended job counts on nodes when reconfiguring, which prevented nodes with suspended jobs from being powered down or rebooted once the jobs completed. -- Fix backfill not scheduling jobs submitted with --prefer and --constraint properly. -- Avoid possible slurmctld segfault caused by race condition with already completed slurmdbd_conn connections. -- Slurmdbd.conf checks included conf files for 0600 permissions -- slurmrestd - fix regression "oversubscribe" fields were removed from job descriptions and submissions from v0.0.39 end points. -- accounting_storage/mysql - Query for indiviual QOS correctly when you have more than 10. -- Add warning message about ignoring --tres-per-tasks=license when used on a step. -- sshare - Fix command to work when using priority/basic. -- Avoid loading cli_filter plugins outside of salloc/sbatch/scron/srun. This fixes a number of missing symbol problems that can manifest for executables linked against libslurm (and not libslurmfull). -- Allow cloud_reg_addrs to update dynamically registered node's addrs on subsequent registrations. -- switch/hpe_slingshot - Fix hetjob components being assigned different vnis. -- Revert a change in 22.05.5 that prevented tasks from sharing a core if --cpus-per-task > threads per core, but caused incorrect accounting and cpu binding. Instead, --ntasks-per-core=1 may be requested to prevent tasks from sharing a core. -- Correctly send assoc_mgr lock to mcs plugin. -- Fix regression in 23.02 leading to error() messages being sent at INFO instead of ERR in syslog. -- switch/hpe_slingshot - Fix bad instant-on data due to incorrect parsing of data from jackaloped. -- Fix TresUsageIn[Tot|Ave] calculation for gres/gpumem and gres/gpuutil. -- Avoid unnecessary gres/gpumem and gres/gpuutil TRES position lookups. -- Fix issue in the gpu plugins where gpu frequencies would only be set if both gpu memory and gpu frequencies were set, while one or the other suffices. -- Fix reservations group ACL's not working with the root group. -- slurmctld - Fix backup slurmctld crash when it takes control multiple times. -- Fix updating a job with a ReqNodeList greater than the job's node count. -- Fix inadvertent permission denied error for --task-prolog and --task-epilog with filesystems mounted with root_squash. -- switch/hpe_slingshot - remove the unused vni_pids option. -- Fix missing detailed cpu and gres information in json/yaml output from scontrol, squeue and sinfo. -- Fix regression in 23.02 that causes a failure to allocate job steps that request --cpus-per-gpu and gpus with types. -- sacct - when printing PLANNED time, use end time instead of start time for jobs cancelled before they started. -- Fix potentially waiting indefinitely for a defunct process to finish, which affects various scripts including Prolog and Epilog. This could have various symptoms, such as jobs getting stuck in a completing state. -- Hold the job with "(Reservation ... invalid)" state reason if the reservation is not usable by the job. -- Fix losing list of reservations on job when updating job with list of reservations and restarting the controller. -- Fix nodes resuming after down and drain state update requests from clients older than 23.02. -- Fix advanced reservation creation/update when an association that should have access to it is composed with partition(s). -- auth/jwt - Fix memory leak. -- sbatch - Added new --export=NIL option. -- Fix job layout calculations with --ntasks-per-gpu, especially when --nodes has not been explicitly provided. -- Fix X11 forwarding for jobs submitted from the slurmctld host. -- When a job requests --no-kill and one or more nodes fail during the job, fix subsequent job steps unable to use some of the remaining resources allocated to the job. -- Fix shared gres allocation when using --tres-per-task with tasks that span multiple sockets. * Changes in Slurm 23.02.2 ========================== -- Fix regression introduced with the migration to interfaces which caused sshare to core dump. Sshare now initialized the priority context correctly when calculating with PriorityFlags=NO_FAIR_TREE. -- Fix IPMI DCMI sensor initialization. -- For the select/cons_tres plugin, improve the best effort GPU to core binding, for requests with per job task count (-n) and GPU (--gpus) specification. -- scrontab - don't update the cron job tasks if the whole crontab file is left untouched after opening it with "scrontab -e". -- mpi/pmix - avoid crashing when running PMIx v5.0 branch with shmem support. -- Fix building switch topology after a reconfig with the correct nodes. -- Allow a dynamic node to register with a reason, using --conf, when the state is DOWN or DRAIN. -- Fix slurmd running tasks before RPC Prolog is run. -- Fix slurmd deadlock iff the controller were to give a bad alias_list. -- slurmrestd - correctly process job submission field "exclusive" with boolean True or False. -- slurmrestd - correctly process job submission field "exclusive" with strings "true" or "false". -- slurmctld/step_mgr - prevent non-allocatable steps from decrementing values that weren't previously incremented when trying to allocate them. -- auth/jwt - Fix memory leak in slurmctld with 'scontrol token'. -- Fix shared gres (shard/mps) leak when using --tres-per-task -- Fix sacctmgr segfault when listing accounts with coordinators. -- slurmrestd - improve error logging when client connections experience polling errors. -- slurmrestd - improve handling of sockets in different states of shutdown to avoid infinite poll() loop causing a thread to max CPU usage until process is killed. -- slurmrestd - avoid possible segfault caused by race condition of already completed connections. -- mpi/cray_shasta - Fix PMI shared secret for hetjobs. -- gpu/oneapi - Fix CPU affinity handling. -- Fix dynamic nodes powering up when already up after adding/deleting nodes when using power_save logic. -- slurmrestd - Add support for setting max connections. -- data_parser/v0.0.39 - fix sacct --json matching associations from a different cluster. -- Fix segfault when clearing reqnodelist of a pending job. -- Fix memory leak of argv when submitting jobs via slurmrestd or CLI commands. -- slurmrestd - correct miscalculation of job argument count that could cause memory leak when job submission fails. -- slurmdbd - add warning on startup if max_allowed_packet is too small. -- gpu/nvml - Remove E-cores from NVML's cpu affinity bitmap when "allow_ecores" is not set in SlurmdParameters. -- Fix regression from 23.02.0rc1 causing a FrontEnd slurmd to assert fail on startup and don't be configured with the appropriate port. -- Fix dynamic nodes not being sorted and not being included in topology, which resulted in suboptimal dynamic node selection for jobs. -- Fix slurmstepd crash due to potential division by zero (SIGFPE) in certain edge-cases using the PMIx plugin. -- Fix issue with PMIx HetJob requests where certain use-cases would end up with communication errors due to incorrect PMIx hostname info setup. -- openapi/v0.0.39 - revert regression in job update requests to accept job description for changes instead of requiring job description in "job" field. -- Fix regression in 23.02.0rc1 that caused a step to crash with a bad --gpu-bind=single request. -- job_container/tmpfs - skip more in-depth attempt to clean up the base path when not required. This prevents unhelpful, and possibly misleading, debug2 messages when not using the new "shared" mode. -- gpu/nvml - Fix gpu usage when graphics processes are running on the gpu. -- slurmrestd - fix regression where "exclusive" field was removed from job descriptions and submissions. -- Fix issue where requeued jobs had bad gres allocations leading to gres not being deallocated at the end of the job, preventing other jobs from using those resources. -- Fix regression in 23.02.0rc1 which caused incorrect values for SLURM_TASKS_PER_NODE when the job requests --ntasks-per-node and --exclusive or --ntasks-per-core=1 (or CR_ONE_TASK_PER_CORE) and without requesting --ntasks. SLURM_TASKS_PER_NODE is used by mpirun, so this regression caused mpirun to launch the wrong number of tasks and to sometimes fail to launch tasks. -- Prevent jobs running on shards from being canceled on slurmctld restart. -- Fix SPANK prolog and epilog hooks that rely on slurm_init() for access to internal Slurm API calls. -- oci.conf - Populate %m pattern with ContainerPath or SlurmdSpoolDir if ContainerPath is not configured. -- Removed zero padding for numeric values in container spool directory names. -- Avoid creating an unused task-4294967295 directory in container spooldir. -- Cleanup container step directories at step completion. -- sacctmgr - Fix segfault when printing empty tres. -- srun - fix communication issue that prevented slurmctld from connecting to an srun running outside of a compute node. * Changes in Slurm 23.02.1 ========================== -- job_container/tmpfs - cleanup job container even if namespace mount is already unmounted. -- When cluster specific tables are be removed also remove the job_env_table and job_script_table. -- Fix the way bf_max_job_test is applied to job arrays in backfill. -- data_parser/v0.0.39 - Avoid dumping -1 value or NULL when step's consumed_energy is unset. -- scontrol - Fix showing Array Job Steps. -- scontrol - Fix showing Job HetStep. -- openapi/dbv0.0.38 - Fix not displaying an error when updating QOS or associations fails. -- data_parser/v0.0.39 - Avoid crash while parsing composite structures. -- sched/backfill - fix deleted planned node staying in planned node bitmap. -- Fix nodes remaining as PLANNED after slurmctld save state recovery. -- Fix parsing of cgroup.controllers file with a blank line at the end. -- Add cgroup.conf EnableControllers option for cgroup/v2. -- Get correct cgroup root to allow slurmd to run in containers like Docker. -- Fix "(null)" cluster name in SLURM_WORKING_CLUSTER env. -- slurmctld - add missing PrivateData=jobs check to step ContainerID lookup requests originated from 'scontrol show step container-id=' or certain scrun operations when container state can't be directly queried. -- Automatically sort the TaskPlugin list reverse-alphabetically. This addresses an issue where cpu masks were reset if task/affinity was listed before task/cgroup on cgroup/v2 systems with Linux kernel < 6.2. -- Fix some failed terminate job requests from a 23.02 slurmctld to a 22.05 or 21.08 slurmd. -- Fix compile issues on 32-bit systems. -- Fix nodes un-draining after being drained due to unkillable step. -- Fix remote licenses allowed percentages reset to 0 during upgrade. -- sacct - Avoid truncating time strings when using SLURM_TIME_FORMAT with the --parsable option. -- data_parser/v0.0.39 - fix segfault when default qos is not set. -- Fix regression in 22.05.0rc1 that broke Nodes=ALL in a NodeSet. -- openapi/v0.0.39 - fix jobs submitted via slurmrestd being allocated fewer CPUs than tasks when requesting multiple tasks. -- Fix job not being scheduled on valid nodes and potentially being rejected when using parentheses at the beginning of square brackets in a feature request, for example: "feat1&[(feat2|feat3)]". -- Fix a job being scheduled on nodes that do not match a feature request that uses parentheses inside of brackets and requests additional features outside of brackets, for example: "feat1&[feat2|(feat3|feat4)]". -- Fix regression in 23.02.0rc1 which made --gres-flags=enforce-binding no longer enforce optimal core-gpu job placement. -- switch/hpe_slingshot - add option to disable VNI allocation per-job. -- switch/hpe_slingshot - restrict CXI services to the requesting user. -- switch/hpe_slingshot - Only output tcs once in SLINGSHOT_TCS env. -- switch/hpe_slingshot - Fix updating LEs and ACs limits. -- switch/hpe_slingshot - Use correct Max for EQs and CTs. -- switch/hpe_slingshot - support configuring network options per-job. -- switch/hpe_slingshot - retry destroying CXI service if necessary. -- Fix memory leak caused by job preemption when licenses are configured. -- mpi/pmix - Fix v5 to load correctly when libpmix.so isn't in the normal lib path. -- data_parser/v0.0.39 - fix regression where "memory_per_node" would be rejected for job submission. -- data_parser/v0.0.39 - fix regression where "memory_per_cpu" would be rejected for job submission. -- slurmctld - add an assert to check for magic number presence before deleting a partition record and clear the magic afterwards to better diagnose potential memory problems. -- Clean up OCI containers task directories correctly. -- slurm.spec - add "--with jwt" option. -- scrun - Run under existing job when SLURM_JOB_ID is present. -- Prevent a slurmstepd crash when the I/O subsystem has hung. -- common/conmgr - fix memory leak of complete connection list. -- data_parser/v0.0.39 - fix memory leak when parsing every field in a struct. -- job_container/tmpfs - avoid printing extraneous error messages when running a spank plugin that implements slurm_spank_job_prolog() or slurm_spank_job_epilog(). -- Fix srun < 23.02 always getting an "exact" core allocation. -- Prevent scontrol < 23.02 from setting MaxCPUsPerSocket to 0. -- Add ScronParameters=explicit_scancel and corresponding scancel --cron option. * Changes in Slurm 23.02.0 ========================== -- scrun - Install into /bin instead of /sbin, which also ensures it is included in the RPM packages. -- data_parser/v0.0.39 - avoid erroring on unknown association id. -- data_parser/v0.0.39 - avoid fatal while dumping TRES values. -- Improve error message for using --cpus-per-gpus without any GPUs. -- switch/hpe_slingshot - fix rolling upgrades from 22.05 to 23.02 -- Change cgroup/v1 behavior when waiting for a pid to be moved by waiting some time between retries. This helps with stray cgroup dirs on slow kernels. -- Workaround a bug in kernels < 3.18 with cpuset which randomly leaves stray cgroups after jobs have ended. -- Rebuild the prepacked buf going to the slurmstepd if TRES changed. -- Add new 'make clean-contrib' target to build system to clean contribs dir. -- Treat newlines as delimeters for hostlist_create(). (This is used internally by "scontrol show hostlist" and for handling SLURM_JOB_NODELIST.) -- scontrol - Print "Invalid job id specified" for 'scontrol show step ...' when the job does not exist, rather than "Unexpected error". -- Handle mismatched TRES count from the slurmstepd's instead of fatal()'ing. -- Fix GPU setup on CRAY systems when using the CRAY_CUDA_MPS environment variable. GPUs are now correctly detected in such scenarios. -- data_parser/v0.0.39 - improve flag handling to avoid flags having bits left unset after parsing. -- topology/tree - Add new TopologyParam=SwitchAsNodeRank option to reorder nodes based on switch layout. This can be useful if the naming convention for the nodes does not natually map to the network topology. -- openapi/v0.0.38 - avoid signed math errors while dumping node resources. -- openapi/dbv0.0.39 - avoid error while dumping account coordinators. -- data_parser/v0.0.39 - correct dump of stats backfill mean table size. -- openapi/dbv0.0.39 - avoid error while adding account coordinators. -- scrun - catch invalid values of environment SCRUN_FILE_DEBUG. -- scrun - avoid memory leak due to invalid annotation. -- scrun - avoid false error when pidfile not requested. -- Do not constrain memory in task/cgroup unless CR_Memory is set. -- Fix the job prolog not running for jobs with the interactive step (salloc jobs with LaunchParameters=use_interactive_step set in slurm.conf) that were scheduled on powered down nodes. The prolog not running also broke job_container/tmpfs, pam_slurm_adopt, and x11 forwarding. -- swich/hpe_slingshot - fix issues using 22.05 commands with 23.02 slurmd. -- swith/hpe_slingshot - avoid slurmd segfault if there is a protocol error. -- task/affinity - fix slurmd segfault when request launch task requests of type "--cpu-bind=[map,mask]_cpu:" have no provided. -- salloc/sbatch/srun - error out if "--cpu-bind=[map,mask]_cpu:" fails to extract a list of cpus. -- job_container/tmpfs - cleanup job_mount when container creation fails. -- job_container/tmpfs - don't attempt to remove job_mount directory when it is still mounted. -- Fix regression in rc1 causing sacctmgr to segfault when printing unrelated fields. -- jobcomp/kafka - don't use the purge API if librdkafka < v1.0.0. -- Change shown start time of pending array job to be start time of earliest pending array task. -- job_container/tmpfs - ensure that step_ns_fd is closed before cleaning up the namespace. -- Fix assert when suspending or requeueing array tasks with scontrol. -- Fix federated job submissions. -- Fix sinfo returning non-zero exit code when querying specific clusters. -- openapi/v0.0.39 - correct OpenAPI schema for job update requests. -- openapi/v0.0.39 - prevent assertion on jobs endpoint by altering how the "memory_per_cpu" and "memory_per_node" fields are managed. -- Fix regression in 23.02.0rc1 which caused slurmctld to segfault when gres is added or removed in the configuration. -- Removed the default setting for GpuFreqDef. If unset, no attempt to change the GPU frequency will be made if --gpu-freq is not set for the step. -- Fixed GpuFreqDef option. When set in slurm.conf, it will be used if --gpu-freq was not explicitly set by the job step. -- Update database index for usage tables used for archive and purge. -- Fix configure script on FreeBSD. * Changes in Slurm 23.02.0rc1 ============================= -- Make scontrol reconfigure and sending a SIGHUP to the slurmctld behave the same. If you were using SIGHUP as a 'lighter' scontrol reconfigure to rotate logs please update your scripts to use SIGUSR2 instead. -- Add Account and QOS name type specifications for sprio output formatting. -- openapi/[db]v0.0.36 - plugins have been removed. -- openapi/[db]v0.0.37 - tagged as deprecated. -- openapi/[db]v0.0.39 - forked plugin openapi/[db]v0.0.38. -- Add SRUN_{ERROR,INPUT,OUTPUT} input environment variables for --error, --input and --output options respectively. -- Add MaxCPUsPerSocket to partition configuration, similar to MaxCPUsPerNode. -- openapi/v0.0.39 - change nice request field from string to integer. -- sacctmgr - no longer force updates to the AdminComment, Comment, or SystemComment to lower-case. -- openapi/dbv0.0.39 - more graceful handling of POSTs when expected field lists are missing or unparsable. -- burst_buffer/lua - pass the job's UID and GID to slurm_bb_pre_run, slurm_bb_data_in, slurm_bb_post_run, and slurm_bb_data_out in burst_buffer.lua. -- openapi/v0.0.39 - add new field default_memory_per_node for partitions. -- Change cloud nodes to show by default. PrivateData=cloud is no longer needed. -- Add -F/--future option to sinfo to display future nodes. -- openapi/dbv0.0.39 - resolve job user from uid when user is not provided. -- sacct/sinfo/squeue - use openapi/[db]v0.0.39 for --json and --yaml modes. -- sdiag - Add --json and --yaml arguments. -- Cgroupv2 plugin will not show "Controller is not enabled" error when started from cmd. -- sreport - Count planned (FKA reserved) time for jobs running in IGNORE_JOBS reservations. Previously was lumped into IDLE time. -- job_container/tmpfs - support running with an arbitrary list of private mount points (/tmp and /dev/shm are nolonger required, but are the default). -- sacct - Rename 'Reserved' field to 'Planned' to match sreport and the nomenclature of the 'Planned' node. -- oci.conf: Add %U as OCI pattern replacement to numeric user id. -- cli_filter/lua, jobcomp/lua, job_submit/lua - expose all Slurm error codes. -- jobcomp/elasticsearch - expose features field. -- jobcomp/elasticsearch - post non-NULL but empty string fields. -- Make advanced reservation flag MAINT not replace nodes, similar to STATIC_ALLOC -- NVML - Add usage gathering for Nvidia gpus. -- node_features plugins - invalid users specified for AllowUserBoot will now result in fatal() rather than just an error. -- job_container/tmpfs - Set more environment variables in InitScript. -- Make all cgroup directories created by Slurm owned by root. -- sbatch - add parsing of #PBS -d and #PBS -w. -- Avoid network receive error on heavily loaded machines where the network operation can be restarted. -- Deprecate AllowedKmemSpace, ConstrainKmemSpace, MaxKmemPercent, and MinKmemSpace. -- accounting_storage/mysql - change purge/archive to calculate record ages based on end time, rather than start or submission times. -- Add strigger --draining and -R/--resume options. -- Allow updating SLURM_NTASKS environment variable (from scontrol update job). -- Limit het job license requests to the leader job only. -- Change --oversubscribe and --exclusive to be mutually exclusive for job submission. Job submission commands will now fatal if both are set. Previously, these options would override each other, with the last one in the job submission command taking effect. -- job_submit/lua - add support for log_user() from slurm_job_modify(). -- openapi/v0.0.39 - add submission of job->prefer value. -- Fix tasks binding to GPUs when using --ntasks-per-gpu and GPUs have different core/socket affinity. -- Add support for padding StdOut/StdErr/StdIn format specifiers to scontrol show job. -- Fix sacctmgr qos filter to properly select assocs by qos inheritance rules. -- Prevent slurmctld from starting with invalid jobcomp plugin. -- Fix `sinfo -i` and interactive `scontrol> show node` replies for nodes going from powering down to powered down state. -- auth/jwt - support Azure (among other) JWKS files by preferring x5c field over locally reconstructing an RSA256 key file. -- auth/jwt - add optional AuthAltParameters field "userclaimfield=" to allow overriding "sun" claim to site specific field. -- scontrol - Requested TRES and allocated TRES will now always be printed when showing jobs, instead of one TRES output that was either the requested or allocated. -- Improve scheduling performance for dynamic nodes and nodes with features by consolidating like config records. -- Expose argc and entire argv to job_submit and cli_filter plugins for jobs submitted via salloc/srun or sbatch