# Licensed to the LF AI & Data foundation under one # or more contributor license agreements. See the NOTICE file # distributed with this work for additional information # regarding copyright ownership. The ASF licenses this file # to you under the Apache License, Version 2.0 (the # "License"); you may not use this file except in compliance # with the License. You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # Related configuration of etcd, used to store Milvus metadata & service discovery. etcd: # Endpoints used to access etcd service. You can change this parameter as the endpoints of your own etcd cluster. # Environment variable: ETCD_ENDPOINTS # etcd preferentially acquires valid address from environment variable ETCD_ENDPOINTS when Milvus is started. endpoints: localhost:2379 # Root prefix of the key to where Milvus stores data in etcd. # It is recommended to change this parameter before starting Milvus for the first time. # To share an etcd instance among multiple Milvus instances, consider changing this to a different value for each Milvus instance before you start them. # Set an easy-to-identify root path for Milvus if etcd service already exists. # Changing this for an already running Milvus instance may result in failures to read legacy data. rootPath: by-dev # Sub-prefix of the key to where Milvus stores metadata-related information in etcd. # Caution: Changing this parameter after using Milvus for a period of time will affect your access to old data. # It is recommended to change this parameter before starting Milvus for the first time. metaSubPath: meta # Sub-prefix of the key to where Milvus stores timestamps in etcd. # Caution: Changing this parameter after using Milvus for a period of time will affect your access to old data. # It is recommended not to change this parameter if there is no specific reason. kvSubPath: kv log: level: info # Only supports debug, info, warn, error, panic, or fatal. Default 'info'. # path is one of: # - "default" as os.Stderr, # - "stderr" as os.Stderr, # - "stdout" as os.Stdout, # - file path to append server logs to. # please adjust in embedded Milvus: /tmp/milvus/logs/etcd.log path: stdout ssl: enabled: false # Whether to support ETCD secure connection mode tlsCert: /path/to/etcd-client.pem # path to your cert file tlsKey: /path/to/etcd-client-key.pem # path to your key file tlsCACert: /path/to/ca.pem # path to your CACert file # TLS min version # Optional values: 1.0, 1.1, 1.2, 1.3。 # We recommend using version 1.2 and above. tlsMinVersion: 1.3 requestTimeout: 10000 # Etcd operation timeout in milliseconds use: embed: false # Whether to enable embedded Etcd (an in-process EtcdServer). data: dir: default.etcd # Embedded Etcd only. please adjust in embedded Milvus: /tmp/milvus/etcdData/ auth: enabled: false # Whether to enable authentication userName: # username for etcd authentication password: # password for etcd authentication metastore: type: etcd # Default value: etcd, Valid values: [etcd, tikv] # Related configuration of tikv, used to store Milvus metadata. # Notice that when TiKV is enabled for metastore, you still need to have etcd for service discovery. # TiKV is a good option when the metadata size requires better horizontal scalability. tikv: endpoints: 127.0.0.1:2389 # Note that the default pd port of tikv is 2379, which conflicts with etcd. rootPath: by-dev # The root path where data is stored in tikv metaSubPath: meta # metaRootPath = rootPath + '/' + metaSubPath kvSubPath: kv # kvRootPath = rootPath + '/' + kvSubPath requestTimeout: 10000 # ms, tikv request timeout snapshotScanSize: 256 # batch size of tikv snapshot scan ssl: enabled: false # Whether to support TiKV secure connection mode tlsCert: # path to your cert file tlsKey: # path to your key file tlsCACert: # path to your CACert file localStorage: # Local path to where vector data are stored during a search or a query to avoid repetitve access to MinIO or S3 service. # Caution: Changing this parameter after using Milvus for a period of time will affect your access to old data. # It is recommended to change this parameter before starting Milvus for the first time. path: /var/lib/milvus/data/ # Related configuration of MinIO/S3/GCS or any other service supports S3 API, which is responsible for data persistence for Milvus. # We refer to the storage service as MinIO/S3 in the following description for simplicity. minio: # IP address of MinIO or S3 service. # Environment variable: MINIO_ADDRESS # minio.address and minio.port together generate the valid access to MinIO or S3 service. # MinIO preferentially acquires the valid IP address from the environment variable MINIO_ADDRESS when Milvus is started. # Default value applies when MinIO or S3 is running on the same network with Milvus. address: localhost port: 9000 # Port of MinIO or S3 service. # Access key ID that MinIO or S3 issues to user for authorized access. # Environment variable: MINIO_ACCESS_KEY_ID or minio.accessKeyID # minio.accessKeyID and minio.secretAccessKey together are used for identity authentication to access the MinIO or S3 service. # This configuration must be set identical to the environment variable MINIO_ACCESS_KEY_ID, which is necessary for starting MinIO or S3. # The default value applies to MinIO or S3 service that started with the default docker-compose.yml file. accessKeyID: minioadmin # Secret key used to encrypt the signature string and verify the signature string on server. It must be kept strictly confidential and accessible only to the MinIO or S3 server and users. # Environment variable: MINIO_SECRET_ACCESS_KEY or minio.secretAccessKey # minio.accessKeyID and minio.secretAccessKey together are used for identity authentication to access the MinIO or S3 service. # This configuration must be set identical to the environment variable MINIO_SECRET_ACCESS_KEY, which is necessary for starting MinIO or S3. # The default value applies to MinIO or S3 service that started with the default docker-compose.yml file. secretAccessKey: minioadmin useSSL: false # Switch value to control if to access the MinIO or S3 service through SSL. ssl: tlsCACert: /path/to/public.crt # path to your CACert file # Name of the bucket where Milvus stores data in MinIO or S3. # Milvus 2.0.0 does not support storing data in multiple buckets. # Bucket with this name will be created if it does not exist. If the bucket already exists and is accessible, it will be used directly. Otherwise, there will be an error. # To share an MinIO instance among multiple Milvus instances, consider changing this to a different value for each Milvus instance before you start them. For details, see Operation FAQs. # The data will be stored in the local Docker if Docker is used to start the MinIO service locally. Ensure that there is sufficient storage space. # A bucket name is globally unique in one MinIO or S3 instance. bucketName: a-bucket # Root prefix of the key to where Milvus stores data in MinIO or S3. # It is recommended to change this parameter before starting Milvus for the first time. # To share an MinIO instance among multiple Milvus instances, consider changing this to a different value for each Milvus instance before you start them. For details, see Operation FAQs. # Set an easy-to-identify root key prefix for Milvus if etcd service already exists. # Changing this for an already running Milvus instance may result in failures to read legacy data. rootPath: files # Whether to useIAM role to access S3/GCS instead of access/secret keys # For more information, refer to # aws: https://docs.aws.amazon.com/IAM/latest/UserGuide/id_roles_use.html # gcp: https://cloud.google.com/storage/docs/access-control/iam # aliyun (ack): https://www.alibabacloud.com/help/en/container-service-for-kubernetes/latest/use-rrsa-to-enforce-access-control # aliyun (ecs): https://www.alibabacloud.com/help/en/elastic-compute-service/latest/attach-an-instance-ram-role useIAM: false # Cloud Provider of S3. Supports: "aws", "gcp", "aliyun". # You can use "aws" for other cloud provider supports S3 API with signature v4, e.g.: minio # You can use "gcp" for other cloud provider supports S3 API with signature v2 # You can use "aliyun" for other cloud provider uses virtual host style bucket # When useIAM enabled, only "aws", "gcp", "aliyun" is supported for now cloudProvider: aws # Custom endpoint for fetch IAM role credentials. when useIAM is true & cloudProvider is "aws". # Leave it empty if you want to use AWS default endpoint iamEndpoint: logLevel: fatal # Log level for aws sdk log. Supported level: off, fatal, error, warn, info, debug, trace region: # Specify minio storage system location region useVirtualHost: false # Whether use virtual host mode for bucket requestTimeoutMs: 10000 # minio timeout for request time in milliseconds # The maximum number of objects requested per batch in minio ListObjects rpc, # 0 means using oss client by default, decrease these configration if ListObjects timeout listObjectsMaxKeys: 0 # Milvus supports four MQ: rocksmq(based on RockDB), natsmq(embedded nats-server), Pulsar and Kafka. # You can change your mq by setting mq.type field. # If you don't set mq.type field as default, there is a note about enabling priority if we config multiple mq in this file. # 1. standalone(local) mode: rocksmq(default) > natsmq > Pulsar > Kafka # 2. cluster mode: Pulsar(default) > Kafka (rocksmq and natsmq is unsupported in cluster mode) mq: # Default value: "default" # Valid values: [default, pulsar, kafka, rocksmq, natsmq] type: default enablePursuitMode: true # Default value: "true" pursuitLag: 10 # time tick lag threshold to enter pursuit mode, in seconds pursuitBufferSize: 8388608 # pursuit mode buffer size in bytes mqBufSize: 16 # MQ client consumer buffer length dispatcher: mergeCheckInterval: 1 # the interval time(in seconds) for dispatcher to check whether to merge targetBufSize: 16 # the lenth of channel buffer for targe maxTolerantLag: 3 # Default value: "3", the timeout(in seconds) that target sends msgPack # Related configuration of pulsar, used to manage Milvus logs of recent mutation operations, output streaming log, and provide log publish-subscribe services. pulsar: # IP address of Pulsar service. # Environment variable: PULSAR_ADDRESS # pulsar.address and pulsar.port together generate the valid access to Pulsar. # Pulsar preferentially acquires the valid IP address from the environment variable PULSAR_ADDRESS when Milvus is started. # Default value applies when Pulsar is running on the same network with Milvus. address: localhost port: 6650 # Port of Pulsar service. webport: 80 # Web port of of Pulsar service. If you connect direcly without proxy, should use 8080. # The maximum size of each message in Pulsar. Unit: Byte. # By default, Pulsar can transmit at most 5 MB of data in a single message. When the size of inserted data is greater than this value, proxy fragments the data into multiple messages to ensure that they can be transmitted correctly. # If the corresponding parameter in Pulsar remains unchanged, increasing this configuration will cause Milvus to fail, and reducing it produces no advantage. maxMessageSize: 5242880 # Pulsar can be provisioned for specific tenants with appropriate capacity allocated to the tenant. # To share a Pulsar instance among multiple Milvus instances, you can change this to an Pulsar tenant rather than the default one for each Milvus instance before you start them. However, if you do not want Pulsar multi-tenancy, you are advised to change msgChannel.chanNamePrefix.cluster to the different value. tenant: public namespace: default # A Pulsar namespace is the administrative unit nomenclature within a tenant. requestTimeout: 60 # pulsar client global request timeout in seconds enableClientMetrics: false # Whether to register pulsar client metrics into milvus metrics path. # If you want to enable kafka, needs to comment the pulsar configs # kafka: # brokerList: # saslUsername: # saslPassword: # saslMechanisms: # securityProtocol: # ssl: # enabled: false # whether to enable ssl mode # tlsCert: # path to client's public key (PEM) used for authentication # tlsKey: # path to client's private key (PEM) used for authentication # tlsCaCert: # file or directory path to CA certificate(s) for verifying the broker's key # tlsKeyPassword: # private key passphrase for use with ssl.key.location and set_ssl_cert(), if any # readTimeout: 10 rocksmq: # Prefix of the key to where Milvus stores data in RocksMQ. # Caution: Changing this parameter after using Milvus for a period of time will affect your access to old data. # It is recommended to change this parameter before starting Milvus for the first time. # Set an easy-to-identify root key prefix for Milvus if etcd service already exists. path: /var/lib/milvus/rdb_data lrucacheratio: 0.06 # rocksdb cache memory ratio rocksmqPageSize: 67108864 # The maximum size of messages in each page in RocksMQ. Messages in RocksMQ are checked and cleared (when expired) in batch based on this parameters. Unit: Byte. retentionTimeInMinutes: 4320 # The maximum retention time of acked messages in RocksMQ. Acked messages in RocksMQ are retained for the specified period of time and then cleared. Unit: Minute. retentionSizeInMB: 8192 # The maximum retention size of acked messages of each topic in RocksMQ. Acked messages in each topic are cleared if their size exceed this parameter. Unit: MB. compactionInterval: 86400 # Time interval to trigger rocksdb compaction to remove deleted data. Unit: Second compressionTypes: 0,0,7,7,7 # compaction compression type, only support use 0,7. 0 means not compress, 7 will use zstd. Length of types means num of rocksdb level. # natsmq configuration. # more detail: https://docs.nats.io/running-a-nats-service/configuration natsmq: server: port: 4222 # Listening port of the NATS server. storeDir: /var/lib/milvus/nats # Directory to use for JetStream storage of nats maxFileStore: 17179869184 # Maximum size of the 'file' storage maxPayload: 8388608 # Maximum number of bytes in a message payload maxPending: 67108864 # Maximum number of bytes buffered for a connection Applies to client connections initializeTimeout: 4000 # waiting for initialization of natsmq finished monitor: trace: false # If true enable protocol trace log messages debug: false # If true enable debug log messages logTime: true # If set to false, log without timestamps. logFile: /tmp/milvus/logs/nats.log # Log file path relative to .. of milvus binary if use relative path logSizeLimit: 536870912 # Size in bytes after the log file rolls over to a new one retention: maxAge: 4320 # Maximum age of any message in the P-channel maxBytes: # How many bytes the single P-channel may contain. Removing oldest messages if the P-channel exceeds this size maxMsgs: # How many message the single P-channel may contain. Removing oldest messages if the P-channel exceeds this limit # Related configuration of rootCoord, used to handle data definition language (DDL) and data control language (DCL) requests rootCoord: dmlChannelNum: 16 # The number of DML-Channels to create at the root coord startup. # The maximum number of partitions in each collection. # New partitions cannot be created if this parameter is set as 0 or 1. # Range: [0, INT64MAX] maxPartitionNum: 1024 # The minimum row count of a segment required for creating index. # Segments with smaller size than this parameter will not be indexed, and will be searched with brute force. minSegmentSizeToEnableIndex: 1024 enableActiveStandby: false maxDatabaseNum: 64 # Maximum number of database maxGeneralCapacity: 65536 # upper limit for the sum of of product of partitionNumber and shardNumber gracefulStopTimeout: 5 # seconds. force stop node without graceful stop ip: # TCP/IP address of rootCoord. If not specified, use the first unicastable address port: 53100 # TCP port of rootCoord grpc: serverMaxSendSize: 536870912 # The maximum size of each RPC request that the rootCoord can send, unit: byte serverMaxRecvSize: 268435456 # The maximum size of each RPC request that the rootCoord can receive, unit: byte clientMaxSendSize: 268435456 # The maximum size of each RPC request that the clients on rootCoord can send, unit: byte clientMaxRecvSize: 536870912 # The maximum size of each RPC request that the clients on rootCoord can receive, unit: byte # Related configuration of proxy, used to validate client requests and reduce the returned results. proxy: timeTickInterval: 200 # The interval at which proxy synchronizes the time tick, unit: ms. healthCheckTimeout: 3000 # ms, the interval that to do component healthy check msgStream: timeTick: bufSize: 512 # The maximum number of messages can be buffered in the timeTick message stream of the proxy when producing messages. maxNameLength: 255 # The maximum length of the name or alias that can be created in Milvus, including the collection name, collection alias, partition name, and field name. maxFieldNum: 64 # The maximum number of field can be created when creating in a collection. It is strongly DISCOURAGED to set maxFieldNum >= 64. maxVectorFieldNum: 4 # The maximum number of vector fields that can be specified in a collection. Value range: [1, 10]. maxShardNum: 16 # The maximum number of shards can be created when creating in a collection. maxDimension: 32768 # The maximum number of dimensions of a vector can have when creating in a collection. # Whether to produce gin logs.\n # please adjust in embedded Milvus: false ginLogging: true ginLogSkipPaths: / # skip url path for gin log maxTaskNum: 1024 # The maximum number of tasks in the task queue of the proxy. mustUsePartitionKey: false # switch for whether proxy must use partition key for the collection accessLog: enable: false # Whether to enable the access log feature. minioEnable: false # Whether to upload local access log files to MinIO. This parameter can be specified when proxy.accessLog.filename is not empty. localPath: /tmp/milvus_access # The local folder path where the access log file is stored. This parameter can be specified when proxy.accessLog.filename is not empty. filename: # The name of the access log file. If you leave this parameter empty, access logs will be printed to stdout. maxSize: 64 # The maximum size allowed for a single access log file. If the log file size reaches this limit, a rotation process will be triggered. This process seals the current access log file, creates a new log file, and clears the contents of the original log file. Unit: MB. rotatedTime: 0 # The maximum time interval allowed for rotating a single access log file. Upon reaching the specified time interval, a rotation process is triggered, resulting in the creation of a new access log file and sealing of the previous one. Unit: seconds remotePath: access_log/ # The path of the object storage for uploading access log files. remoteMaxTime: 0 # The time interval allowed for uploading access log files. If the upload time of a log file exceeds this interval, the file will be deleted. Setting the value to 0 disables this feature. formatters: base: format: "[$time_now] [ACCESS] <$user_name: $user_addr> $method_name [status: $method_status] [code: $error_code] [sdk: $sdk_version] [msg: $error_msg] [traceID: $trace_id] [timeCost: $time_cost]" query: format: "[$time_now] [ACCESS] <$user_name: $user_addr> $method_name [status: $method_status] [code: $error_code] [sdk: $sdk_version] [msg: $error_msg] [traceID: $trace_id] [timeCost: $time_cost] [database: $database_name] [collection: $collection_name] [partitions: $partition_name] [expr: $method_expr]" methods: "Query,Search,Delete" cacheSize: 0 # Size of log of write cache, in byte. (Close write cache if size was 0) cacheFlushInterval: 3 # time interval of auto flush write cache, in seconds. (Close auto flush if interval was 0) connectionCheckIntervalSeconds: 120 # the interval time(in seconds) for connection manager to scan inactive client info connectionClientInfoTTLSeconds: 86400 # inactive client info TTL duration, in seconds maxConnectionNum: 10000 # the max client info numbers that proxy should manage, avoid too many client infos gracefulStopTimeout: 30 # seconds. force stop node without graceful stop slowQuerySpanInSeconds: 5 # query whose executed time exceeds the `slowQuerySpanInSeconds` can be considered slow, in seconds. queryNodePooling: size: 10 # the size for shardleader(querynode) client pool http: enabled: true # Whether to enable the http server debug_mode: false # Whether to enable http server debug mode port: # high-level restful api acceptTypeAllowInt64: true # high-level restful api, whether http client can deal with int64 enablePprof: true # Whether to enable pprof middleware on the metrics port ip: # TCP/IP address of proxy. If not specified, use the first unicastable address port: 19530 # TCP port of proxy internalPort: 19529 grpc: serverMaxSendSize: 268435456 # The maximum size of each RPC request that the proxy can send, unit: byte serverMaxRecvSize: 67108864 # The maximum size of each RPC request that the proxy can receive, unit: byte clientMaxSendSize: 268435456 # The maximum size of each RPC request that the clients on proxy can send, unit: byte clientMaxRecvSize: 67108864 # The maximum size of each RPC request that the clients on proxy can receive, unit: byte # Related configuration of queryCoord, used to manage topology and load balancing for the query nodes, and handoff from growing segments to sealed segments. queryCoord: taskMergeCap: 1 taskExecutionCap: 256 # Switch value to control if to automatically replace a growing segment with the corresponding indexed sealed segment when the growing segment reaches the sealing threshold. # If this parameter is set false, Milvus simply searches the growing segments with brute force. autoHandoff: true autoBalance: true # Switch value to control if to automatically balance the memory usage among query nodes by distributing segment loading and releasing operations evenly. autoBalanceChannel: true # Enable auto balance channel balancer: ScoreBasedBalancer # auto balancer used for segments on queryNodes globalRowCountFactor: 0.1 # the weight used when balancing segments among queryNodes scoreUnbalanceTolerationFactor: 0.05 # the least value for unbalanced extent between from and to nodes when doing balance reverseUnBalanceTolerationFactor: 1.3 # the largest value for unbalanced extent between from and to nodes after doing balance overloadedMemoryThresholdPercentage: 90 # The threshold of memory usage (in percentage) in a query node to trigger the sealed segment balancing. balanceIntervalSeconds: 60 # The interval at which query coord balances the memory usage among query nodes. memoryUsageMaxDifferencePercentage: 30 # The threshold of memory usage difference (in percentage) between any two query nodes to trigger the sealed segment balancing. rowCountFactor: 0.4 # the row count weight used when balancing segments among queryNodes segmentCountFactor: 0.4 # the segment count weight used when balancing segments among queryNodes globalSegmentCountFactor: 0.1 # the segment count weight used when balancing segments among queryNodes segmentCountMaxSteps: 50 # segment count based plan generator max steps rowCountMaxSteps: 50 # segment count based plan generator max steps randomMaxSteps: 10 # segment count based plan generator max steps growingRowCountWeight: 4 # the memory weight of growing segment row count delegatorMemoryOverloadFactor: 0.1 # the factor of delegator overloaded memory balanceCostThreshold: 0.001 # the threshold of balance cost, if the difference of cluster's cost after executing the balance plan is less than this value, the plan will not be executed checkSegmentInterval: 1000 checkChannelInterval: 1000 checkBalanceInterval: 10000 checkIndexInterval: 10000 channelTaskTimeout: 60000 # 1 minute segmentTaskTimeout: 120000 # 2 minute distPullInterval: 500 heartbeatAvailableInterval: 10000 # 10s, Only QueryNodes which fetched heartbeats within the duration are available loadTimeoutSeconds: 600 distRequestTimeout: 5000 # the request timeout for querycoord fetching data distribution from querynodes, in milliseconds heatbeatWarningLag: 5000 # the lag value for querycoord report warning when last heatbeat is too old, in milliseconds checkHandoffInterval: 5000 enableActiveStandby: false checkInterval: 1000 checkHealthInterval: 3000 # 3s, the interval when query coord try to check health of query node checkHealthRPCTimeout: 2000 # 100ms, the timeout of check health rpc to query node brokerTimeout: 5000 # 5000ms, querycoord broker rpc timeout collectionRecoverTimes: 3 # if collection recover times reach the limit during loading state, release it observerTaskParallel: 16 # the parallel observer dispatcher task number checkAutoBalanceConfigInterval: 10 # the interval of check auto balance config checkNodeSessionInterval: 60 # the interval(in seconds) of check querynode cluster session gracefulStopTimeout: 5 # seconds. force stop node without graceful stop enableStoppingBalance: true # whether enable stopping balance channelExclusiveNodeFactor: 4 # the least node number for enable channel's exclusive mode collectionObserverInterval: 200 # the interval of collection observer checkExecutedFlagInterval: 100 # the interval of check executed flag to force to pull dist updateCollectionLoadStatusInterval: 5 # 5m, max interval for updating collection loaded status cleanExcludeSegmentInterval: 60 # the time duration of clean pipeline exclude segment which used for filter invalid data, in seconds ip: # TCP/IP address of queryCoord. If not specified, use the first unicastable address port: 19531 # TCP port of queryCoord grpc: serverMaxSendSize: 536870912 # The maximum size of each RPC request that the queryCoord can send, unit: byte serverMaxRecvSize: 268435456 # The maximum size of each RPC request that the queryCoord can receive, unit: byte clientMaxSendSize: 268435456 # The maximum size of each RPC request that the clients on queryCoord can send, unit: byte clientMaxRecvSize: 536870912 # The maximum size of each RPC request that the clients on queryCoord can receive, unit: byte # Related configuration of queryNode, used to run hybrid search between vector and scalar data. queryNode: stats: publishInterval: 1000 # The interval that query node publishes the node statistics information, including segment status, cpu usage, memory usage, health status, etc. Unit: ms. segcore: knowhereThreadPoolNumRatio: 4 # The number of threads in knowhere's thread pool. If disk is enabled, the pool size will multiply with knowhereThreadPoolNumRatio([1, 32]). chunkRows: 128 # Row count by which Segcore divides a segment into chunks. interimIndex: # Whether to create a temporary index for growing segments and sealed segments not yet indexed, improving search performance. # Milvus will eventually seals and indexes all segments, but enabling this optimizes search performance for immediate queries following data insertion. # This defaults to true, indicating that Milvus creates temporary index for growing segments and the sealed segments that are not indexed upon searches. enableIndex: true nlist: 128 # temp index nlist, recommend to set sqrt(chunkRows), must smaller than chunkRows/8 nprobe: 16 # nprobe to search small index, based on your accuracy requirement, must smaller than nlist memExpansionRate: 1.15 # extra memory needed by building interim index buildParallelRate: 0.5 # the ratio of building interim index parallel matched with cpu num knowhereScoreConsistency: false # Enable knowhere strong consistency score computation logic loadMemoryUsageFactor: 1 # The multiply factor of calculating the memory usage while loading segments enableDisk: false # enable querynode load disk index, and search on disk index maxDiskUsagePercentage: 95 cache: enabled: true memoryLimit: 2147483648 # 2 GB, 2 * 1024 *1024 *1024 readAheadPolicy: willneed # The read ahead policy of chunk cache, options: `normal, random, sequential, willneed, dontneed` # options: async, sync, disable. # Specifies the necessity for warming up the chunk cache. # 1. If set to "sync" or "async" the original vector data will be synchronously/asynchronously loaded into the # chunk cache during the load process. This approach has the potential to substantially reduce query/search latency # for a specific duration post-load, albeit accompanied by a concurrent increase in disk usage; # 2. If set to "disable" original vector data will only be loaded into the chunk cache during search/query. warmup: disable mmap: mmapEnabled: false # Enable mmap for loading data growingMmapEnabled: false # Enable mmap for using in growing raw data fixedFileSizeForMmapAlloc: 1 # tmp file size for mmap chunk manager maxDiskUsagePercentageForMmapAlloc: 50 # disk percentage used in mmap chunk manager lazyload: enabled: false # Enable lazyload for loading data waitTimeout: 30000 # max wait timeout duration in milliseconds before start to do lazyload search and retrieve requestResourceTimeout: 5000 # max timeout in milliseconds for waiting request resource for lazy load, 5s by default requestResourceRetryInterval: 2000 # retry interval in milliseconds for waiting request resource for lazy load, 2s by default maxRetryTimes: 1 # max retry times for lazy load, 1 by default maxEvictPerRetry: 1 # max evict count for lazy load, 1 by default grouping: enabled: true maxNQ: 1000 topKMergeRatio: 20 scheduler: receiveChanSize: 10240 unsolvedQueueSize: 10240 # maxReadConcurrentRatio is the concurrency ratio of read task (search task and query task). # Max read concurrency would be the value of hardware.GetCPUNum * maxReadConcurrentRatio. # It defaults to 2.0, which means max read concurrency would be the value of hardware.GetCPUNum * 2. # Max read concurrency must greater than or equal to 1, and less than or equal to hardware.GetCPUNum * 100. # (0, 100] maxReadConcurrentRatio: 1 cpuRatio: 10 # ratio used to estimate read task cpu usage. maxTimestampLag: 86400 scheduleReadPolicy: # fifo: A FIFO queue support the schedule. # user-task-polling: # The user's tasks will be polled one by one and scheduled. # Scheduling is fair on task granularity. # The policy is based on the username for authentication. # And an empty username is considered the same user. # When there are no multi-users, the policy decay into FIFO" name: fifo taskQueueExpire: 60 # Control how long (many seconds) that queue retains since queue is empty enableCrossUserGrouping: false # Enable Cross user grouping when using user-task-polling policy. (Disable it if user's task can not merge each other) maxPendingTaskPerUser: 1024 # Max pending task per user in scheduler dataSync: flowGraph: maxQueueLength: 16 # The maximum size of task queue cache in flow graph in query node. maxParallelism: 1024 # Maximum number of tasks executed in parallel in the flowgraph enableSegmentPrune: false # use partition stats to prune data in search/query on shard delegator bloomFilterApplyParallelFactor: 4 # parallel factor when to apply pk to bloom filter, default to 4*CPU_CORE_NUM queryStreamBatchSize: 4194304 # return batch size of stream query workerPooling: size: 10 # the size for worker querynode client pool ip: # TCP/IP address of queryNode. If not specified, use the first unicastable address port: 21123 # TCP port of queryNode grpc: serverMaxSendSize: 536870912 # The maximum size of each RPC request that the queryNode can send, unit: byte serverMaxRecvSize: 268435456 # The maximum size of each RPC request that the queryNode can receive, unit: byte clientMaxSendSize: 268435456 # The maximum size of each RPC request that the clients on queryNode can send, unit: byte clientMaxRecvSize: 536870912 # The maximum size of each RPC request that the clients on queryNode can receive, unit: byte indexCoord: bindIndexNodeMode: enable: false address: localhost:22930 withCred: false nodeID: 0 segment: minSegmentNumRowsToEnableIndex: 1024 # It's a threshold. When the segment num rows is less than this value, the segment will not be indexed indexNode: scheduler: buildParallel: 1 enableDisk: true # enable index node build disk vector index maxDiskUsagePercentage: 95 ip: # TCP/IP address of indexNode. If not specified, use the first unicastable address port: 21121 # TCP port of indexNode grpc: serverMaxSendSize: 536870912 # The maximum size of each RPC request that the indexNode can send, unit: byte serverMaxRecvSize: 268435456 # The maximum size of each RPC request that the indexNode can receive, unit: byte clientMaxSendSize: 268435456 # The maximum size of each RPC request that the clients on indexNode can send, unit: byte clientMaxRecvSize: 536870912 # The maximum size of each RPC request that the clients on indexNode can receive, unit: byte dataCoord: channel: watchTimeoutInterval: 300 # Timeout on watching channels (in seconds). Datanode tickler update watch progress will reset timeout timer. balanceWithRpc: true # Whether to enable balance with RPC, default to use etcd watch legacyVersionWithoutRPCWatch: 2.4.1 # Datanodes <= this version are considered as legacy nodes, which doesn't have rpc based watch(). This is only used during rolling upgrade where legacy nodes won't get new channels balanceSilentDuration: 300 # The duration after which the channel manager start background channel balancing balanceInterval: 360 # The interval with which the channel manager check dml channel balance status checkInterval: 1 # The interval in seconds with which the channel manager advances channel states notifyChannelOperationTimeout: 5 # Timeout notifing channel operations (in seconds). segment: maxSize: 1024 # The maximum size of a segment, unit: MB. datacoord.segment.maxSize and datacoord.segment.sealProportion together determine if a segment can be sealed. diskSegmentMaxSize: 2048 # Maximun size of a segment in MB for collection which has Disk index sealProportion: 0.12 # The minimum proportion to datacoord.segment.maxSize to seal a segment. datacoord.segment.maxSize and datacoord.segment.sealProportion together determine if a segment can be sealed. assignmentExpiration: 2000 # Expiration time of the segment assignment, unit: ms allocLatestExpireAttempt: 200 # The time attempting to alloc latest lastExpire from rootCoord after restart maxLife: 86400 # The max lifetime of segment in seconds, 24*60*60 # If a segment didn't accept dml records in maxIdleTime and the size of segment is greater than # minSizeFromIdleToSealed, Milvus will automatically seal it. # The max idle time of segment in seconds, 10*60. maxIdleTime: 600 minSizeFromIdleToSealed: 16 # The min size in MB of segment which can be idle from sealed. # The max number of binlog file for one segment, the segment will be sealed if # the number of binlog file reaches to max value. maxBinlogFileNumber: 32 smallProportion: 0.5 # The segment is considered as "small segment" when its # of rows is smaller than # (smallProportion * segment max # of rows). # A compaction will happen on small segments if the segment after compaction will have compactableProportion: 0.85 # over (compactableProportion * segment max # of rows) rows. # MUST BE GREATER THAN OR EQUAL TO !!! # During compaction, the size of segment # of rows is able to exceed segment max # of rows by (expansionRate-1) * 100%. expansionRate: 1.25 sealPolicy: channel: # The size threshold in MB, if the total size of growing segments of each shard # exceeds this threshold, the largest growing segment will be sealed. growingSegmentsMemSize: 4096 autoUpgradeSegmentIndex: false # whether auto upgrade segment index to index engine's version segmentFlushInterval: 2 # the minimal interval duration(unit: Seconds) between flusing operation on same segment # Switch value to control if to enable segment compaction. # Compaction merges small-size segments into a large segment, and clears the entities deleted beyond the rentention duration of Time Travel. enableCompaction: true compaction: # Switch value to control if to enable automatic segment compaction during which data coord locates and merges compactable segments in the background. # This configuration takes effect only when dataCoord.enableCompaction is set as true. enableAutoCompaction: true indexBasedCompaction: true rpcTimeout: 10 maxParallelTaskNum: 10 workerMaxParallelTaskNum: 2 clustering: enable: true # Enable clustering compaction autoEnable: false # Enable auto clustering compaction triggerInterval: 600 # clustering compaction trigger interval in seconds minInterval: 3600 # The minimum interval between clustering compaction executions of one collection, to avoid redundant compaction maxInterval: 259200 # If a collection haven't been clustering compacted for longer than maxInterval, force compact newDataSizeThreshold: 512m # If new data size is large than newDataSizeThreshold, execute clustering compaction preferSegmentSizeRatio: 0.8 maxSegmentSizeRatio: 1 maxTrainSizeRatio: 0.8 # max data size ratio in Kmeans train, if larger than it, will down sampling to meet this limit maxCentroidsNum: 10240 # maximum centroids number in Kmeans train minCentroidsNum: 16 # minimum centroids number in Kmeans train minClusterSizeRatio: 0.01 # minimum cluster size / avg size in Kmeans train maxClusterSizeRatio: 10 # maximum cluster size / avg size in Kmeans train maxClusterSize: 5g # maximum cluster size in Kmeans train levelzero: forceTrigger: minSize: 8388608 # The minmum size in bytes to force trigger a LevelZero Compaction, default as 8MB maxSize: 67108864 # The maxmum size in bytes to force trigger a LevelZero Compaction, default as 64MB deltalogMinNum: 10 # The minimum number of deltalog files to force trigger a LevelZero Compaction deltalogMaxNum: 30 # The maxmum number of deltalog files to force trigger a LevelZero Compaction, default as 30 syncSegmentsInterval: 300 # The time interval for regularly syncing segments enableGarbageCollection: true # Switch value to control if to enable garbage collection to clear the discarded data in MinIO or S3 service. gc: interval: 3600 # The interval at which data coord performs garbage collection, unit: second. missingTolerance: 86400 # The retention duration of the unrecorded binary log (binlog) files. Setting a reasonably large value for this parameter avoids erroneously deleting the newly created binlog files that lack metadata. Unit: second. dropTolerance: 10800 # The retention duration of the binlog files of the deleted segments before they are cleared, unit: second. removeConcurrent: 32 # number of concurrent goroutines to remove dropped s3 objects scanInterval: 168 # orphan file (file on oss but has not been registered on meta) on object storage garbage collection scanning interval in hours enableActiveStandby: false brokerTimeout: 5000 # 5000ms, dataCoord broker rpc timeout autoBalance: true # Enable auto balance checkAutoBalanceConfigInterval: 10 # the interval of check auto balance config import: filesPerPreImportTask: 2 # The maximum number of files allowed per pre-import task. taskRetention: 10800 # The retention period in seconds for tasks in the Completed or Failed state. maxSizeInMBPerImportTask: 6144 # To prevent generating of small segments, we will re-group imported files. This parameter represents the sum of file sizes in each group (each ImportTask). scheduleInterval: 2 # The interval for scheduling import, measured in seconds. checkIntervalHigh: 2 # The interval for checking import, measured in seconds, is set to a high frequency for the import checker. checkIntervalLow: 120 # The interval for checking import, measured in seconds, is set to a low frequency for the import checker. maxImportFileNumPerReq: 1024 # The maximum number of files allowed per single import request. waitForIndex: true # Indicates whether the import operation waits for the completion of index building. gracefulStopTimeout: 5 # seconds. force stop node without graceful stop slot: clusteringCompactionUsage: 16 # slot usage of clustering compaction job. mixCompactionUsage: 8 # slot usage of mix compaction job. l0DeleteCompactionUsage: 8 # slot usage of l0 compaction job. ip: # TCP/IP address of dataCoord. If not specified, use the first unicastable address port: 13333 # TCP port of dataCoord grpc: serverMaxSendSize: 536870912 # The maximum size of each RPC request that the dataCoord can send, unit: byte serverMaxRecvSize: 268435456 # The maximum size of each RPC request that the dataCoord can receive, unit: byte clientMaxSendSize: 268435456 # The maximum size of each RPC request that the clients on dataCoord can send, unit: byte clientMaxRecvSize: 536870912 # The maximum size of each RPC request that the clients on dataCoord can receive, unit: byte dataNode: dataSync: flowGraph: maxQueueLength: 16 # Maximum length of task queue in flowgraph maxParallelism: 1024 # Maximum number of tasks executed in parallel in the flowgraph maxParallelSyncMgrTasks: 256 # The max concurrent sync task number of datanode sync mgr globally skipMode: enable: true # Support skip some timetick message to reduce CPU usage skipNum: 4 # Consume one for every n records skipped coldTime: 60 # Turn on skip mode after there are only timetick msg for x seconds segment: # The maximum size of each binlog file in a segment buffered in memory. Binlog files whose size exceeds this value are then flushed to MinIO or S3 service. # Unit: Byte # Setting this parameter too small causes the system to store a small amount of data too frequently. Setting it too large increases the system's demand for memory. insertBufSize: 16777216 deleteBufBytes: 16777216 # Max buffer size in bytes to flush del for a single channel, default as 16MB syncPeriod: 600 # The period to sync segments if buffer is not empty. memory: forceSyncEnable: true # Set true to force sync if memory usage is too high forceSyncSegmentNum: 1 # number of segments to sync, segments with top largest buffer will be synced. checkInterval: 3000 # the interal to check datanode memory usage, in milliseconds forceSyncWatermark: 0.5 # memory watermark for standalone, upon reaching this watermark, segments will be synced. timetick: byRPC: true interval: 500 channel: # specify the size of global work pool of all channels # if this parameter <= 0, will set it as the maximum number of CPUs that can be executing # suggest to set it bigger on large collection numbers to avoid blocking workPoolSize: -1 # specify the size of global work pool for channel checkpoint updating # if this parameter <= 0, will set it as 10 updateChannelCheckpointMaxParallel: 10 updateChannelCheckpointInterval: 60 # the interval duration(in seconds) for datanode to update channel checkpoint of each channel updateChannelCheckpointRPCTimeout: 20 # timeout in seconds for UpdateChannelCheckpoint RPC call maxChannelCheckpointsPerPRC: 128 # The maximum number of channel checkpoints per UpdateChannelCheckpoint RPC. channelCheckpointUpdateTickInSeconds: 10 # The frequency, in seconds, at which the channel checkpoint updater executes updates. import: maxConcurrentTaskNum: 16 # The maximum number of import/pre-import tasks allowed to run concurrently on a datanode. maxImportFileSizeInGB: 16 # The maximum file size (in GB) for an import file, where an import file refers to either a Row-Based file or a set of Column-Based files. readBufferSizeInMB: 16 # The data block size (in MB) read from chunk manager by the datanode during import. compaction: levelZeroBatchMemoryRatio: 0.05 # The minimal memory ratio of free memory for level zero compaction executing in batch mode levelZeroMaxBatchSize: -1 # Max batch size refers to the max number of L1/L2 segments in a batch when executing L0 compaction. Default to -1, any value that is less than 1 means no limit. Valid range: >= 1. gracefulStopTimeout: 1800 # seconds. force stop node without graceful stop slot: slotCap: 16 # The maximum number of tasks(e.g. compaction, importing) allowed to run concurrently on a datanode clusteringCompaction: memoryBufferRatio: 0.1 # The ratio of memory buffer of clustering compaction. Data larger than threshold will be flushed to storage. workPoolSize: 8 # worker pool size for one clustering compaction job. ip: # TCP/IP address of dataNode. If not specified, use the first unicastable address port: 21124 # TCP port of dataNode grpc: serverMaxSendSize: 536870912 # The maximum size of each RPC request that the dataNode can send, unit: byte serverMaxRecvSize: 268435456 # The maximum size of each RPC request that the dataNode can receive, unit: byte clientMaxSendSize: 268435456 # The maximum size of each RPC request that the clients on dataNode can send, unit: byte clientMaxRecvSize: 536870912 # The maximum size of each RPC request that the clients on dataNode can receive, unit: byte # This topic introduces the message channel-related configurations of Milvus. msgChannel: chanNamePrefix: # Root name prefix of the channel when a message channel is created. # It is recommended to change this parameter before starting Milvus for the first time. # To share a Pulsar instance among multiple Milvus instances, consider changing this to a name rather than the default one for each Milvus instance before you start them. cluster: by-dev # Sub-name prefix of the message channel where the root coord publishes time tick messages. # The complete channel name prefix is ${msgChannel.chanNamePrefix.cluster}-${msgChannel.chanNamePrefix.rootCoordTimeTick} # Caution: Changing this parameter after using Milvus for a period of time will affect your access to old data. # It is recommended to change this parameter before starting Milvus for the first time. rootCoordTimeTick: rootcoord-timetick # Sub-name prefix of the message channel where the root coord publishes its own statistics messages. # The complete channel name prefix is ${msgChannel.chanNamePrefix.cluster}-${msgChannel.chanNamePrefix.rootCoordStatistics} # Caution: Changing this parameter after using Milvus for a period of time will affect your access to old data. # It is recommended to change this parameter before starting Milvus for the first time. rootCoordStatistics: rootcoord-statistics # Sub-name prefix of the message channel where the root coord publishes Data Manipulation Language (DML) messages. # The complete channel name prefix is ${msgChannel.chanNamePrefix.cluster}-${msgChannel.chanNamePrefix.rootCoordDml} # Caution: Changing this parameter after using Milvus for a period of time will affect your access to old data. # It is recommended to change this parameter before starting Milvus for the first time. rootCoordDml: rootcoord-dml replicateMsg: replicate-msg # Sub-name prefix of the message channel where the query node publishes time tick messages. # The complete channel name prefix is ${msgChannel.chanNamePrefix.cluster}-${msgChannel.chanNamePrefix.queryTimeTick} # Caution: Changing this parameter after using Milvus for a period of time will affect your access to old data. # It is recommended to change this parameter before starting Milvus for the first time. queryTimeTick: queryTimeTick # Sub-name prefix of the message channel where the data coord publishes time tick messages. # The complete channel name prefix is ${msgChannel.chanNamePrefix.cluster}-${msgChannel.chanNamePrefix.dataCoordTimeTick} # Caution: Changing this parameter after using Milvus for a period of time will affect your access to old data. # It is recommended to change this parameter before starting Milvus for the first time. dataCoordTimeTick: datacoord-timetick-channel # Sub-name prefix of the message channel where the data coord publishes segment information messages. # The complete channel name prefix is ${msgChannel.chanNamePrefix.cluster}-${msgChannel.chanNamePrefix.dataCoordSegmentInfo} # Caution: Changing this parameter after using Milvus for a period of time will affect your access to old data. # It is recommended to change this parameter before starting Milvus for the first time. dataCoordSegmentInfo: segment-info-channel subNamePrefix: # Subscription name prefix of the data coord. # Caution: Changing this parameter after using Milvus for a period of time will affect your access to old data. # It is recommended to change this parameter before starting Milvus for the first time. dataCoordSubNamePrefix: dataCoord # Subscription name prefix of the data node. # Caution: Changing this parameter after using Milvus for a period of time will affect your access to old data. # It is recommended to change this parameter before starting Milvus for the first time. dataNodeSubNamePrefix: dataNode # Configures the system log output. log: # Milvus log level. Option: debug, info, warn, error, panic, and fatal. # It is recommended to use debug level under test and development environments, and info level in production environment. level: info file: # Root path to the log files. # The default value is set empty, indicating to output log files to standard output (stdout) and standard error (stderr). # If this parameter is set to a valid local path, Milvus writes and stores log files in this path. # Set this parameter as the path that you have permission to write. rootPath: maxSize: 300 # The maximum size of a log file, unit: MB. maxAge: 10 # The maximum retention time before a log file is automatically cleared, unit: day. The minimum value is 1. maxBackups: 20 # The maximum number of log files to back up, unit: day. The minimum value is 1. format: text # Milvus log format. Option: text and JSON stdout: true # Stdout enable or not grpc: log: level: WARNING gracefulStopTimeout: 10 # second, time to wait graceful stop finish client: compressionEnabled: false dialTimeout: 200 keepAliveTime: 10000 keepAliveTimeout: 20000 maxMaxAttempts: 10 initialBackoff: 0.2 maxBackoff: 10 minResetInterval: 1000 maxCancelError: 32 minSessionCheckInterval: 200 # Configure the proxy tls enable. tls: serverPemPath: configs/cert/server.pem serverKeyPath: configs/cert/server.key caPemPath: configs/cert/ca.pem common: defaultPartitionName: _default # Name of the default partition when a collection is created defaultIndexName: _default_idx # Name of the index when it is created with name unspecified entityExpiration: -1 # Entity expiration in seconds, CAUTION -1 means never expire indexSliceSize: 16 # Index slice size in MB threadCoreCoefficient: highPriority: 10 # This parameter specify how many times the number of threads is the number of cores in high priority pool middlePriority: 5 # This parameter specify how many times the number of threads is the number of cores in middle priority pool lowPriority: 1 # This parameter specify how many times the number of threads is the number of cores in low priority pool buildIndexThreadPoolRatio: 0.75 DiskIndex: MaxDegree: 56 SearchListSize: 100 PQCodeBudgetGBRatio: 0.125 BuildNumThreadsRatio: 1 SearchCacheBudgetGBRatio: 0.1 LoadNumThreadRatio: 8 BeamWidthRatio: 4 gracefulTime: 5000 # milliseconds. it represents the interval (in ms) by which the request arrival time needs to be subtracted in the case of Bounded Consistency. gracefulStopTimeout: 1800 # seconds. it will force quit the server if the graceful stop process is not completed during this time. storageType: remote # please adjust in embedded Milvus: local, available values are [local, remote, opendal], value minio is deprecated, use remote instead # Default value: auto # Valid values: [auto, avx512, avx2, avx, sse4_2] # This configuration is only used by querynode and indexnode, it selects CPU instruction set for Searching and Index-building. simdType: auto security: authorizationEnabled: false # The superusers will ignore some system check processes, # like the old password verification when updating the credential superUsers: defaultRootPassword: Milvus # default password for root user tlsMode: 0 session: ttl: 30 # ttl value when session granting a lease to register service retryTimes: 30 # retry times when session sending etcd requests locks: metrics: enable: false # whether gather statistics for metrics locks threshold: info: 500 # minimum milliseconds for printing durations in info level warn: 1000 # minimum milliseconds for printing durations in warn level storage: scheme: s3 enablev2: false # Whether to disable the internal time messaging mechanism for the system. # If disabled (set to false), the system will not allow DML operations, including insertion, deletion, queries, and searches. # This helps Milvus-CDC synchronize incremental data ttMsgEnabled: true traceLogMode: 0 # trace request info bloomFilterSize: 100000 # bloom filter initial size maxBloomFalsePositive: 0.001 # max false positive rate for bloom filter bloomFilterType: BasicBloomFilter # bloom filter type, support BasicBloomFilter and BlockedBloomFilter bloomFilterApplyBatchSize: 1000 # batch size when to apply pk to bloom filter usePartitionKeyAsClusteringKey: false # if true, do clustering compaction and segment prune on partition key field useVectorAsClusteringKey: false # if true, do clustering compaction and segment prune on vector field enableVectorClusteringKey: false # if true, enable vector clustering key and vector clustering compaction # QuotaConfig, configurations of Milvus quota and limits. # By default, we enable: # 1. TT protection; # 2. Memory protection. # 3. Disk quota protection. # You can enable: # 1. DML throughput limitation; # 2. DDL, DQL qps/rps limitation; # 3. DQL Queue length/latency protection; # 4. DQL result rate protection; # If necessary, you can also manually force to deny RW requests. quotaAndLimits: enabled: true # `true` to enable quota and limits, `false` to disable. # quotaCenterCollectInterval is the time interval that quotaCenter # collects metrics from Proxies, Query cluster and Data cluster. # seconds, (0 ~ 65536) quotaCenterCollectInterval: 3 limits: allocRetryTimes: 15 # retry times when delete alloc forward data from rate limit failed allocWaitInterval: 1000 # retry wait duration when delete alloc forward data rate failed, in millisecond complexDeleteLimitEnable: false # whether complex delete check forward data by limiter maxCollectionNum: 65536 maxCollectionNumPerDB: 65536 # Maximum number of collections per database. maxInsertSize: -1 # maximum size of a single insert request, in bytes, -1 means no limit maxResourceGroupNumOfQueryNode: 1024 # maximum number of resource groups of query nodes ddl: enabled: false # Whether DDL request throttling is enabled. # Maximum number of collection-related DDL requests per second. # Setting this item to 10 indicates that Milvus processes no more than 10 collection-related DDL requests per second, including collection creation requests, collection drop requests, collection load requests, and collection release requests. # To use this setting, set quotaAndLimits.ddl.enabled to true at the same time. collectionRate: -1 # Maximum number of partition-related DDL requests per second. # Setting this item to 10 indicates that Milvus processes no more than 10 partition-related requests per second, including partition creation requests, partition drop requests, partition load requests, and partition release requests. # To use this setting, set quotaAndLimits.ddl.enabled to true at the same time. partitionRate: -1 db: collectionRate: -1 # qps of db level , default no limit, rate for CreateCollection, DropCollection, LoadCollection, ReleaseCollection partitionRate: -1 # qps of db level, default no limit, rate for CreatePartition, DropPartition, LoadPartition, ReleasePartition indexRate: enabled: false # Whether index-related request throttling is enabled. # Maximum number of index-related requests per second. # Setting this item to 10 indicates that Milvus processes no more than 10 partition-related requests per second, including index creation requests and index drop requests. # To use this setting, set quotaAndLimits.indexRate.enabled to true at the same time. max: -1 db: max: -1 # qps of db level, default no limit, rate for CreateIndex, DropIndex flushRate: enabled: true # Whether flush request throttling is enabled. # Maximum number of flush requests per second. # Setting this item to 10 indicates that Milvus processes no more than 10 flush requests per second. # To use this setting, set quotaAndLimits.flushRate.enabled to true at the same time. max: -1 collection: max: 0.1 # qps, default no limit, rate for flush at collection level. db: max: -1 # qps of db level, default no limit, rate for flush compactionRate: enabled: false # Whether manual compaction request throttling is enabled. # Maximum number of manual-compaction requests per second. # Setting this item to 10 indicates that Milvus processes no more than 10 manual-compaction requests per second. # To use this setting, set quotaAndLimits.compaction.enabled to true at the same time. max: -1 db: max: -1 # qps of db level, default no limit, rate for manualCompaction dml: enabled: false # Whether DML request throttling is enabled. insertRate: # Highest data insertion rate per second. # Setting this item to 5 indicates that Milvus only allows data insertion at the rate of 5 MB/s. # To use this setting, set quotaAndLimits.dml.enabled to true at the same time. max: -1 db: max: -1 # MB/s, default no limit collection: # Highest data insertion rate per collection per second. # Setting this item to 5 indicates that Milvus only allows data insertion to any collection at the rate of 5 MB/s. # To use this setting, set quotaAndLimits.dml.enabled to true at the same time. max: -1 partition: max: -1 # MB/s, default no limit upsertRate: max: -1 # MB/s, default no limit db: max: -1 # MB/s, default no limit collection: max: -1 # MB/s, default no limit partition: max: -1 # MB/s, default no limit deleteRate: # Highest data deletion rate per second. # Setting this item to 0.1 indicates that Milvus only allows data deletion at the rate of 0.1 MB/s. # To use this setting, set quotaAndLimits.dml.enabled to true at the same time. max: -1 db: max: -1 # MB/s, default no limit collection: # Highest data deletion rate per second. # Setting this item to 0.1 indicates that Milvus only allows data deletion from any collection at the rate of 0.1 MB/s. # To use this setting, set quotaAndLimits.dml.enabled to true at the same time. max: -1 partition: max: -1 # MB/s, default no limit bulkLoadRate: max: -1 # MB/s, default no limit, not support yet. TODO: limit bulkLoad rate db: max: -1 # MB/s, default no limit, not support yet. TODO: limit db bulkLoad rate collection: max: -1 # MB/s, default no limit, not support yet. TODO: limit collection bulkLoad rate partition: max: -1 # MB/s, default no limit, not support yet. TODO: limit partition bulkLoad rate dql: enabled: false # Whether DQL request throttling is enabled. searchRate: # Maximum number of vectors to search per second. # Setting this item to 100 indicates that Milvus only allows searching 100 vectors per second no matter whether these 100 vectors are all in one search or scattered across multiple searches. # To use this setting, set quotaAndLimits.dql.enabled to true at the same time. max: -1 db: max: -1 # vps (vectors per second), default no limit collection: # Maximum number of vectors to search per collection per second. # Setting this item to 100 indicates that Milvus only allows searching 100 vectors per second per collection no matter whether these 100 vectors are all in one search or scattered across multiple searches. # To use this setting, set quotaAndLimits.dql.enabled to true at the same time. max: -1 partition: max: -1 # vps (vectors per second), default no limit queryRate: # Maximum number of queries per second. # Setting this item to 100 indicates that Milvus only allows 100 queries per second. # To use this setting, set quotaAndLimits.dql.enabled to true at the same time. max: -1 db: max: -1 # qps, default no limit collection: # Maximum number of queries per collection per second. # Setting this item to 100 indicates that Milvus only allows 100 queries per collection per second. # To use this setting, set quotaAndLimits.dql.enabled to true at the same time. max: -1 partition: max: -1 # qps, default no limit limitWriting: # forceDeny false means dml requests are allowed (except for some # specific conditions, such as memory of nodes to water marker), true means always reject all dml requests. forceDeny: false ttProtection: enabled: false # maxTimeTickDelay indicates the backpressure for DML Operations. # DML rates would be reduced according to the ratio of time tick delay to maxTimeTickDelay, # if time tick delay is greater than maxTimeTickDelay, all DML requests would be rejected. # seconds maxTimeTickDelay: 300 memProtection: # When memory usage > memoryHighWaterLevel, all dml requests would be rejected; # When memoryLowWaterLevel < memory usage < memoryHighWaterLevel, reduce the dml rate; # When memory usage < memoryLowWaterLevel, no action. enabled: true dataNodeMemoryLowWaterLevel: 0.85 # (0, 1], memoryLowWaterLevel in DataNodes dataNodeMemoryHighWaterLevel: 0.95 # (0, 1], memoryHighWaterLevel in DataNodes queryNodeMemoryLowWaterLevel: 0.85 # (0, 1], memoryLowWaterLevel in QueryNodes queryNodeMemoryHighWaterLevel: 0.95 # (0, 1], memoryHighWaterLevel in QueryNodes growingSegmentsSizeProtection: # No action will be taken if the growing segments size is less than the low watermark. # When the growing segments size exceeds the low watermark, the dml rate will be reduced, # but the rate will not be lower than minRateRatio * dmlRate. enabled: false minRateRatio: 0.5 lowWaterLevel: 0.2 highWaterLevel: 0.4 diskProtection: enabled: true # When the total file size of object storage is greater than `diskQuota`, all dml requests would be rejected; diskQuota: -1 # MB, (0, +inf), default no limit diskQuotaPerDB: -1 # MB, (0, +inf), default no limit diskQuotaPerCollection: -1 # MB, (0, +inf), default no limit diskQuotaPerPartition: -1 # MB, (0, +inf), default no limit l0SegmentsRowCountProtection: enabled: false # switch to enable l0 segment row count quota lowWaterLevel: 32768 # l0 segment row count quota, low water level highWaterLevel: 65536 # l0 segment row count quota, low water level limitReading: # forceDeny false means dql requests are allowed (except for some # specific conditions, such as collection has been dropped), true means always reject all dql requests. forceDeny: false queueProtection: enabled: false # nqInQueueThreshold indicated that the system was under backpressure for Search/Query path. # If NQ in any QueryNode's queue is greater than nqInQueueThreshold, search&query rates would gradually cool off # until the NQ in queue no longer exceeds nqInQueueThreshold. We think of the NQ of query request as 1. # int, default no limit nqInQueueThreshold: -1 # queueLatencyThreshold indicated that the system was under backpressure for Search/Query path. # If dql latency of queuing is greater than queueLatencyThreshold, search&query rates would gradually cool off # until the latency of queuing no longer exceeds queueLatencyThreshold. # The latency here refers to the averaged latency over a period of time. # milliseconds, default no limit queueLatencyThreshold: -1 resultProtection: enabled: false # maxReadResultRate indicated that the system was under backpressure for Search/Query path. # If dql result rate is greater than maxReadResultRate, search&query rates would gradually cool off # until the read result rate no longer exceeds maxReadResultRate. # MB/s, default no limit maxReadResultRate: -1 maxReadResultRatePerDB: -1 maxReadResultRatePerCollection: -1 # colOffSpeed is the speed of search&query rates cool off. # (0, 1] coolOffSpeed: 0.9 trace: # trace exporter type, default is stdout, # optional values: ['noop','stdout', 'jaeger', 'otlp'] exporter: noop # fraction of traceID based sampler, # optional values: [0, 1] # Fractions >= 1 will always sample. Fractions < 0 are treated as zero. sampleFraction: 0 jaeger: url: # when exporter is jaeger should set the jaeger's URL otlp: endpoint: # example: "127.0.0.1:4317" for grpc, "127.0.0.1:4318" for http method: # otlp export method, acceptable values: ["grpc", "http"], using "grpc" by default secure: true initTimeoutSeconds: 10 # segcore initialization timeout in seconds, preventing otlp grpc hangs forever #when using GPU indexing, Milvus will utilize a memory pool to avoid frequent memory allocation and deallocation. #here, you can set the size of the memory occupied by the memory pool, with the unit being MB. #note that there is a possibility of Milvus crashing when the actual memory demand exceeds the value set by maxMemSize. #if initMemSize and MaxMemSize both set zero, #milvus will automatically initialize half of the available GPU memory, #maxMemSize will the whole available GPU memory. gpu: initMemSize: # Gpu Memory Pool init size maxMemSize: # Gpu Memory Pool Max size