1# Related configuration of rootCoord, used to handle data definition language (DDL) and data control language (DCL) requests
2rootCoord:
3  maxDatabaseNum: 64 # Maximum number of database
4  maxPartitionNum: 4096 # Maximum number of partitions in a collection
5  minSegmentSizeToEnableIndex: 1024 # It's a threshold. When the segment size is less than this value, the segment will not be indexed
6  maxGeneralCapacity: 65536 # upper limit for the sum of of product of partitionNumber and shardNumber
7  grpc:
8    serverMaxSendSize: 536870912
9    serverMaxRecvSize: 268435456
10    clientMaxSendSize: 268435456
11    clientMaxRecvSize: 536870912
12    
13# Related configuration of proxy, used to validate client requests and reduce the returned results.
14proxy:
15  timeTickInterval: 200 # ms, the interval that proxy synchronize the time tick
16  healthCheckTimeout: 3000 # ms, the interval that to do component healthy check
17  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.
18  maxFieldNum: 64 # The maximum number of field can be created when creating in a collection. It is strongly DISCOURAGED to set maxFieldNum >= 64.
19  maxTaskNum: 1024 # max task number of proxy task queue
20  maxDimension: 32768 # The maximum number of dimensions of a vector can have when creating in a collection.
21  slowQuerySpanInSeconds: 1 # threshold for slow query detection in seconds.
22  grpc:
23    serverMaxSendSize: 268435456
24    serverMaxRecvSize: 67108864
25    clientMaxSendSize: 268435456
26    clientMaxRecvSize: 67108864
27
28# Related configuration of queryCoord, used to manage topology and load balancing for the query nodes, and handoff from growing segments to sealed segments.
29queryCoord:
30  autoHandoff: true # Enable auto handoff
31  autoBalance: true # Enable auto balance
32  autoBalanceChannel: true # Enable auto balance channel
33  balancer: ScoreBasedBalancer # Balancer to use
34  overloadedMemoryThresholdPercentage: 90 # The threshold percentage that memory overload
35  balanceIntervalSeconds: 60
36  memoryUsageMaxDifferencePercentage: 30
37  checkInterval: 1000
38  brokerTimeout: 5000 # 5000ms, querycoord broker rpc timeout
39  channelTaskTimeout: 60000 # 1 minute
40  segmentTaskTimeout: 120000 # 2 minute
41  distPullInterval: 500
42  heartbeatAvailableInterval: 10000 # 10s, Only QueryNodes which fetched heartbeats within the duration are available
43  loadTimeoutSeconds: 600
44  checkHandoffInterval: 5000
45  enableActiveStandby: false
46  channelExclusiveNodeFactor: 4 # the least node number for enable channel's exclusive mode
47  updateCollectionLoadStatusInterval: 5 # 5m, max interval of updating collection loaded status for check health
48  checkExecutedFlagInterval: 100 # the interval of check executed flag to force to pull dist
49  channelTaskCapFraction: 0.3 # fraction of total task execution capacity reserved for channel tasks per node (0.0-1.0)
50  checkHealthInterval: 3000 # 3s, the interval when query coord try to check health of query node
51  checkHealthRPCTimeout: 2000 # 100ms, the timeout of check health rpc to query node
52  collectionObserverInterval: 200 # the interval of collection observer
53  collectionRecoverTimes: 3 # if collection recover times reach the limit during loading state, release it
54  observerTaskParallel: 16 # the parallel observer dispatcher task number
55  resourceExhaustionPenaltyDuration: 30
56  resourceExhaustionCleanupInterval: 10 # Interval (in seconds) for cleaning up expired resource exhaustion marks on query nodes.
57  grpc:
58    serverMaxSendSize: 536870912
59    serverMaxRecvSize: 268435456
60    clientMaxSendSize: 268435456
61    clientMaxRecvSize: 536870912
62
63# Related configuration of queryNode, used to run hybrid search between vector and scalar data.
64queryNode:
65  dataSync:
66    flowGraph:
67      maxQueueLength: 16 # The maximum size of task queue cache in flow graph in query node.
68      maxParallelism: 1024 # Maximum number of tasks executed in parallel in the flowgraph
69    maxParallelSyncMgrTasksPerCPUCore: 16 # The max concurrent sync task number of datanode sync mgr per CPU core
70  stats:
71    publishInterval: 1000 # The interval that query node publishes the node statistics information, including segment status, cpu usage, memory usage, health status, etc. Unit: ms.
72  segcore:
73    knowhereThreadPoolNumRatio: 4
74    # Use more threads to make better use of SSD throughput in disk index.
75    # This parameter is only useful when enable-disk = true.
76    # And this value should be a number greater than 1 and less than 32.
77    chunkRows: 128 # The number of vectors in a chunk.
78    interimIndex: # build a vector temperate index for growing segment or binlog to accelerate search
79      enableIndex: true
80      nlist: 128 # segment index nlist
81      nprobe: 16 # nprobe to search segment, based on your accuracy requirement, must smaller than nlist
82      memExpansionRate: 1.15 # the ratio of building interim index memory usage to raw data
83  loadMemoryUsageFactor: 1 # The multiply factor of calculating the memory usage while loading segments
84  enableDisk: false # enable querynode load disk index, and search on disk index
85  maxDiskUsagePercentage: 95
86  grouping:
87    maxNQ: 1000
88    topKMergeRatio: 20
89  scheduler:
90    receiveChanSize: 10240
91    unsolvedQueueSize: 10240
92    # maxReadConcurrentRatio is the concurrency ratio of read task (search task and query task).
93    # Max read concurrency would be the value of runtime.NumCPU * maxReadConcurrentRatio.
94    # It defaults to 2.0, which means max read concurrency would be the value of runtime.NumCPU * 2.
95    # Max read concurrency must greater than or equal to 1, and less than or equal to runtime.NumCPU * 100.
96    # (0, 100]
97    maxReadConcurrentRatio: 1
98    cpuRatio: 10 # ratio used to estimate read task cpu usage.
99    maxTimestampLag: 86400
100    # read task schedule policy: fifo(by default), user-task-polling.
101    scheduleReadPolicy:
102      # fifo: A FIFO queue support the schedule.
103      # user-task-polling:
104      #     The user's tasks will be polled one by one and scheduled.
105      #     Scheduling is fair on task granularity.
106      #     The policy is based on the username for authentication.
107      #     And an empty username is considered the same user.
108      #     When there are no multi-users, the policy decay into FIFO
109      name: fifo
110      # user-task-polling configure:
111      taskQueueExpire: 60 # 1 min by default, expire time of inner user task queue since queue is empty.
112      enableCrossUserGrouping: false # false by default Enable Cross user grouping when using user-task-polling policy. (close it if task of any user can not merge others).
113      maxPendingTaskPerUser: 1024 # 50 by default, max pending task in scheduler per user.
114  grpc:
115    serverMaxSendSize: 536870912
116    serverMaxRecvSize: 268435456
117    clientMaxSendSize: 268435456
118    clientMaxRecvSize: 536870912
119
120indexCoord:
121  bindIndexNodeMode:
122    enable: false
123    withCred: false
124  segment:
125    minSegmentNumRowsToEnableIndex: 1024 # It's a threshold. When the segment num rows is less than this value, the segment will not be indexed
126
127indexNode:
128  scheduler:
129    buildParallel: 1
130
131dataCoord:
132  channel:
133    watchTimeoutInterval: 300 # Timeout on watching channels (in seconds). Datanode tickler update watch progress will reset timeout timer.
134    balanceSilentDuration: 300 # The duration before the channelBalancer on datacoord to run
135    balanceInterval: 360 #The interval for the channelBalancer on datacoord to check balance status
136  segment:
137    maxSize: 1024 # Maximum size of a segment in MB
138    diskSegmentMaxSize: 2048 # Maximum size of a segment in MB for collection which has Disk index
139    sealProportion: 0.12
140    # The time of the assignment expiration in ms
141    # Warning! this parameter is an expert variable and closely related to data integrity. Without specific
142    # target and solid understanding of the scenarios, it should not be changed. If it's necessary to alter
143    # this parameter, make sure that the newly changed value is larger than the previous value used before restart
144    # otherwise there could be a large possibility of data loss
145    assignmentExpiration: 2000
146    maxLife: 86400 # The max lifetime of segment in seconds, 24*60*60
147    # If a segment didn't accept dml records in maxIdleTime and the size of segment is greater than
148    # minSizeFromIdleToSealed, Milvus will automatically seal it.
149    # The max idle time of segment in seconds, 10*60.
150    maxIdleTime: 600
151    minSizeFromIdleToSealed: 16 # The min size in MB of segment which can be idle from sealed.
152    # The max number of binlog file for one segment, the segment will be sealed if
153    # the number of binlog file reaches to max value.
154    maxBinlogFileNumber: 32
155    smallProportion: 0.5 # The segment is considered as "small segment" when its # of rows is smaller than
156    # (smallProportion * segment max # of rows).
157    # A compaction will happen on small segments if the segment after compaction will have
158    compactableProportion: 0.85
159    # over (compactableProportion * segment max # of rows) rows.
160    # MUST BE GREATER THAN OR EQUAL TO <smallProportion>!!!
161    # During compaction, the size of segment # of rows is able to exceed segment max # of rows by (expansionRate-1) * 100%.
162    expansionRate: 1.25
163  enableCompaction: true # Enable data segment compaction
164  compaction:
165    enableAutoCompaction: true
166    rpcTimeout: 10 # compaction rpc request timeout in seconds
167    maxParallelTaskNum: 10 # max parallel compaction task number
168    indexBasedCompaction: true
169
170    levelzero:
171      forceTrigger:
172        minSize: 8 # The minmum size in MB to force trigger a LevelZero Compaction
173        deltalogMinNum: 10 # the minimum number of deltalog files to force trigger a LevelZero Compaction
174
175  enableGarbageCollection: true
176  gc:
177    interval: 3600 # gc interval in seconds
178    missingTolerance: 3600 # file meta missing tolerance duration in seconds, 3600
179    dropTolerance: 10800 # file belongs to dropped entity tolerance duration in seconds. 10800
180  enableActiveStandby: false
181  grpc:
182    serverMaxSendSize: 536870912
183    serverMaxRecvSize: 268435456
184    clientMaxSendSize: 268435456
185    clientMaxRecvSize: 536870912
186
187dataNode:
188  dataSync:
189    flowGraph:
190      maxQueueLength: 16 # Maximum length of task queue in flowgraph
191      maxParallelism: 1024 # Maximum number of tasks executed in parallel in the flowgraph
192    skipMode:
193      # when there are only timetick msg in flowgraph for a while (longer than coldTime),
194      # flowGraph will turn on skip mode to skip most timeticks to reduce cost, especially there are a lot of channels
195      enable: true
196      skipNum: 4
197      coldTime: 60
198  segment:
199    insertBufSize: 16777216 # Max buffer size to flush for a single segment.
200    deleteBufBytes: 16777216 # Max buffer size to flush del for a single channel
201    syncPeriod: 600 # The period to sync segments if buffer is not empty.
202  # can specify ip for example
203  # ip: 127.0.0.1
204  grpc:
205    serverMaxSendSize: 536870912
206    serverMaxRecvSize: 268435456
207    clientMaxSendSize: 268435456
208    clientMaxRecvSize: 536870912
209  memory:
210    forceSyncEnable: true # Set true to force sync if memory usage is too high
211    forceSyncSegmentNum: 1 # number of segments to sync, segments with top largest buffer will be synced.
212    forceSyncWatermark: 0.5 # memory watermark for standalone, upon reaching this watermark, segments will be synced.
213  channel:
214    # specify the size of global work pool of all channels
215    # if this parameter <= 0, will set it as the maximum number of CPUs that can be executing
216    # suggest to set it bigger on large collection numbers to avoid blocking
217    workPoolSize: -1
218    # specify the size of global work pool for channel checkpoint updating
219    # if this parameter <= 0, will set it as 1000
220    # suggest to set it bigger on large collection numbers to avoid blocking
221    updateChannelCheckpointMaxParallel: 10
222
223grpc:
224  client:
225    compressionEnabled: false
226    dialTimeout: 200
227    keepAliveTime: 10000
228    keepAliveTimeout: 20000
229    maxMaxAttempts: 10
230    initialBackoff: 0.2
231    initialBackOff: 0.2 # seconds
232    maxBackoff: 10 # seconds
233    
234quotaAndLimits:
235  enabled: true # `true` to enable quota and limits, `false` to disable.
236  limits:
237    maxCollectionNum: 65536
238    maxCollectionNumPerDB: 65536
239  # quotaCenterCollectInterval is the time interval that quotaCenter
240  # collects metrics from Proxies, Query cluster and Data cluster.
241  # seconds, (0 ~ 65536)
242  quotaCenterCollectInterval: 3
243  # FactorChangeThreshold defines the minimum relative change in factor to trigger an update.
244  # If the factor change is less than this threshold (e.g., 5%), the update is skipped
245  # to reduce unnecessary proxy updates. Range: (0, 1]
246  factorChangeThreshold: 0.05
247  ddl:
248    enabled: false
249    collectionRate: -1 # qps, default no limit, rate for CreateCollection, DropCollection, LoadCollection, ReleaseCollection
250    partitionRate: -1 # qps, default no limit, rate for CreatePartition, DropPartition, LoadPartition, ReleasePartition
251  indexRate:
252    enabled: false
253    max: -1 # qps, default no limit, rate for CreateIndex, DropIndex
254  flushRate:
255    enabled: false
256    max: -1 # qps, default no limit, rate for flush
257  compactionRate:
258    enabled: false
259    max: -1 # qps, default no limit, rate for manualCompaction
260  dml:
261    # dml limit rates, default no limit.
262    # The maximum rate will not be greater than max.
263    enabled: false
264    insertRate:
265      collection:
266        max: -1 # MB/s, default no limit
267      max: -1 # MB/s, default no limit
268    deleteRate:
269      collection:
270        max: -1 # MB/s, default no limit
271      max: -1 # MB/s, default no limit
272    bulkLoadRate:
273      collection:
274        max: -1 # MB/s, default no limit, not support yet. TODO: limit bulkLoad rate
275      max: -1 # MB/s, default no limit, not support yet. TODO: limit bulkLoad rate
276  dql:
277    # dql limit rates, default no limit.
278    # The maximum rate will not be greater than max.
279    enabled: false
280    searchRate:
281      collection:
282        max: -1 # vps (vectors per second), default no limit
283      max: -1 # vps (vectors per second), default no limit
284    queryRate:
285      collection:
286        max: -1 # qps, default no limit
287      max: -1 # qps, default no limit
288  limitWriting:
289    # forceDeny false means dml requests are allowed (except for some
290    # specific conditions, such as memory of nodes to water marker), true means always reject all dml requests.
291    forceDeny: false
292    ttProtection:
293      enabled: false
294      # maxTimeTickDelay indicates the backpressure for DML Operations.
295      # DML rates would be reduced according to the ratio of time tick delay to maxTimeTickDelay,
296      # if time tick delay is greater than maxTimeTickDelay, all DML requests would be rejected.
297      # seconds
298      maxTimeTickDelay: 1200
299    memProtection:
300      # When memory usage > memoryHighWaterLevel, all dml requests would be rejected;
301      # When memoryLowWaterLevel < memory usage < memoryHighWaterLevel, reduce the dml rate;
302      # When memory usage < memoryLowWaterLevel, no action.
303      enabled: true
304      dataNodeMemoryLowWaterLevel: 0.85 # (0, 1], memoryLowWaterLevel in DataNodes
305      dataNodeMemoryHighWaterLevel: 0.95 # (0, 1], memoryHighWaterLevel in DataNodes
306      queryNodeMemoryLowWaterLevel: 0.85 # (0, 1], memoryLowWaterLevel in QueryNodes
307      queryNodeMemoryHighWaterLevel: 0.95 # (0, 1], memoryHighWaterLevel in QueryNodes
308    growingSegmentsSizeProtection:
309      # No action will be taken if the growing segments size is less than the low watermark.
310      # When the growing segments size exceeds the low watermark, the dml rate will be reduced,
311      # but the rate will not be lower than `minRateRatio * dmlRate`.
312      enabled: false
313      minRateRatio: 0.5
314      lowWaterLevel: 0.2
315      highWaterLevel: 0.4
316    diskProtection:
317      enabled: true # When the total file size of object storage is greater than `diskQuota`, all dml requests would be rejected;
318      diskQuota: -1 # MB, (0, +inf), default no limit
319      diskQuotaPerCollection: -1 # MB, (0, +inf), default no limit
320      loadedDiskQuota: -1 # MB, (0, +inf), default no limit
321  limitReading:
322    # forceDeny false means dql requests are allowed (except for some
323    # specific conditions, such as collection has been dropped), true means always reject all dql requests.
324    forceDeny: false