跳到主要内容
版本:dev

Monitoring Milvus

Collect and monitor general performance metrics of Milvus vector database.

Protocol Used: milvus

Milvus exposes a Prometheus-compatible metrics endpoint. HertzBeat scrapes this endpoint directly — no SDK or agent required.

Default endpoint: http://<host>:9091/metrics

Ensure the Milvus metrics port (default 9091) is accessible from the HertzBeat collector. You can verify it manually:

curl http://<milvus-host>:9091/metrics

If Milvus is deployed via Helm on Kubernetes, the metrics port is exposed by default. For standalone deployments it is enabled out of the box.

Reference: https://milvus.io/docs/monitor.md

Configuration Parameters​

Parameter NameDescription
Monitoring HostMonitored IPv4, IPv6 or domain name. Note ⚠️ Do not include protocol headers (e.g. https://, http://)
Monitoring NameIdentifies this monitoring instance. Must be unique.
PortMilvus metrics port. Default: 9091
HTTPSWhether to connect over HTTPS. Default: false
TimeoutConnection and read timeout in milliseconds. Default: 6000
Collection IntervalInterval for periodic data collection, unit: second. Minimum: 30 seconds
Whether to DetectWhether to test availability before saving. The monitor is only saved if the availability check passes.
DescriptionAdditional notes or remarks for this monitoring instance.

Collected Metrics​

Metric Group: milvus_proxy_req_count​

Priority 0 — collected first to determine availability.

Metric NameUnitDescription
function_nameName of the RPC function called on the proxy
statusRequest result status (e.g. OK, fail)
valueNumber of requests

Metric Group: milvus_proxy_slow_query_count​

Priority 1

Metric NameUnitDescription
function_nameName of the slow query RPC function
valueNumber of slow queries recorded

Metric Group: milvus_proxy_search_vectors_count​

Priority 2

Metric NameUnitDescription
valueCumulative number of vectors searched

Metric Group: milvus_proxy_insert_vectors_count​

Priority 3

Metric NameUnitDescription
valueCumulative number of vectors inserted

Metric Group: milvus_proxy_sq_latency_bucket​

Priority 4 — search/query request latency histogram buckets.

Metric NameUnitDescription
leHistogram bucket upper bound (e.g. 1, 5, +Inf)
valuemsNumber of requests whose latency falls within this bucket

Metric Group: milvus_proxy_mutation_latency_bucket​

Priority 5 — insert/delete request latency histogram buckets.

Metric NameUnitDescription
leHistogram bucket upper bound
valuemsNumber of mutation requests whose latency falls within this bucket

Metric Group: milvus_querynode_collection_num​

Priority 6

Metric NameUnitDescription
node_idQueryNode instance identifier
valueNumber of collections currently loaded on this node

Metric Group: milvus_querynode_entity_num​

Priority 7

Metric NameUnitDescription
node_idQueryNode instance identifier
valueNumber of entities currently loaded on this node

Metric Group: milvus_rootcoord_collection_num​

Priority 8

Metric NameUnitDescription
valueTotal number of collections managed by RootCoord

Metric Group: milvus_rootcoord_partition_num​

Priority 9

Metric NameUnitDescription
valueTotal number of partitions managed by RootCoord

Metric Group: milvus_rootcoord_dml_req_count​

Priority 10

Metric NameUnitDescription
statusDML request result status
valueTotal DML (insert/delete) request count

Metric Group: milvus_rootcoord_ddl_req_count​

Priority 11

Metric NameUnitDescription
statusDDL request result status
valueTotal DDL (schema/collection) request count