enhance: [Cherry-pick]Add collection name label for some metric (#36951…

…) (#37159) pr: #36951 Signed-off-by: aoiasd <[email protected]>
milvus-io · Oct 29, 2024 · 8370caa · 8370caa
1 parent 05c4052
commit 8370caa
Show file tree

Hide file tree

Showing 7 changed files with 70 additions and 7 deletions.
diff --git a/internal/datacoord/index_meta.go b/internal/datacoord/index_meta.go
@@ -800,7 +800,7 @@ func (m *indexMeta) SetStoredIndexFileSizeMetric(collections map[UniqueID]*colle
 	for _, segmentIdx := range m.buildID2SegmentIndex {
 		coll, ok := collections[segmentIdx.CollectionID]
 		if ok {
-			metrics.DataCoordStoredIndexFilesSize.WithLabelValues(coll.DatabaseName,
+			metrics.DataCoordStoredIndexFilesSize.WithLabelValues(coll.DatabaseName, coll.Schema.GetName(),
 				fmt.Sprint(segmentIdx.CollectionID), fmt.Sprint(segmentIdx.SegmentID)).Set(float64(segmentIdx.IndexSize))
 			total += segmentIdx.IndexSize
 		}

diff --git a/internal/datacoord/meta.go b/internal/datacoord/meta.go
@@ -437,7 +437,7 @@ func (m *meta) GetQuotaInfo() *metricsinfo.DataCoordQuotaMetrics {
 		coll, ok := m.collections[collectionID]
 		if ok {
 			for state, rows := range statesRows {
-				metrics.DataCoordNumStoredRows.WithLabelValues(coll.DatabaseName, fmt.Sprint(collectionID), state.String()).Set(float64(rows))
+				metrics.DataCoordNumStoredRows.WithLabelValues(coll.DatabaseName, fmt.Sprint(collectionID), coll.Schema.GetName(), state.String()).Set(float64(rows))
 			}
 		}
 	}

diff --git a/internal/querynodev2/metrics_info.go b/internal/querynodev2/metrics_info.go
@@ -59,7 +59,7 @@ func getQuotaMetrics(node *QueryNode) (*metricsinfo.QueryNodeQuotaMetrics, error
 	}
 
 	minTsafeChannel, minTsafe := node.tSafeManager.Min()
-	collections := node.manager.Collection.List()
+	collections := node.manager.Collection.ListWithName()
 	nodeID := fmt.Sprint(node.GetNodeID())
 
 	metrics.QueryNodeNumEntities.Reset()
@@ -70,7 +70,7 @@ func getQuotaMetrics(node *QueryNode) (*metricsinfo.QueryNodeQuotaMetrics, error
 	growingGroupByCollection := lo.GroupBy(growingSegments, func(seg segments.Segment) int64 {
 		return seg.Collection()
 	})
-	for _, collection := range collections {
+	for collection := range collections {
 		segs := growingGroupByCollection[collection]
 		size := lo.SumBy(segs, func(seg segments.Segment) int64 {
 			return seg.MemSize()
@@ -90,6 +90,7 @@ func getQuotaMetrics(node *QueryNode) (*metricsinfo.QueryNodeQuotaMetrics, error
 		segment := segs[0]
 		metrics.QueryNodeNumEntities.WithLabelValues(
 			segment.DatabaseName(),
+			collections[segment.Collection()],
 			nodeID,
 			fmt.Sprint(segment.Collection()),
 			fmt.Sprint(segment.Partition()),
@@ -101,7 +102,7 @@ func getQuotaMetrics(node *QueryNode) (*metricsinfo.QueryNodeQuotaMetrics, error
 	sealedGroupByCollection := lo.GroupBy(sealedSegments, func(seg segments.Segment) int64 {
 		return seg.Collection()
 	})
-	for _, collection := range collections {
+	for collection := range collections {
 		segs := sealedGroupByCollection[collection]
 		size := lo.SumBy(segs, func(seg segments.Segment) int64 {
 			return seg.MemSize()
@@ -119,6 +120,7 @@ func getQuotaMetrics(node *QueryNode) (*metricsinfo.QueryNodeQuotaMetrics, error
 		segment := segs[0]
 		metrics.QueryNodeNumEntities.WithLabelValues(
 			segment.DatabaseName(),
+			collections[segment.Collection()],
 			nodeID,
 			fmt.Sprint(segment.Collection()),
 			fmt.Sprint(segment.Partition()),
@@ -148,7 +150,7 @@ func getQuotaMetrics(node *QueryNode) (*metricsinfo.QueryNodeQuotaMetrics, error
 		GrowingSegmentsSize: totalGrowingSize,
 		Effect: metricsinfo.NodeEffect{
 			NodeID:        node.GetNodeID(),
-			CollectionIDs: collections,
+			CollectionIDs: lo.Keys(collections),
 		},
 		DeleteBufferInfo: metricsinfo.DeleteBufferInfo{
 			CollectionDeleteBufferNum:  deleteBufferNum,

diff --git a/internal/querynodev2/segments/collection.go b/internal/querynodev2/segments/collection.go
@@ -48,6 +48,7 @@ import (
 
 type CollectionManager interface {
 	List() []int64
+	ListWithName() map[int64]string
 	Get(collectionID int64) *Collection
 	PutOrRef(collectionID int64, schema *schemapb.CollectionSchema, meta *segcorepb.CollectionIndexMeta, loadMeta *querypb.LoadMetaInfo)
 	Ref(collectionID int64, count uint32) bool
@@ -75,6 +76,16 @@ func (m *collectionManager) List() []int64 {
 	return lo.Keys(m.collections)
 }
 
+// return all collections by map id --> name
+func (m *collectionManager) ListWithName() map[int64]string {
+	m.mut.RLock()
+	defer m.mut.RUnlock()
+
+	return lo.MapValues(m.collections, func(coll *Collection, _ int64) string {
+		return coll.Schema().GetName()
+	})
+}
+
 func (m *collectionManager) Get(collectionID int64) *Collection {
 	m.mut.RLock()
 	defer m.mut.RUnlock()

diff --git a/internal/querynodev2/segments/mock_collection_manager.go b/internal/querynodev2/segments/mock_collection_manager.go
diff --git a/pkg/metrics/datacoord_metrics.go b/pkg/metrics/datacoord_metrics.go
@@ -95,6 +95,7 @@ var (
 		}, []string{
 			databaseLabelName,
 			collectionIDLabelName,
+			collectionName,
 			segmentStateLabelName,
 		})
 
@@ -162,6 +163,7 @@ var (
 			Help:      "index files size of the segments",
 		}, []string{
 			databaseLabelName,
+			collectionName,
 			collectionIDLabelName,
 			segmentIDLabelName,
 		})
@@ -390,7 +392,7 @@ func CleanupDataCoordSegmentMetrics(dbName string, collectionID int64, segmentID
 		collectionIDLabelName: fmt.Sprint(collectionID),
 		segmentIDLabelName:    fmt.Sprint(segmentID),
 	})
-	DataCoordStoredIndexFilesSize.Delete(prometheus.Labels{
+	DataCoordStoredIndexFilesSize.DeletePartialMatch(prometheus.Labels{
 		databaseLabelName:     dbName,
 		collectionIDLabelName: fmt.Sprint(collectionID),
 		segmentIDLabelName:    fmt.Sprint(segmentID),

diff --git a/pkg/metrics/querynode_metrics.go b/pkg/metrics/querynode_metrics.go
@@ -439,6 +439,7 @@ var (
 			Help:      "number of entities which can be searched/queried, clustered by collection, partition and state",
 		}, []string{
 			databaseLabelName,
+			collectionName,
 			nodeIDLabelName,
 			collectionIDLabelName,
 			partitionIDLabelName,