enhance: Add collection name label for some metric (#36951)

Signed-off-by: aoiasd <zhicheng.yue@zilliz.com>

internal/datacoord/index_meta.go

@@ -824,7 +824,7 @@ 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
 		}

internal/datacoord/meta.go

@@ -446,7 +446,7 @@ 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))
 			}
 		}
 	}

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,

internal/querynodev2/segments/collection.go

@@ -47,6 +47,7 @@ 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
@@ -74,6 +75,16 @@ 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()

internal/querynodev2/segments/mock_collection_manager.go

@@ -118,6 +118,53 @@ func (_c *MockCollectionManager_List_Call) RunAndReturn(run func() []int64) *Moc
 	return _c
 }
 
+// ListWithName provides a mock function with given fields:
+func (_m *MockCollectionManager) ListWithName() map[int64]string {
+	ret := _m.Called()
+
+	if len(ret) == 0 {
+		panic("no return value specified for ListWithName")
+	}
+
+	var r0 map[int64]string
+	if rf, ok := ret.Get(0).(func() map[int64]string); ok {
+		r0 = rf()
+	} else {
+		if ret.Get(0) != nil {
+			r0 = ret.Get(0).(map[int64]string)
+		}
+	}
+
+	return r0
+}
+
+// MockCollectionManager_ListWithName_Call is a *mock.Call that shadows Run/Return methods with type explicit version for method 'ListWithName'
+type MockCollectionManager_ListWithName_Call struct {
+	*mock.Call
+}
+
+// ListWithName is a helper method to define mock.On call
+func (_e *MockCollectionManager_Expecter) ListWithName() *MockCollectionManager_ListWithName_Call {
+	return &MockCollectionManager_ListWithName_Call{Call: _e.mock.On("ListWithName")}
+}
+
+func (_c *MockCollectionManager_ListWithName_Call) Run(run func()) *MockCollectionManager_ListWithName_Call {
+	_c.Call.Run(func(args mock.Arguments) {
+		run()
+	})
+	return _c
+}
+
+func (_c *MockCollectionManager_ListWithName_Call) Return(_a0 map[int64]string) *MockCollectionManager_ListWithName_Call {
+	_c.Call.Return(_a0)
+	return _c
+}
+
+func (_c *MockCollectionManager_ListWithName_Call) RunAndReturn(run func() map[int64]string) *MockCollectionManager_ListWithName_Call {
+	_c.Call.Return(run)
+	return _c
+}
+
 // PutOrRef provides a mock function with given fields: collectionID, schema, meta, loadMeta
 func (_m *MockCollectionManager) PutOrRef(collectionID int64, schema *schemapb.CollectionSchema, meta *segcorepb.CollectionIndexMeta, loadMeta *querypb.LoadMetaInfo) {
 	_m.Called(collectionID, schema, meta, loadMeta)

pkg/metrics/datacoord_metrics.go

@@ -96,6 +96,7 @@ var (
 		}, []string{
 			databaseLabelName,
 			collectionIDLabelName,
+			collectionName,
 			segmentStateLabelName,
 		})
 
@@ -163,6 +164,7 @@ var (
 			Help:      "index files size of the segments",
 		}, []string{
 			databaseLabelName,
+			collectionName,
 			collectionIDLabelName,
 			segmentIDLabelName,
 		})
@@ -403,7 +405,7 @@ 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),

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,