Fix collection metrics leakage (#19944)

Signed-off-by: Nxiaofan-luan <xiaofan.luan@zilliz.com>
Signed-off-by: Nxiaofan-luan <xiaofan.luan@zilliz.com>

internal/proxy/impl.go

@@ -110,12 +110,14 @@ func (node *Proxy) InvalidateCollectionMetaCache(ctx context.Context, request *p
 
 	collectionName := request.CollectionName
 	collectionID := request.CollectionID
+
+	var aliasName []string
 	if globalMetaCache != nil {
 		if collectionName != "" {
 			globalMetaCache.RemoveCollection(ctx, collectionName) // no need to return error, though collection may be not cached
 		}
 		if request.CollectionID != UniqueID(0) {
-			globalMetaCache.RemoveCollectionsByID(ctx, collectionID)
+			aliasName = globalMetaCache.RemoveCollectionsByID(ctx, collectionID)
 		}
 	}
 	if request.GetBase().GetMsgType() == commonpb.MsgType_DropCollection {
@@ -123,6 +125,9 @@ func (node *Proxy) InvalidateCollectionMetaCache(ctx context.Context, request *p
 		node.chMgr.removeDMLStream(request.GetCollectionID())
 		// clean up collection level metrics
 		metrics.CleanupCollectionMetrics(Params.ProxyCfg.GetNodeID(), collectionName)
+		for _, alias := range aliasName {
+			metrics.CleanupCollectionMetrics(Params.ProxyCfg.GetNodeID(), alias)
+		}
 	}
 	logutil.Logger(ctx).Info("complete to invalidate collection meta cache",
 		zap.String("role", typeutil.ProxyRole),

internal/proxy/meta_cache.go

@@ -63,7 +63,7 @@ type Cache interface {
 	GetShards(ctx context.Context, withCache bool, collectionName string) (map[string][]nodeInfo, error)
 	ClearShards(collectionName string)
 	RemoveCollection(ctx context.Context, collectionName string)
-	RemoveCollectionsByID(ctx context.Context, collectionID UniqueID)
+	RemoveCollectionsByID(ctx context.Context, collectionID UniqueID) []string
 	RemovePartition(ctx context.Context, collectionName string, partitionName string)
 
 	// GetCredentialInfo operate credential cache
@@ -515,14 +515,17 @@ func (m *MetaCache) RemoveCollection(ctx context.Context, collectionName string)
 	delete(m.collInfo, collectionName)
 }
 
-func (m *MetaCache) RemoveCollectionsByID(ctx context.Context, collectionID UniqueID) {
+func (m *MetaCache) RemoveCollectionsByID(ctx context.Context, collectionID UniqueID) []string {
 	m.mu.Lock()
 	defer m.mu.Unlock()
+	var collNames []string
 	for k, v := range m.collInfo {
 		if v.collID == collectionID {
 			delete(m.collInfo, k)
+			collNames = append(collNames, k)
 		}
 	}
+	return collNames
 }
 
 func (m *MetaCache) RemovePartition(ctx context.Context, collectionName, partitionName string) {

internal/rootcoord/alter_collection_task.go

@@ -50,9 +50,10 @@ func (a *alterCollectionTask) Execute(ctx context.Context) error {
 	})
 
 	redoTask.AddSyncStep(&expireCacheStep{
-		baseStep:     baseStep{core: a.core},
-		collectionID: oldColl.CollectionID,
-		ts:           ts,
+		baseStep:        baseStep{core: a.core},
+		collectionNames: []string{oldColl.Name},
+		collectionID:    oldColl.CollectionID,
+		ts:              ts,
 	})
 
 	a.Req.CollectionID = oldColl.CollectionID

internal/rootcoord/create_partition_task.go

@@ -56,7 +56,7 @@ func (t *createPartitionTask) Execute(ctx context.Context) error {
 	undoTask := newBaseUndoTask(t.core.stepExecutor)
 	undoTask.AddStep(&expireCacheStep{
 		baseStep:        baseStep{core: t.core},
-		collectionNames: []string{t.Req.GetCollectionName()},
+		collectionNames: []string{t.collMeta.Name},
 		collectionID:    t.collMeta.CollectionID,
 		ts:              t.GetTs(),
 	}, &nullStep{})

internal/rootcoord/drop_partition_task.go

@@ -63,7 +63,7 @@ func (t *dropPartitionTask) Execute(ctx context.Context) error {
 	})
 	redoTask.AddSyncStep(&expireCacheStep{
 		baseStep:        baseStep{core: t.core},
-		collectionNames: []string{t.Req.GetCollectionName()},
+		collectionNames: []string{t.collMeta.Name},
 		collectionID:    t.collMeta.CollectionID,
 		ts:              t.GetTs(),
 	})