checker, statistic: avoid leak in label statistic (#8703) (#8735)

close #8700

Signed-off-by: lhy1024 <[email protected]>

Co-authored-by: lhy1024 <[email protected]>

Author: ti-chi-bot

pkg/cluster/cluster.go

@@ -68,7 +68,7 @@ func HandleOverlaps(ctx context.Context, c Cluster, overlaps []*core.RegionInfo)
 		if c.GetRegionStats() != nil {
 			c.GetRegionStats().ClearDefunctRegion(item.GetID())
 		}
-		c.GetLabelStats().ClearDefunctRegion(item.GetID())
+		c.GetLabelStats().MarkDefunctRegion(item.GetID())
 		c.GetRuleManager().InvalidCache(item.GetID())
 	}
 }

pkg/mcs/scheduling/server/cluster.go

@@ -379,13 +379,12 @@ func (c *Cluster) waitSchedulersInitialized() {
 	}
 }
 
-// TODO: implement the following methods
-
 // UpdateRegionsLabelLevelStats updates the status of the region label level by types.
 func (c *Cluster) UpdateRegionsLabelLevelStats(regions []*core.RegionInfo) {
 	for _, region := range regions {
 		c.labelStats.Observe(region, c.getStoresWithoutLabelLocked(region, core.EngineKey, core.EngineTiFlash), c.persistConfig.GetLocationLabels())
 	}
+	c.labelStats.ClearDefunctRegions()
 }
 
 func (c *Cluster) getStoresWithoutLabelLocked(region *core.RegionInfo, key, value string) []*core.StoreInfo {

pkg/statistics/region_collection.go

@@ -365,13 +365,15 @@ type LabelStatistics struct {
 	syncutil.RWMutex
 	regionLabelStats map[uint64]string
 	labelCounter     map[string]int
+	defunctRegions   map[uint64]struct{}
 }
 
 // NewLabelStatistics creates a new LabelStatistics.
 func NewLabelStatistics() *LabelStatistics {
 	return &LabelStatistics{
 		regionLabelStats: make(map[uint64]string),
 		labelCounter:     make(map[string]int),
+		defunctRegions:   make(map[uint64]struct{}),
 	}
 }
 
@@ -405,14 +407,26 @@ func ResetLabelStatsMetrics() {
 	regionLabelLevelGauge.Reset()
 }
 
-// ClearDefunctRegion is used to handle the overlap region.
-func (l *LabelStatistics) ClearDefunctRegion(regionID uint64) {
+// MarkDefunctRegion is used to handle the overlap region.
+// It is used to mark the region as defunct and remove it from the label statistics later.
+func (l *LabelStatistics) MarkDefunctRegion(regionID uint64) {
 	l.Lock()
 	defer l.Unlock()
-	if label, ok := l.regionLabelStats[regionID]; ok {
-		l.labelCounter[label]--
-		delete(l.regionLabelStats, regionID)
+	l.defunctRegions[regionID] = struct{}{}
+}
+
+// ClearDefunctRegions is used to handle the overlap region.
+// It is used to remove the defunct regions from the label statistics.
+func (l *LabelStatistics) ClearDefunctRegions() {
+	l.Lock()
+	defer l.Unlock()
+	for regionID := range l.defunctRegions {
+		if label, ok := l.regionLabelStats[regionID]; ok {
+			l.labelCounter[label]--
+			delete(l.regionLabelStats, regionID)
+		}
 	}
+	l.defunctRegions = make(map[uint64]struct{})
 }
 
 // GetLabelCounter is only used for tests.

server/cluster/cluster_test.go

@@ -1125,6 +1125,7 @@ func TestRegionLabelIsolationLevel(t *testing.T) {
 	opt.SetReplicationConfig(cfg)
 	re.NoError(err)
 	cluster := newTestRaftCluster(ctx, mockid.NewIDAllocator(), opt, storage.NewStorageWithMemoryBackend())
+	cluster.coordinator = schedule.NewCoordinator(ctx, cluster, nil)
 
 	for i := uint64(1); i <= 4; i++ {
 		var labels []*metapb.StoreLabel
@@ -1159,13 +1160,42 @@ func TestRegionLabelIsolationLevel(t *testing.T) {
 		StartKey: []byte{byte(1)},
 		EndKey:   []byte{byte(2)},
 	}
-	r := core.NewRegionInfo(region, peers[0])
-	re.NoError(cluster.putRegion(r))
+	r1 := core.NewRegionInfo(region, peers[0])
+	re.NoError(cluster.putRegion(r1))
 
-	cluster.UpdateRegionsLabelLevelStats([]*core.RegionInfo{r})
+	cluster.UpdateRegionsLabelLevelStats([]*core.RegionInfo{r1})
 	counter := cluster.labelStats.GetLabelCounter()
 	re.Equal(0, counter["none"])
 	re.Equal(1, counter["zone"])
+
+	region = &metapb.Region{
+		Id:       10,
+		Peers:    peers,
+		StartKey: []byte{byte(2)},
+		EndKey:   []byte{byte(3)},
+	}
+	r2 := core.NewRegionInfo(region, peers[0])
+	re.NoError(cluster.putRegion(r2))
+
+	cluster.UpdateRegionsLabelLevelStats([]*core.RegionInfo{r2})
+	counter = cluster.labelStats.GetLabelCounter()
+	re.Equal(0, counter["none"])
+	re.Equal(2, counter["zone"])
+
+	// issue: https://github.com/tikv/pd/issues/8700
+	// step1: heartbeat a overlap region, which is used to simulate the case that the region is merged.
+	// step2: update region 9 and region 10, which is used to simulate the case that patrol is triggered.
+	// We should only count region 9.
+	overlapRegion := r1.Clone(
+		core.WithStartKey(r1.GetStartKey()),
+		core.WithEndKey(r2.GetEndKey()),
+		core.WithLeader(r2.GetPeer(8)),
+	)
+	re.NoError(cluster.HandleRegionHeartbeat(overlapRegion))
+	cluster.UpdateRegionsLabelLevelStats([]*core.RegionInfo{r1, r2})
+	counter = cluster.labelStats.GetLabelCounter()
+	re.Equal(0, counter["none"])
+	re.Equal(1, counter["zone"])
 }
 
 func heartbeatRegions(re *require.Assertions, cluster *RaftCluster, regions []*core.RegionInfo) {

server/cluster/scheduling_controller.go

@@ -236,6 +236,7 @@ func (sc *schedulingController) UpdateRegionsLabelLevelStats(regions []*core.Reg
 	for _, region := range regions {
 		sc.labelStats.Observe(region, sc.getStoresWithoutLabelLocked(region, core.EngineKey, core.EngineTiFlash), sc.opt.GetLocationLabels())
 	}
+	sc.labelStats.ClearDefunctRegions()
 }
 
 func (sc *schedulingController) getStoresWithoutLabelLocked(region *core.RegionInfo, key, value string) []*core.StoreInfo {