statistics: add gc in hot peer cache (tikv#8702) tikv#8750

Signed-off-by: lhy1024 <admin@liudos.us>
ti-chi-bot · Oct 30, 2024 · 63a02b0 · 63a02b0
1 parent 4fbf433
commit 63a02b0
Show file tree

Hide file tree

Showing 10 changed files with 188 additions and 80 deletions.
diff --git a/pkg/mock/mockcluster/mockcluster.go b/pkg/mock/mockcluster/mockcluster.go
@@ -62,10 +62,11 @@ type Cluster struct {
 
 // NewCluster creates a new Cluster
 func NewCluster(ctx context.Context, opts *config.PersistOptions) *Cluster {
+	basicCluster:= core.NewBasicCluster()
 	clus := &Cluster{
-		BasicCluster:       core.NewBasicCluster(),
+		BasicCluster:       basicCluster,
 		IDAllocator:        mockid.NewIDAllocator(),
-		HotStat:            statistics.NewHotStat(ctx),
+		HotStat:            statistics.NewHotStat(ctx, basicCluster),
 		HotBucketCache:     buckets.NewBucketsCache(ctx),
 		PersistOptions:     opts,
 		suspectRegions:     map[uint64]struct{}{},

diff --git a/pkg/schedule/schedulers/hot_region_test.go b/pkg/schedule/schedulers/hot_region_test.go
@@ -1494,6 +1494,9 @@ func TestHotCacheUpdateCache(t *testing.T) {
 	re := require.New(t)
 	cancel, _, tc, _ := prepareSchedulersTest()
 	defer cancel()
+	for i := 0; i < 3; i++ {
+		tc.PutStore(core.NewStoreInfo(&metapb.Store{Id: uint64(i + 1)}))
+	}
 	tc.SetHotRegionCacheHitsThreshold(0)
 
 	// For read flow
@@ -1561,6 +1564,9 @@ func TestHotCacheKeyThresholds(t *testing.T) {
 	{ // only a few regions
 		cancel, _, tc, _ := prepareSchedulersTest()
 		defer cancel()
+		for i := 0; i < 6; i++ {
+			tc.PutStore(core.NewStoreInfo(&metapb.Store{Id: uint64(i + 1)}))
+		}
 		tc.SetHotRegionCacheHitsThreshold(0)
 		addRegionInfo(tc, statistics.Read, []testRegionInfo{
 			{1, []uint64{1, 2, 3}, 0, 1, 0},
@@ -1580,6 +1586,9 @@ func TestHotCacheKeyThresholds(t *testing.T) {
 	{ // many regions
 		cancel, _, tc, _ := prepareSchedulersTest()
 		defer cancel()
+		for i := 0; i < 3; i++ {
+			tc.PutStore(core.NewStoreInfo(&metapb.Store{Id: uint64(i + 1)}))
+		}
 		regions := []testRegionInfo{}
 		for i := 1; i <= 1000; i += 2 {
 			regions = append(regions,
@@ -1633,6 +1642,9 @@ func TestHotCacheByteAndKey(t *testing.T) {
 	re := require.New(t)
 	cancel, _, tc, _ := prepareSchedulersTest()
 	defer cancel()
+	for i := 0; i < 3; i++ {
+		tc.PutStore(core.NewStoreInfo(&metapb.Store{Id: uint64(i + 1)}))
+	}
 	tc.SetHotRegionCacheHitsThreshold(0)
 	statistics.ThresholdsUpdateInterval = 0
 	defer func() {
@@ -1760,6 +1772,9 @@ func TestHotCacheCheckRegionFlow(t *testing.T) {
 func checkHotCacheCheckRegionFlow(re *require.Assertions, testCase testHotCacheCheckRegionFlowCase, enablePlacementRules bool) {
 	cancel, _, tc, oc := prepareSchedulersTest()
 	defer cancel()
+	for i := 0; i < 3; i++ {
+		tc.PutStore(core.NewStoreInfo(&metapb.Store{Id: uint64(i + 1)}))
+	}
 	tc.SetClusterVersion(versioninfo.MinSupportedVersion(versioninfo.Version4_0))
 	tc.SetEnablePlacementRules(enablePlacementRules)
 	labels := []string{"zone", "host"}
@@ -1835,6 +1850,9 @@ func TestHotCacheCheckRegionFlowWithDifferentThreshold(t *testing.T) {
 func checkHotCacheCheckRegionFlowWithDifferentThreshold(re *require.Assertions, enablePlacementRules bool) {
 	cancel, _, tc, _ := prepareSchedulersTest()
 	defer cancel()
+	for i := 0; i < 3; i++ {
+		tc.PutStore(core.NewStoreInfo(&metapb.Store{Id: uint64(i + 1)}))
+	}
 	tc.SetClusterVersion(versioninfo.MinSupportedVersion(versioninfo.Version4_0))
 	tc.SetEnablePlacementRules(enablePlacementRules)
 	labels := []string{"zone", "host"}

diff --git a/pkg/statistics/hot_cache.go b/pkg/statistics/hot_cache.go
@@ -37,11 +37,11 @@ type HotCache struct {
 }
 
 // NewHotCache creates a new hot spot cache.
-func NewHotCache(ctx context.Context) *HotCache {
+func NewHotCache(ctx context.Context, cluster *core.BasicCluster) *HotCache {
 	w := &HotCache{
 		ctx:        ctx,
-		writeCache: NewHotPeerCache(ctx, Write),
-		readCache:  NewHotPeerCache(ctx, Read),
+		writeCache: NewHotPeerCache(ctx, cluster, Write),
+		readCache:  NewHotPeerCache(ctx, cluster, Read),
 	}
 	go w.updateItems(w.readCache.taskQueue, w.runReadTask)
 	go w.updateItems(w.writeCache.taskQueue, w.runWriteTask)

diff --git a/pkg/statistics/hot_peer_cache.go b/pkg/statistics/hot_peer_cache.go
@@ -76,20 +76,22 @@ type thresholds struct {
 // hotPeerCache saves the hot peer's statistics.
 type hotPeerCache struct {
 	kind              RWType
+	cluster           *core.BasicCluster
 	peersOfStore      map[uint64]*TopN               // storeID -> hot peers
 	storesOfRegion    map[uint64]map[uint64]struct{} // regionID -> storeIDs
 	regionsOfStore    map[uint64]map[uint64]struct{} // storeID -> regionIDs
 	topNTTL           time.Duration
 	taskQueue         *chanx.UnboundedChan[FlowItemTask]
 	thresholdsOfStore map[uint64]*thresholds                     // storeID -> thresholds
 	metrics           map[uint64][ActionTypeLen]prometheus.Gauge // storeID -> metrics
-	// TODO: consider to remove store info when store is offline.
+	lastGCTime        time.Time
 }
 
 // NewHotPeerCache creates a hotPeerCache
-func NewHotPeerCache(ctx context.Context, kind RWType) *hotPeerCache {
+func NewHotPeerCache(ctx context.Context, cluster *core.BasicCluster, kind RWType) *hotPeerCache {
 	return &hotPeerCache{
 		kind:              kind,
+		cluster:           cluster,
 		peersOfStore:      make(map[uint64]*TopN),
 		storesOfRegion:    make(map[uint64]map[uint64]struct{}),
 		regionsOfStore:    make(map[uint64]map[uint64]struct{}),
@@ -130,6 +132,7 @@ func (f *hotPeerCache) updateStat(item *HotPeerStat) {
 		return
 	}
 	f.incMetrics(item.actionType, item.StoreID)
+	f.gc()
 }
 
 func (f *hotPeerCache) incMetrics(action ActionType, storeID uint64) {
@@ -560,6 +563,36 @@ func (f *hotPeerCache) removeItem(item *HotPeerStat) {
 	}
 }
 
+func (f *hotPeerCache) gc() {
+	if time.Since(f.lastGCTime) < f.topNTTL {
+		return
+	}
+	f.lastGCTime = time.Now()
+	// remove tombstone stores
+	stores := make(map[uint64]struct{})
+	for _, storeID := range f.cluster.GetStores() {
+		stores[storeID.GetID()] = struct{}{}
+	}
+	for storeID := range f.peersOfStore {
+		if _, ok := stores[storeID]; !ok {
+			delete(f.peersOfStore, storeID)
+			delete(f.regionsOfStore, storeID)
+			delete(f.thresholdsOfStore, storeID)
+			delete(f.metrics, storeID)
+		}
+	}
+	// remove expired items
+	for _, peers := range f.peersOfStore {
+		regions := peers.RemoveExpired()
+		for _, regionID := range regions {
+			delete(f.storesOfRegion, regionID)
+			for storeID := range f.regionsOfStore {
+				delete(f.regionsOfStore[storeID], regionID)
+			}
+		}
+	}
+}
+
 func (f *hotPeerCache) coldItem(newItem, oldItem *HotPeerStat) {
 	newItem.HotDegree = oldItem.HotDegree - 1
 	newItem.AntiCount = oldItem.AntiCount - 1