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

close #8698

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

pkg/mcs/scheduling/server/cluster.go

@@ -93,7 +93,7 @@ func NewCluster(parentCtx context.Context, persistConfig *config.PersistConfig,
 		ruleManager:       ruleManager,
 		labelerManager:    labelerManager,
 		persistConfig:     persistConfig,
-		hotStat:           statistics.NewHotStat(ctx),
+		hotStat:           statistics.NewHotStat(ctx, basicCluster),
 		labelStats:        statistics.NewLabelStatistics(),
 		regionStats:       statistics.NewRegionStatistics(basicCluster, persistConfig, ruleManager),
 		storage:           storage,

pkg/mock/mockcluster/mockcluster.go

@@ -63,11 +63,12 @@ type Cluster struct {
 
 // NewCluster creates a new Cluster
 func NewCluster(ctx context.Context, opts *config.PersistOptions) *Cluster {
+	bc := core.NewBasicCluster()
 	c := &Cluster{
 		ctx:                     ctx,
-		BasicCluster:            core.NewBasicCluster(),
+		BasicCluster:            bc,
 		IDAllocator:             mockid.NewIDAllocator(),
-		HotStat:                 statistics.NewHotStat(ctx),
+		HotStat:                 statistics.NewHotStat(ctx, bc),
 		HotBucketCache:          buckets.NewBucketsCache(ctx),
 		PersistOptions:          opts,
 		pendingProcessedRegions: map[uint64]struct{}{},

pkg/schedule/schedulers/hot_region_test.go

@@ -1614,6 +1614,9 @@ func TestHotCacheUpdateCache(t *testing.T) {
 	re := require.New(t)
 	cancel, _, tc, _ := prepareSchedulersTest()
 	defer cancel()
+	for i := range 3 {
+		tc.PutStore(core.NewStoreInfo(&metapb.Store{Id: uint64(i + 1)}))
+	}
 
 	// For read flow
 	addRegionInfo(tc, utils.Read, []testRegionInfo{
@@ -1680,6 +1683,9 @@ func TestHotCacheKeyThresholds(t *testing.T) {
 	{ // only a few regions
 		cancel, _, tc, _ := prepareSchedulersTest()
 		defer cancel()
+		for i := range 6 {
+			tc.PutStore(core.NewStoreInfo(&metapb.Store{Id: uint64(i + 1)}))
+		}
 		addRegionInfo(tc, utils.Read, []testRegionInfo{
 			{1, []uint64{1, 2, 3}, 0, 1, 0},
 			{2, []uint64{1, 2, 3}, 0, 1 * units.KiB, 0},
@@ -1698,6 +1704,9 @@ func TestHotCacheKeyThresholds(t *testing.T) {
 	{ // many regions
 		cancel, _, tc, _ := prepareSchedulersTest()
 		defer cancel()
+		for i := range 3 {
+			tc.PutStore(core.NewStoreInfo(&metapb.Store{Id: uint64(i + 1)}))
+		}
 		regions := []testRegionInfo{}
 		for i := 1; i <= 1000; i += 2 {
 			regions = append(regions,
@@ -1751,6 +1760,9 @@ func TestHotCacheByteAndKey(t *testing.T) {
 	re := require.New(t)
 	cancel, _, tc, _ := prepareSchedulersTest()
 	defer cancel()
+	for i := range 3 {
+		tc.PutStore(core.NewStoreInfo(&metapb.Store{Id: uint64(i + 1)}))
+	}
 	statistics.ThresholdsUpdateInterval = 0
 	defer func() {
 		statistics.ThresholdsUpdateInterval = 8 * time.Second
@@ -1877,6 +1889,9 @@ func TestHotCacheCheckRegionFlow(t *testing.T) {
 func checkHotCacheCheckRegionFlow(re *require.Assertions, testCase testHotCacheCheckRegionFlowCase, enablePlacementRules bool) {
 	cancel, _, tc, oc := prepareSchedulersTest()
 	defer cancel()
+	for i := range 3 {
+		tc.PutStore(core.NewStoreInfo(&metapb.Store{Id: uint64(i + 1)}))
+	}
 	tc.SetClusterVersion(versioninfo.MinSupportedVersion(versioninfo.Version4_0))
 	tc.SetEnablePlacementRules(enablePlacementRules)
 	labels := []string{"zone", "host"}
@@ -1953,6 +1968,9 @@ func TestHotCacheCheckRegionFlowWithDifferentThreshold(t *testing.T) {
 func checkHotCacheCheckRegionFlowWithDifferentThreshold(re *require.Assertions, enablePlacementRules bool) {
 	cancel, _, tc, _ := prepareSchedulersTest()
 	defer cancel()
+	for i := range 3 {
+		tc.PutStore(core.NewStoreInfo(&metapb.Store{Id: uint64(i + 1)}))
+	}
 	tc.SetClusterVersion(versioninfo.MinSupportedVersion(versioninfo.Version4_0))
 	tc.SetEnablePlacementRules(enablePlacementRules)
 	labels := []string{"zone", "host"}

pkg/statistics/hot_cache.go

@@ -39,11 +39,11 @@ 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, utils.Write),
-		readCache:  NewHotPeerCache(ctx, utils.Read),
+		writeCache: NewHotPeerCache(ctx, cluster, utils.Write),
+		readCache:  NewHotPeerCache(ctx, cluster, utils.Read),
 	}
 	go w.updateItems(w.readCache.taskQueue, w.runReadTask)
 	go w.updateItems(w.writeCache.taskQueue, w.runWriteTask)

pkg/statistics/hot_cache_test.go

@@ -19,6 +19,7 @@ import (
 	"testing"
 
 	"github.com/stretchr/testify/require"
+	"github.com/tikv/pd/pkg/core"
 	"github.com/tikv/pd/pkg/statistics/utils"
 )
 
@@ -27,8 +28,9 @@ func TestIsHot(t *testing.T) {
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()
 	for i := utils.RWType(0); i < utils.RWTypeLen; i++ {
-		cache := NewHotCache(ctx)
-		region := buildRegion(i, 3, 60)
+		cluster := core.NewBasicCluster()
+		cache := NewHotCache(ctx, cluster)
+		region := buildRegion(cluster, i, 3, 60)
 		stats := cache.CheckReadPeerSync(region, region.GetPeers(), []float64{100000000, 1000, 1000}, 60)
 		cache.Update(stats[0], i)
 		for range 100 {

pkg/statistics/hot_peer_cache.go

@@ -60,20 +60,22 @@ type thresholds struct {
 // HotPeerCache saves the hot peer's statistics.
 type HotPeerCache struct {
 	kind              utils.RWType
+	cluster           *core.BasicCluster
 	peersOfStore      map[uint64]*utils.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[func(*HotPeerCache)]
 	thresholdsOfStore map[uint64]*thresholds                           // storeID -> thresholds
 	metrics           map[uint64][utils.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 utils.RWType) *HotPeerCache {
+func NewHotPeerCache(ctx context.Context, cluster *core.BasicCluster, kind utils.RWType) *HotPeerCache {
 	return &HotPeerCache{
 		kind:              kind,
+		cluster:           cluster,
 		peersOfStore:      make(map[uint64]*utils.TopN),
 		storesOfRegion:    make(map[uint64]map[uint64]struct{}),
 		regionsOfStore:    make(map[uint64]map[uint64]struct{}),
@@ -115,6 +117,7 @@ func (f *HotPeerCache) UpdateStat(item *HotPeerStat) {
 		return
 	}
 	f.incMetrics(item.actionType, item.StoreID)
+	f.gc()
 }
 
 func (f *HotPeerCache) incMetrics(action utils.ActionType, storeID uint64) {
@@ -548,6 +551,36 @@ 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)
+			}
+		}
+	}
+}
+
 // removeAllItem removes all items of the cache.
 // It is used for test.
 func (f *HotPeerCache) removeAllItem() {