statistics: fix anticount and cold peer statistics (tikv#5479)

close tikv#5478 Signed-off-by: lhy1024 <[email protected]> Co-authored-by: 混沌DM <[email protected]>
lhy1024 · Sep 13, 2022 · b95e5f8 · b95e5f8
1 parent ea25ac1
commit b95e5f8
Show file tree

Hide file tree

Showing 4 changed files with 20 additions and 15 deletions.
diff --git a/server/cluster/cluster_test.go b/server/cluster/cluster_test.go
@@ -155,8 +155,7 @@ func TestStoreHeartbeat(t *testing.T) {
 	re.Nil(cluster.HandleStoreHeartbeat(hotHeartBeat))
 	time.Sleep(20 * time.Millisecond)
 	storeStats = cluster.hotStat.RegionStats(statistics.Read, 1)
-	re.Len(storeStats[1], 1)
-	re.Equal(uint64(1), storeStats[1][0].RegionID)
+	re.Len(storeStats[1], 0)
 	storeStats = cluster.hotStat.RegionStats(statistics.Read, 3)
 	re.Empty(storeStats[1])
 	// after 2 hot heartbeats, wo can find region 1 peer again
@@ -593,7 +592,7 @@ func TestRegionHeartbeatHotStat(t *testing.T) {
 		EndKey:      []byte{byte(1 + 1)},
 		RegionEpoch: &metapb.RegionEpoch{ConfVer: 2, Version: 2},
 	}
-	region := core.NewRegionInfo(regionMeta, leader, core.WithInterval(&pdpb.TimeInterval{StartTimestamp: 0, EndTimestamp: 10}),
+	region := core.NewRegionInfo(regionMeta, leader, core.WithInterval(&pdpb.TimeInterval{StartTimestamp: 0, EndTimestamp: statistics.RegionHeartBeatReportInterval}),
 		core.SetWrittenBytes(30000*10),
 		core.SetWrittenKeys(300000*10))
 	err = cluster.processRegionHeartbeat(region)

diff --git a/server/statistics/hot_peer.go b/server/statistics/hot_peer.go
@@ -154,7 +154,7 @@ func (stat *HotPeerStat) GetThresholds() []float64 {
 	return stat.thresholds
 }
 
-// Clone clones the HotPeerStat
+// Clone clones the HotPeerStat.
 func (stat *HotPeerStat) Clone() *HotPeerStat {
 	ret := *stat
 	ret.Loads = make([]float64, DimLen)
@@ -204,3 +204,10 @@ func (stat *HotPeerStat) GetStores() []uint64 {
 func (stat *HotPeerStat) IsLearner() bool {
 	return stat.isLearner
 }
+
+func (stat *HotPeerStat) defaultAntiCount() int {
+	if stat.Kind == Read {
+		return hotRegionAntiCount * (RegionHeartBeatReportInterval / StoreHeartBeatReportInterval)
+	}
+	return hotRegionAntiCount
+}
diff --git a/server/statistics/hot_peer_cache.go b/server/statistics/hot_peer_cache.go
@@ -98,7 +98,7 @@ func (f *hotPeerCache) RegionStats(minHotDegree int) map[uint64][]*HotPeerStat {
 		values := peers.GetAll()
 		stat := make([]*HotPeerStat, 0, len(values))
 		for _, v := range values {
-			if peer := v.(*HotPeerStat); peer.HotDegree >= minHotDegree && !peer.inCold {
+			if peer := v.(*HotPeerStat); peer.HotDegree >= minHotDegree && !peer.inCold && peer.AntiCount == peer.defaultAntiCount() {
 				stat = append(stat, peer)
 			}
 		}
@@ -255,8 +255,8 @@ func (f *hotPeerCache) checkColdPeer(storeID uint64, reportRegions map[uint64]*c
 				StoreID:  storeID,
 				RegionID: regionID,
 				Kind:     f.kind,
-				// use oldItem.thresholds to make the newItem won't affect the threshold
-				Loads:          oldItem.thresholds,
+				// use 0 to make the cold newItem won't affect the loads.
+				Loads:          make([]float64, len(oldItem.Loads)),
 				LastUpdateTime: time.Now(),
 				isLeader:       oldItem.isLeader,
 				isLearner:      oldItem.isLearner,
@@ -539,20 +539,18 @@ func coldItem(newItem, oldItem *HotPeerStat) {
 
 func hotItem(newItem, oldItem *HotPeerStat) {
 	newItem.HotDegree = oldItem.HotDegree + 1
-	newItem.AntiCount = hotRegionAntiCount
-	newItem.allowInherited = true
-	if newItem.Kind == Read {
-		newItem.AntiCount = hotRegionAntiCount * (RegionHeartBeatReportInterval / StoreHeartBeatReportInterval)
+	if oldItem.AntiCount < oldItem.defaultAntiCount() {
+		newItem.AntiCount = oldItem.AntiCount + 1
+	} else {
+		newItem.AntiCount = oldItem.AntiCount
 	}
+	newItem.allowInherited = true
 }
 
 func initItem(item *HotPeerStat) {
 	item.HotDegree = 1
-	item.AntiCount = hotRegionAntiCount
+	item.AntiCount = item.defaultAntiCount()
 	item.allowInherited = true
-	if item.Kind == Read {
-		item.AntiCount = hotRegionAntiCount * (RegionHeartBeatReportInterval / StoreHeartBeatReportInterval)
-	}
 }
 
 func inheritItem(newItem, oldItem *HotPeerStat) {

diff --git a/server/statistics/hot_peer_cache_test.go b/server/statistics/hot_peer_cache_test.go
@@ -331,6 +331,7 @@ func TestUpdateHotPeerStat(t *testing.T) {
 	re.Equal(0, newItem.AntiCount)
 	// sum of interval is larger than report interval, and hot
 	oldItem = newItem
+	oldItem.AntiCount = oldItem.defaultAntiCount()
 	newItem = cache.updateHotPeerStat(nil, newItem, oldItem, []float64{60.0, 60.0, 60.0}, 4*time.Second)
 	re.Equal(1, newItem.HotDegree)
 	re.Equal(2*m, newItem.AntiCount)