From 10caef09a743179b3f976e83fa26c3d532f2ee0d Mon Sep 17 00:00:00 2001
From: Ti Chi Robot <ti-community-prow-bot@tidb.io>
Date: Mon, 18 Dec 2023 16:37:54 +0800
Subject: [PATCH] *: provide a option to wait for init stats to finish before
 providing service during startup (#43381) (#43445)

ref pingcap/tidb#42160, close pingcap/tidb#43385
---
 config/config.go            |  6 ++++++
 config/config.toml.example  |  3 +++
 config/config_test.go       |  2 ++
 domain/domain.go            | 23 ++++++++++++++++-------
 statistics/handle/handle.go |  3 +++
 tidb-server/main.go         |  3 +++
 6 files changed, 33 insertions(+), 7 deletions(-)

diff --git a/config/config.go b/config/config.go
index 1ef50edb01fd5..00217028a6acc 100644
--- a/config/config.go
+++ b/config/config.go
@@ -683,6 +683,11 @@ type Performance struct {
 	MemoryUsageAlarmRatio float64 `toml:"memory-usage-alarm-ratio" json:"memory-usage-alarm-ratio"`
 
 	EnableLoadFMSketch bool `toml:"enable-load-fmsketch" json:"enable-load-fmsketch"`
+
+	// If ForceInitStats is true, when tidb starts up, it doesn't provide service until init stats is finished.
+	// If ForceInitStats is false, tidb can provide service before init stats is finished. Note that during the period
+	// of init stats the optimizer may make bad decisions due to pseudo stats.
+	ForceInitStats bool `toml:"force-init-stats" json:"force-init-stats"`
 }
 
 // PlanCache is the PlanCache section of the config.
@@ -941,6 +946,7 @@ var defaultConf = Config{
 		EnableStatsCacheMemQuota:         false,
 		RunAutoAnalyze:                   true,
 		EnableLoadFMSketch:               false,
+		ForceInitStats:                   false,
 	},
 	ProxyProtocol: ProxyProtocol{
 		Networks:      "",
diff --git a/config/config.toml.example b/config/config.toml.example
index 2d3b1acd0b29c..f9ae48fcf6fc3 100644
--- a/config/config.toml.example
+++ b/config/config.toml.example
@@ -255,6 +255,9 @@ max-txn-ttl = 3600000
 # If you find the CPU used by GC is too high or GC is too frequent and impact your business you can increase this value.
 gogc = 100
 
+# Whether to wait for init stats to finish before providing service during startup
+force-init-stats = false
+
 [proxy-protocol]
 # PROXY protocol acceptable client networks.
 # Empty string means disable PROXY protocol, * means all networks.
diff --git a/config/config_test.go b/config/config_test.go
index f4909ee15c23f..b4b439744045e 100644
--- a/config/config_test.go
+++ b/config/config_test.go
@@ -736,6 +736,7 @@ tidb-max-reuse-column = 20
 txn-total-size-limit=2000
 tcp-no-delay = false
 enable-load-fmsketch = true
+force-init-stats = true
 [tikv-client]
 commit-timeout="41s"
 max-batch-size=128
@@ -827,6 +828,7 @@ max_connections = 200
 	require.Equal(t, 10240, conf.Status.GRPCInitialWindowSize)
 	require.Equal(t, 40960, conf.Status.GRPCMaxSendMsgSize)
 	require.True(t, conf.Performance.EnableLoadFMSketch)
+	require.True(t, conf.Performance.ForceInitStats)
 
 	err = f.Truncate(0)
 	require.NoError(t, err)
diff --git a/domain/domain.go b/domain/domain.go
index 874c74c16146b..055c9e6bc8601 100644
--- a/domain/domain.go
+++ b/domain/domain.go
@@ -1927,6 +1927,20 @@ func (do *Domain) newOwnerManager(prompt, ownerKey string) owner.Manager {
 	return statsOwner
 }
 
+func (do *Domain) initStats() {
+	statsHandle := do.StatsHandle()
+	defer func() {
+		close(statsHandle.InitStatsDone)
+	}()
+	t := time.Now()
+	err := statsHandle.InitStats(do.InfoSchema())
+	if err != nil {
+		logutil.BgLogger().Error("init stats info failed", zap.Duration("take time", time.Since(t)), zap.Error(err))
+	} else {
+		logutil.BgLogger().Info("init stats info time", zap.Duration("take time", time.Since(t)))
+	}
+}
+
 func (do *Domain) loadStatsWorker() {
 	defer util.Recover(metrics.LabelDomain, "loadStatsWorker", nil, false)
 	lease := do.statsLease
@@ -1938,14 +1952,9 @@ func (do *Domain) loadStatsWorker() {
 		loadTicker.Stop()
 		logutil.BgLogger().Info("loadStatsWorker exited.")
 	}()
+	do.initStats()
 	statsHandle := do.StatsHandle()
-	t := time.Now()
-	err := statsHandle.InitStats(do.InfoSchema())
-	if err != nil {
-		logutil.BgLogger().Error("init stats info failed", zap.Duration("take time", time.Since(t)), zap.Error(err))
-	} else {
-		logutil.BgLogger().Info("init stats info time", zap.Duration("take time", time.Since(t)))
-	}
+	var err error
 	for {
 		select {
 		case <-loadTicker.C:
diff --git a/statistics/handle/handle.go b/statistics/handle/handle.go
index 22da2184ace78..ff2a56935b266 100644
--- a/statistics/handle/handle.go
+++ b/statistics/handle/handle.go
@@ -128,6 +128,8 @@ type Handle struct {
 	serverIDGetter func() uint64
 	// tableLocked used to store locked tables
 	tableLocked []int64
+
+	InitStatsDone chan struct{}
 }
 
 // GetTableLockedAndClearForTest for unit test only
@@ -476,6 +478,7 @@ func NewHandle(ctx, initStatsCtx sessionctx.Context, lease time.Duration, pool s
 		pool:             pool,
 		sysProcTracker:   tracker,
 		serverIDGetter:   serverIDGetter,
+		InitStatsDone:    make(chan struct{}),
 	}
 	handle.initStatsCtx = initStatsCtx
 	handle.lease.Store(lease)
diff --git a/tidb-server/main.go b/tidb-server/main.go
index 80bcf40392b6c..4ff93417a107a 100644
--- a/tidb-server/main.go
+++ b/tidb-server/main.go
@@ -232,6 +232,9 @@ func main() {
 		close(exited)
 	})
 	topsql.SetupTopSQL()
+	if config.GetGlobalConfig().Performance.ForceInitStats {
+		<-dom.StatsHandle().InitStatsDone
+	}
 	terror.MustNil(svr.Run())
 	<-exited
 	syncLog()