statistics: upgrade stats timeout checkpoint after it timeouts (pingcap#52424) (pingcap#52438)

close pingcap#52425

Author: ti-chi-bot

planner/core/plan_stats.go

@@ -36,7 +36,7 @@ func (collectPredicateColumnsPoint) optimize(_ context.Context, plan LogicalPlan
 		return plan, nil
 	}
 	predicateNeeded := variable.EnableColumnTracking.Load()
-	syncWait := plan.SCtx().GetSessionVars().StatsLoadSyncWait * time.Millisecond.Nanoseconds()
+	syncWait := plan.SCtx().GetSessionVars().StatsLoadSyncWait.Load() * time.Millisecond.Nanoseconds()
 	histNeeded := syncWait > 0
 	predicateColumns, histNeededColumns := CollectColumnStatsUsage(plan, predicateNeeded, histNeeded)
 	if len(predicateColumns) > 0 {

sessionctx/variable/session.go

@@ -1196,7 +1196,7 @@ type SessionVars struct {
 	ReadConsistency ReadConsistencyLevel
 
 	// StatsLoadSyncWait indicates how long to wait for stats load before timeout.
-	StatsLoadSyncWait int64
+	StatsLoadSyncWait atomic.Int64
 
 	// SysdateIsNow indicates whether Sysdate is an alias of Now function
 	SysdateIsNow bool
@@ -1710,7 +1710,6 @@ func NewSessionVars(hctx HookContext) *SessionVars {
 		TMPTableSize:                  DefTiDBTmpTableMaxSize,
 		MPPStoreFailTTL:               DefTiDBMPPStoreFailTTL,
 		Rng:                           mathutil.NewWithTime(),
-		StatsLoadSyncWait:             StatsLoadSyncWait.Load(),
 		EnableLegacyInstanceScope:     DefEnableLegacyInstanceScope,
 		RemoveOrderbyInSubquery:       DefTiDBRemoveOrderbyInSubquery,
 		EnableSkewDistinctAgg:         DefTiDBSkewDistinctAgg,
@@ -1762,6 +1761,7 @@ func NewSessionVars(hctx HookContext) *SessionVars {
 	vars.DiskTracker = disk.NewTracker(memory.LabelForSession, -1)
 	vars.MemTracker = memory.NewTracker(memory.LabelForSession, vars.MemQuotaQuery)
 	vars.MemTracker.IsRootTrackerOfSess = true
+	vars.StatsLoadSyncWait.Store(StatsLoadSyncWait.Load())
 
 	for _, engine := range config.GetGlobalConfig().IsolationRead.Engines {
 		switch engine {

sessionctx/variable/sysvar.go

@@ -1941,7 +1941,7 @@ var defaultSysVars = []*SysVar{
 	}},
 	{Scope: ScopeGlobal | ScopeSession, Name: TiDBStatsLoadSyncWait, Value: strconv.Itoa(DefTiDBStatsLoadSyncWait), Type: TypeInt, MinValue: 0, MaxValue: math.MaxInt32,
 		SetSession: func(s *SessionVars, val string) error {
-			s.StatsLoadSyncWait = TidbOptInt64(val, DefTiDBStatsLoadSyncWait)
+			s.StatsLoadSyncWait.Store(TidbOptInt64(val, DefTiDBStatsLoadSyncWait))
 			return nil
 		},
 		GetGlobal: func(_ context.Context, s *SessionVars) (string, error) {

statistics/handle/handle_hist.go

@@ -259,6 +259,7 @@ func (h *Handle) HandleOneTask(lastTask *NeededItemTask, readerCtx *StatsReaderC
 		}
 		return task, sr.Err
 	case <-time.After(timeout):
+		task.ToTimeout.Add(time.Duration(h.mu.ctx.GetSessionVars().StatsLoadSyncWait.Load()) * time.Microsecond)
 		return task, nil
 	}
 }
@@ -445,6 +446,7 @@ func (h *Handle) drainColTask(exit chan struct{}) (*NeededItemTask, error) {
 			// if the task has already timeout, no sql is sync-waiting for it,
 			// so do not handle it just now, put it to another channel with lower priority
 			if time.Now().After(task.ToTimeout) {
+				task.ToTimeout.Add(time.Duration(h.mu.ctx.GetSessionVars().StatsLoadSyncWait.Load()) * time.Microsecond)
 				h.writeToTimeoutChan(h.StatsLoad.TimeoutItemsCh, task)
 				continue
 			}