statistics: add upper bound of retry for sync load

pkg/statistics/handle/syncload/BUILD.bazel

@@ -30,12 +30,13 @@ go_test(
     srcs = ["stats_syncload_test.go"],
     flaky = True,
     race = "on",
-    shard_count = 4,
+    shard_count = 5,
     deps = [
         "//pkg/config",
         "//pkg/parser/model",
         "//pkg/sessionctx",
         "//pkg/sessionctx/stmtctx",
+        "//pkg/statistics/handle/types",
         "//pkg/testkit",
         "//pkg/util/mathutil",
         "@com_github_pingcap_failpoint//:failpoint",

pkg/statistics/handle/syncload/stats_syncload.go

@@ -223,46 +223,58 @@ func (s *statsSyncLoad) HandleOneTask(sctx sessionctx.Context, lastTask *statsty
 	} else {
 		task = lastTask
 	}
+	result := stmtctx.StatsLoadResult{Item: task.Item.TableItemID}
 	resultChan := s.StatsLoad.Singleflight.DoChan(task.Item.Key(), func() (any, error) {
-		return s.handleOneItemTask(sctx, task)
+		err := s.handleOneItemTask(sctx, task)
+		return nil, err
 	})
 	timeout := time.Until(task.ToTimeout)
 	select {
-	case result := <-resultChan:
-		if result.Err == nil {
-			slr := result.Val.(*stmtctx.StatsLoadResult)
-			if slr.Error != nil {
-				return task, slr.Error
-			}
-			task.ResultCh <- *slr
+	case sr := <-resultChan:
+		if sr.Err == nil {
+			task.ResultCh <- result
+			return nil, nil
+		}
+		if !updateNeededItemTaskRetryCountAndCheck(task) {
+			result.Error = sr.Err
+			task.ResultCh <- result
 			return nil, nil
 		}
-		return task, result.Err
+		return task, sr.Err
 	case <-time.After(timeout):
+		if !updateNeededItemTaskRetryCountAndCheck(task) {
+			result.Error = errors.New("stats loading timeout")
+			task.ResultCh <- result
+			return nil, nil
+		}
 		task.ToTimeout.Add(time.Duration(sctx.GetSessionVars().StatsLoadSyncWait.Load()) * time.Microsecond)
 		return task, nil
 	}
 }
 
-func (s *statsSyncLoad) handleOneItemTask(sctx sessionctx.Context, task *statstypes.NeededItemTask) (result *stmtctx.StatsLoadResult, err error) {
+func updateNeededItemTaskRetryCountAndCheck(task *statstypes.NeededItemTask) bool {
+	task.Retry++
+	return task.Retry <= 3
+}
+
+func (s *statsSyncLoad) handleOneItemTask(sctx sessionctx.Context, task *statstypes.NeededItemTask) (err error) {
 	defer func() {
 		// recover for each task, worker keeps working
 		if r := recover(); r != nil {
 			logutil.BgLogger().Error("handleOneItemTask panicked", zap.Any("recover", r), zap.Stack("stack"))
 			err = errors.Errorf("stats loading panicked: %v", r)
 		}
 	}()
-	result = &stmtctx.StatsLoadResult{Item: task.Item.TableItemID}
-	item := result.Item
+	item := task.Item.TableItemID
 	tbl, ok := s.statsHandle.Get(item.TableID)
 	if !ok {
-		return result, nil
+		return nil
 	}
 	wrapper := &statsWrapper{}
 	if item.IsIndex {
 		index, loadNeeded := tbl.IndexIsLoadNeeded(item.ID)
 		if !loadNeeded {
-			return result, nil
+			return nil
 		}
 		if index != nil {
 			wrapper.idxInfo = index.Info
@@ -272,7 +284,7 @@ func (s *statsSyncLoad) handleOneItemTask(sctx sessionctx.Context, task *statsty
 	} else {
 		col, loadNeeded, analyzed := tbl.ColumnIsLoadNeeded(item.ID, task.Item.FullLoad)
 		if !loadNeeded {
-			return result, nil
+			return nil
 		}
 		if col != nil {
 			wrapper.colInfo = col.Info
@@ -288,18 +300,15 @@ func (s *statsSyncLoad) handleOneItemTask(sctx sessionctx.Context, task *statsty
 				Histogram:  *statistics.NewHistogram(item.ID, 0, 0, 0, &wrapper.colInfo.FieldType, 0, 0),
 				IsHandle:   tbl.IsPkIsHandle && mysql.HasPriKeyFlag(wrapper.colInfo.GetFlag()),
 			}
-			if s.updateCachedItem(item, wrapper.col, wrapper.idx, task.Item.FullLoad) {
-				return result, nil
-			}
-			return nil, nil
+			s.updateCachedItem(item, wrapper.col, wrapper.idx, task.Item.FullLoad)
+			return nil
 		}
 	}
 	t := time.Now()
 	needUpdate := false
 	wrapper, err = s.readStatsForOneItem(sctx, item, wrapper, tbl.IsPkIsHandle, task.Item.FullLoad)
 	if err != nil {
-		result.Error = err
-		return result, err
+		return err
 	}
 	if item.IsIndex {
 		if wrapper.idxInfo != nil {
@@ -311,10 +320,10 @@ func (s *statsSyncLoad) handleOneItemTask(sctx sessionctx.Context, task *statsty
 		}
 	}
 	metrics.ReadStatsHistogram.Observe(float64(time.Since(t).Milliseconds()))
-	if needUpdate && s.updateCachedItem(item, wrapper.col, wrapper.idx, task.Item.FullLoad) {
-		return result, nil
+	if needUpdate {
+		s.updateCachedItem(item, wrapper.col, wrapper.idx, task.Item.FullLoad)
 	}
-	return nil, nil
+	return nil
 }
 
 // readStatsForOneItem reads hist for one column/index, TODO load data via kv-get asynchronously
@@ -492,14 +501,14 @@ func (s *statsSyncLoad) updateCachedItem(item model.TableItemID, colHist *statis
 	// like `GetPartitionStats` called in `fmSketchFromStorage` would have modified the stats cache already.
 	tbl, ok := s.statsHandle.Get(item.TableID)
 	if !ok {
-		return true
+		return false
 	}
 	if !item.IsIndex && colHist != nil {
 		c, ok := tbl.Columns[item.ID]
 		// - If the stats is fully loaded,
 		// - If the stats is meta-loaded and we also just need the meta.
 		if ok && (c.IsFullLoad() || !fullLoaded) {
-			return true
+			return false
 		}
 		tbl = tbl.Copy()
 		tbl.Columns[item.ID] = colHist

pkg/statistics/handle/syncload/stats_syncload_test.go

@@ -23,6 +23,7 @@ import (
 	"github.com/pingcap/tidb/pkg/parser/model"
 	"github.com/pingcap/tidb/pkg/sessionctx"
 	"github.com/pingcap/tidb/pkg/sessionctx/stmtctx"
+	"github.com/pingcap/tidb/pkg/statistics/handle/types"
 	"github.com/pingcap/tidb/pkg/testkit"
 	"github.com/pingcap/tidb/pkg/util/mathutil"
 	"github.com/stretchr/testify/require"
@@ -229,3 +230,68 @@ func TestConcurrentLoadHistWithPanicAndFail(t *testing.T) {
 		require.Greater(t, hg.Len()+topn.Num(), 0)
 	}
 }
+
+func TestRetry(t *testing.T) {
+	originConfig := config.GetGlobalConfig()
+	newConfig := config.NewConfig()
+	newConfig.Performance.StatsLoadConcurrency = 0 // no worker to consume channel
+	config.StoreGlobalConfig(newConfig)
+	defer config.StoreGlobalConfig(originConfig)
+	store, dom := testkit.CreateMockStoreAndDomain(t)
+
+	testKit := testkit.NewTestKit(t, store)
+	testKit.MustExec("use test")
+	testKit.MustExec("drop table if exists t")
+	testKit.MustExec("set @@session.tidb_analyze_version=2")
+	testKit.MustExec("create table t(a int, b int, c int, primary key(a), key idx(b))")
+	testKit.MustExec("insert into t values (1,1,1),(2,2,2),(3,3,3)")
+
+	oriLease := dom.StatsHandle().Lease()
+	dom.StatsHandle().SetLease(1)
+	defer func() {
+		dom.StatsHandle().SetLease(oriLease)
+	}()
+	testKit.MustExec("analyze table t")
+
+	is := dom.InfoSchema()
+	tbl, err := is.TableByName(model.NewCIStr("test"), model.NewCIStr("t"))
+	require.NoError(t, err)
+	tableInfo := tbl.Meta()
+
+	h := dom.StatsHandle()
+
+	neededColumns := make([]model.StatsLoadItem, 1)
+	neededColumns[0] = model.StatsLoadItem{TableItemID: model.TableItemID{TableID: tableInfo.ID, ID: tableInfo.Columns[2].ID, IsIndex: false}, FullLoad: true}
+	timeout := time.Nanosecond * mathutil.MaxInt
+
+	// clear statsCache
+	h.Clear()
+	require.NoError(t, dom.StatsHandle().Update(is))
+
+	// no stats at beginning
+	stat := h.GetTableStats(tableInfo)
+	c, ok := stat.Columns[tableInfo.Columns[2].ID]
+	require.True(t, !ok || (c.Histogram.Len()+c.TopN.Num() == 0))
+
+	stmtCtx1 := stmtctx.NewStmtCtx()
+	h.SendLoadRequests(stmtCtx1, neededColumns, timeout)
+	stmtCtx2 := stmtctx.NewStmtCtx()
+	h.SendLoadRequests(stmtCtx2, neededColumns, timeout)
+
+	exitCh := make(chan struct{})
+	require.NoError(t, failpoint.Enable("github.com/pingcap/tidb/pkg/statistics/handle/syncload/mockReadStatsForOneFail", "return(true)"))
+	var (
+		task1 *types.NeededItemTask
+		err1  error
+	)
+
+	for i := 0; i < 3; i++ {
+		task1, err1 = h.HandleOneTask(testKit.Session().(sessionctx.Context), task1, exitCh)
+		require.Error(t, err1)
+		require.NotNil(t, task1)
+	}
+	task1, err1 = h.HandleOneTask(testKit.Session().(sessionctx.Context), task1, exitCh)
+	require.NoError(t, err1)
+	require.Nil(t, task1)
+	require.NoError(t, failpoint.Disable("github.com/pingcap/tidb/pkg/statistics/handle/syncload/mockReadStatsForOneFail"))
+}

pkg/statistics/handle/types/interfaces.go

@@ -370,6 +370,7 @@ type NeededItemTask struct {
 	ToTimeout time.Time
 	ResultCh  chan stmtctx.StatsLoadResult
 	Item      model.StatsLoadItem
+	Retry     int
 }
 
 // StatsLoad is used to load stats concurrently