statistics: make sure PQ can analyze all indexes with stats version 1 (#57258)

close pingcap/tidb#57257

pkg/statistics/handle/autoanalyze/priorityqueue/dynamic_partitioned_table_analysis_job.go

@@ -244,8 +244,10 @@ func (j *DynamicPartitionedTableAnalysisJob) analyzePartitionIndexes(
 	sysProcTracker sysproctrack.Tracker,
 ) (success bool) {
 	analyzePartitionBatchSize := int(variable.AutoAnalyzePartitionBatchSize.Load())
+	// For version 2, analyze one index will analyze all other indexes and columns.
+	// For version 1, analyze one index will only analyze the specified index.
+	analyzeVersion := sctx.GetSessionVars().AnalyzeVersion
 
-OnlyPickOneIndex:
 	for indexName, partitionNames := range j.PartitionIndexes {
 		needAnalyzePartitionNames := make([]any, 0, len(partitionNames))
 		for _, partition := range partitionNames {
@@ -262,10 +264,16 @@ OnlyPickOneIndex:
 			params := append([]any{j.TableSchema, j.GlobalTableName}, needAnalyzePartitionNames[start:end]...)
 			params = append(params, indexName)
 			success = exec.AutoAnalyze(sctx, statsHandle, sysProcTracker, j.TableStatsVer, sql, params...)
+			if !success {
+				return false
+			}
+		}
+		// For version 1, we need to analyze all indexes.
+		if analyzeVersion != 1 {
 			// Halt execution after analyzing one index.
 			// This is because analyzing a single index also analyzes all other indexes and columns.
 			// Therefore, to avoid redundancy, we prevent multiple analyses of the same partition.
-			break OnlyPickOneIndex
+			break
 		}
 	}
 	return

pkg/statistics/handle/autoanalyze/priorityqueue/non_partitioned_table_analysis_job.go

@@ -212,6 +212,18 @@ func (j *NonPartitionedTableAnalysisJob) analyzeIndexes(
 	if len(j.Indexes) == 0 {
 		return true
 	}
+	// For version 2, analyze one index will analyze all other indexes and columns.
+	// For version 1, analyze one index will only analyze the specified index.
+	analyzeVersion := sctx.GetSessionVars().AnalyzeVersion
+	if analyzeVersion == 1 {
+		for _, index := range j.Indexes {
+			sql, params := j.GenSQLForAnalyzeIndex(index)
+			if !exec.AutoAnalyze(sctx, statsHandle, sysProcTracker, j.TableStatsVer, sql, params...) {
+				return false
+			}
+		}
+		return true
+	}
 	// Only analyze the first index.
 	// This is because analyzing a single index also analyzes all other indexes and columns.
 	// Therefore, to avoid redundancy, we prevent multiple analyses of the same table.

pkg/statistics/handle/autoanalyze/priorityqueue/queue.go

@@ -376,7 +376,7 @@ func (pq *AnalysisPriorityQueue) ProcessDMLChanges() {
 
 		// Only update if we've seen a newer version
 		if newMaxVersion > lastFetchTimestamp {
-			statslogutil.StatsLogger().Info("Updating last fetch timestamp", zap.Uint64("new_max_version", newMaxVersion))
+			queueSamplerLogger().Info("Updating last fetch timestamp", zap.Uint64("new_max_version", newMaxVersion))
 			pq.syncFields.lastDMLUpdateFetchTimestamp = newMaxVersion
 		}
 		return nil
@@ -404,12 +404,6 @@ func (pq *AnalysisPriorityQueue) processTableStats(
 		return errors.Trace(err)
 	}
 	jobFactory := NewAnalysisJobFactory(sctx, autoAnalyzeRatio, currentTs)
-	// Check if the table is needed to be analyzed.
-	// Note: Unanalyzed tables will also be considered.
-	changePercent := jobFactory.CalculateChangePercentage(stats)
-	if changePercent == 0 {
-		return nil
-	}
 	is := sctx.GetDomainInfoSchema().(infoschema.InfoSchema)
 	pruneMode := variable.PartitionPruneMode(sctx.GetSessionVars().PartitionPruneMode.Load())
 
@@ -455,7 +449,6 @@ func (pq *AnalysisPriorityQueue) tryCreateJob(
 	}
 
 	tableInfo, ok := pq.statsHandle.TableInfoByID(is, stats.PhysicalID)
-	tableMeta := tableInfo.Meta()
 	if !ok {
 		statslogutil.StatsLogger().Warn(
 			"Table info not found for table id",
@@ -463,6 +456,7 @@ func (pq *AnalysisPriorityQueue) tryCreateJob(
 		)
 		return nil
 	}
+	tableMeta := tableInfo.Meta()
 	schemaName, ok := is.SchemaNameByTableID(tableMeta.ID)
 	if !ok {
 		statslogutil.StatsLogger().Warn(

pkg/statistics/handle/autoanalyze/priorityqueue/queue_ddl_handler_test.go

@@ -902,3 +902,48 @@ func TestVectorIndexTriggerAutoAnalyze(t *testing.T) {
 	// No event is found
 	require.Nil(t, addIndexEvent)
 }
+
+func TestAddIndexTriggerAutoAnalyzeWithStatsVersion1(t *testing.T) {
+	store, do := testkit.CreateMockStoreAndDomain(t)
+	testKit := testkit.NewTestKit(t, store)
+	testKit.MustExec("set @@global.tidb_analyze_version=1;")
+	testKit.MustExec("use test")
+	testKit.MustExec("create table t (c1 int, c2 int, index idx(c1, c2)) partition by range columns (c1) (partition p0 values less than (5), partition p1 values less than (10))")
+	is := do.InfoSchema()
+	tbl, err := is.TableByName(context.Background(), pmodel.NewCIStr("test"), pmodel.NewCIStr("t"))
+	require.NoError(t, err)
+	tableInfo := tbl.Meta()
+	h := do.StatsHandle()
+	// Analyze table.
+	testKit.MustExec("analyze table t")
+	require.NoError(t, h.Update(context.Background(), do.InfoSchema()))
+	// Insert some data.
+	testKit.MustExec("insert into t values (1,2),(2,2)")
+	require.NoError(t, h.DumpStatsDeltaToKV(true))
+	require.NoError(t, h.Update(context.Background(), do.InfoSchema()))
+	// Add two indexes.
+	testKit.MustExec("alter table t add index idx1(c1)")
+	testKit.MustExec("alter table t add index idx2(c2)")
+
+	statistics.AutoAnalyzeMinCnt = 0
+	defer func() {
+		statistics.AutoAnalyzeMinCnt = 1000
+	}()
+
+	pq := priorityqueue.NewAnalysisPriorityQueue(h)
+	defer pq.Close()
+	require.NoError(t, pq.Initialize())
+	isEmpty, err := pq.IsEmpty()
+	require.NoError(t, err)
+	require.False(t, isEmpty)
+	job, err := pq.Peek()
+	require.NoError(t, err)
+	require.Equal(t, tableInfo.ID, job.GetTableID())
+	require.NoError(t, job.Analyze(h, do.SysProcTracker()))
+
+	// Check the stats of the indexes.
+	tableStats := h.GetTableStats(tableInfo)
+	require.True(t, tableStats.GetIdx(1).IsAnalyzed())
+	require.True(t, tableStats.GetIdx(2).IsAnalyzed())
+	require.True(t, tableStats.GetIdx(3).IsAnalyzed())
+}

pkg/statistics/handle/autoanalyze/priorityqueue/static_partitioned_table_analysis_job.go

@@ -225,6 +225,18 @@ func (j *StaticPartitionedTableAnalysisJob) analyzeStaticPartitionIndexes(
 	if len(j.Indexes) == 0 {
 		return true
 	}
+	// For version 2, analyze one index will analyze all other indexes and columns.
+	// For version 1, analyze one index will only analyze the specified index.
+	analyzeVersion := sctx.GetSessionVars().AnalyzeVersion
+	if analyzeVersion == 1 {
+		for _, index := range j.Indexes {
+			sql, params := j.GenSQLForAnalyzeStaticPartitionIndex(index)
+			if !exec.AutoAnalyze(sctx, statsHandle, sysProcTracker, j.TableStatsVer, sql, params...) {
+				return false
+			}
+		}
+		return true
+	}
 	// Only analyze the first index.
 	// This is because analyzing a single index also analyzes all other indexes and columns.
 	// Therefore, to avoid redundancy, we prevent multiple analyses of the same partition.