ddl: release memory when adding index on partitions finish (pingcap#3…

…9190)

close pingcap#39189

ddl/backfilling.go

@@ -620,6 +620,15 @@ func (dc *ddlCtx) writePhysicalTableRecord(sessPool *sessionPool, t table.Physic
 	}()
 	jc := dc.jobContext(job)
 
+	var ingestBeCtx *ingest.BackendContext
+	if bfWorkerType == typeAddIndexWorker && job.ReorgMeta.ReorgTp == model.ReorgTypeLitMerge {
+		if bc, ok := ingest.LitBackCtxMgr.Load(job.ID); ok {
+			ingestBeCtx = bc
+		} else {
+			return errors.New(ingest.LitErrGetBackendFail)
+		}
+	}
+
 	for {
 		kvRanges, err := splitTableRanges(t, reorgInfo.d.store, startKey, endKey)
 		if err != nil {
@@ -661,7 +670,7 @@ func (dc *ddlCtx) writePhysicalTableRecord(sessPool *sessionPool, t table.Physic
 			case typeAddIndexWorker:
 				idxWorker, err := newAddIndexWorker(sessCtx, i, t, decodeColMap, reorgInfo, jc, job)
 				if err != nil {
-					return errors.Trace(err)
+					return handleCreateBackfillWorkerErr(err, len(backfillWorkers), reorgInfo.ID)
 				}
 				backfillWorkers = append(backfillWorkers, idxWorker.backfillWorker)
 				go idxWorker.backfillWorker.run(reorgInfo.d, idxWorker, job)
@@ -716,14 +725,11 @@ func (dc *ddlCtx) writePhysicalTableRecord(sessPool *sessionPool, t table.Physic
 			zap.Int("regionCnt", len(kvRanges)),
 			zap.String("startKey", hex.EncodeToString(startKey)),
 			zap.String("endKey", hex.EncodeToString(endKey)))
-		if bfWorkerType == typeAddIndexWorker && job.ReorgMeta.ReorgTp == model.ReorgTypeLitMerge {
-			if bc, ok := ingest.LitBackCtxMgr.Load(job.ID); ok {
-				err := bc.Flush(reorgInfo.currElement.ID)
-				if err != nil {
-					return errors.Trace(err)
-				}
-			} else {
-				return errors.New(ingest.LitErrGetBackendFail)
+
+		if ingestBeCtx != nil {
+			err := ingestBeCtx.Flush(reorgInfo.currElement.ID)
+			if err != nil {
+				return errors.Trace(err)
 			}
 		}
 		remains, err := dc.handleRangeTasks(sessPool, t, backfillWorkers, reorgInfo, &totalAddedCount, kvRanges)
@@ -732,13 +738,26 @@ func (dc *ddlCtx) writePhysicalTableRecord(sessPool *sessionPool, t table.Physic
 		}
 
 		if len(remains) == 0 {
+			if ingestBeCtx != nil {
+				ingestBeCtx.EngMgr.ResetWorkers(ingestBeCtx, job.ID, reorgInfo.currElement.ID)
+			}
 			break
 		}
 		startKey = remains[0].StartKey
 	}
 	return nil
 }
 
+func handleCreateBackfillWorkerErr(err error, workerCnt int, jobID int64) error {
+	if workerCnt == 0 {
+		return errors.Trace(err)
+	}
+	logutil.BgLogger().Warn("[ddl] create add index backfill worker failed",
+		zap.Int("current worker count", workerCnt),
+		zap.Int64("job ID", jobID), zap.Error(err))
+	return nil
+}
+
 // recordIterFunc is used for low-level record iteration.
 type recordIterFunc func(h kv.Handle, rowKey kv.Key, rawRecord []byte) (more bool, err error)
 

ddl/index.go

@@ -758,6 +758,7 @@ func tryFallbackToTxnMerge(job *model.Job, err error) {
 		logutil.BgLogger().Info("[ddl] fallback to txn-merge backfill process", zap.Error(err))
 		job.ReorgMeta.ReorgTp = model.ReorgTypeTxnMerge
 		job.SnapshotVer = 0
+		job.RowCount = 0
 	}
 }
 
@@ -782,8 +783,10 @@ func doReorgWorkForCreateIndex(w *worker, d *ddlCtx, t *meta.Meta, job *model.Jo
 	}
 	switch indexInfo.BackfillState {
 	case model.BackfillStateRunning:
-		logutil.BgLogger().Info("[ddl] index backfill state running", zap.Int64("job ID", job.ID),
-			zap.String("table", tbl.Meta().Name.O), zap.String("index", indexInfo.Name.O))
+		logutil.BgLogger().Info("[ddl] index backfill state running",
+			zap.Int64("job ID", job.ID), zap.String("table", tbl.Meta().Name.O),
+			zap.Bool("ingest mode", bfProcess == model.ReorgTypeLitMerge),
+			zap.String("index", indexInfo.Name.O))
 		switch bfProcess {
 		case model.ReorgTypeLitMerge:
 			bc, ok := ingest.LitBackCtxMgr.Load(job.ID)
@@ -1204,7 +1207,7 @@ func newAddIndexWorker(sessCtx sessionctx.Context, id int, t table.PhysicalTable
 		}
 		ei, err := bc.EngMgr.Register(bc, job, reorgInfo.currElement.ID)
 		if err != nil {
-			return nil, errors.Trace(errors.New(ingest.LitErrCreateEngineFail))
+			return nil, errors.Trace(err)
 		}
 		lwCtx, err = ei.NewWriterCtx(id)
 		if err != nil {

ddl/ingest/engine_mgr.go

@@ -58,7 +58,9 @@ func (m *engineManager) Register(bc *BackendContext, job *model.Job, indexID int
 		cfg := generateLocalEngineConfig(job.ID, job.SchemaName, job.TableName)
 		openedEn, err := bc.backend.OpenEngine(bc.ctx, cfg, job.TableName, int32(indexID))
 		if err != nil {
-			return nil, errors.New(LitErrCreateEngineFail)
+			logutil.BgLogger().Warn(LitErrCreateEngineFail, zap.Int64("job ID", job.ID),
+				zap.Int64("index ID", indexID), zap.Error(err))
+			return nil, errors.Trace(err)
 		}
 		id := openedEn.GetEngineUUID()
 		en = NewEngineInfo(bc.ctx, job.ID, indexID, cfg, openedEn, id, 1, m.MemRoot, m.DiskRoot)
@@ -99,6 +101,20 @@ func (m *engineManager) Unregister(jobID, indexID int64) {
 	m.MemRoot.Release(StructSizeEngineInfo)
 }
 
+// ResetWorkers reset the writer count of the engineInfo because
+// the goroutines of backfill workers have been terminated.
+func (m *engineManager) ResetWorkers(bc *BackendContext, jobID, indexID int64) {
+	ei, exist := m.Load(indexID)
+	if !exist {
+		return
+	}
+	m.MemRoot.Release(StructSizeWriterCtx * int64(ei.writerCount))
+	m.MemRoot.ReleaseWithTag(encodeEngineTag(jobID, indexID))
+	engineCacheSize := int64(bc.cfg.TikvImporter.EngineMemCacheSize)
+	m.MemRoot.ConsumeWithTag(encodeEngineTag(jobID, indexID), engineCacheSize)
+	ei.writerCount = 0
+}
+
 // UnregisterAll delete all engineInfo from the engineManager.
 func (m *engineManager) UnregisterAll(jobID int64) {
 	for _, idxID := range m.Keys() {

ddl/ingest/mem_root.go

@@ -122,7 +122,7 @@ func (m *memRootImpl) ConsumeWithTag(tag string, size int64) {
 	m.structSize[tag] = size
 }
 
-// TestConsume implements MemRoot.
+// CheckConsume implements MemRoot.
 func (m *memRootImpl) CheckConsume(size int64) bool {
 	m.mu.RLock()
 	defer m.mu.RUnlock()

ddl/ingest/message.go

@@ -28,7 +28,7 @@ const (
 	LitErrStatDirFail       string = "[ddl-ingest] stat lightning sort path error"
 	LitErrDeleteDirFail     string = "[ddl-ingest] delete lightning sort path error"
 	LitErrCreateBackendFail string = "[ddl-ingest] build lightning backend failed, will use kernel index reorg method to backfill the index"
-	LitErrGetBackendFail    string = "[ddl-ingest]: Can not get cached backend"
+	LitErrGetBackendFail    string = "[ddl-ingest] can not get cached backend"
 	LitErrCreateEngineFail  string = "[ddl-ingest] build lightning engine failed, will use kernel index reorg method to backfill the index"
 	LitErrCreateContextFail string = "[ddl-ingest] build lightning worker context failed, will use kernel index reorg method to backfill the index"
 	LitErrGetEngineFail     string = "[ddl-ingest] can not get cached engine info"

ddl/reorg.go

@@ -234,7 +234,7 @@ func (w *worker) runReorgJob(rh *reorgHandler, reorgInfo *reorgInfo, tblInfo *mo
 			return dbterror.ErrCancelledDDLJob
 		}
 		rowCount, _, _ := rc.getRowCountAndKey()
-		logutil.BgLogger().Info("[ddl] run reorg job done", zap.Int64("handled rows", rowCount))
+		logutil.BgLogger().Info("[ddl] run reorg job done", zap.Int64("handled rows", rowCount), zap.Error(err))
 		job.SetRowCount(rowCount)
 
 		// Update a job's warnings.

tests/realtikvtest/addindextest/integration_test.go

@@ -97,3 +97,54 @@ func TestAddIndexIngestLimitOneBackend(t *testing.T) {
 	require.True(t, strings.Contains(rows[0][3].(string) /* job_type */, "ingest"))
 	require.Equal(t, rows[0][7].(string) /* row_count */, "3")
 }
+
+func TestAddIndexIngestWriterCountOnPartitionTable(t *testing.T) {
+	store := realtikvtest.CreateMockStoreAndSetup(t)
+	tk := testkit.NewTestKit(t, store)
+	tk.MustExec("drop database if exists addindexlit;")
+	tk.MustExec("create database addindexlit;")
+	tk.MustExec("use addindexlit;")
+	tk.MustExec(`set global tidb_ddl_enable_fast_reorg=on;`)
+
+	tk.MustExec("create table t (a int primary key) partition by hash(a) partitions 32;")
+	var sb strings.Builder
+	sb.WriteString("insert into t values ")
+	for i := 0; i < 100; i++ {
+		sb.WriteString(fmt.Sprintf("(%d)", i))
+		if i != 99 {
+			sb.WriteString(",")
+		}
+	}
+	tk.MustExec(sb.String())
+	tk.MustExec("alter table t add index idx(a);")
+	rows := tk.MustQuery("admin show ddl jobs 1;").Rows()
+	require.Len(t, rows, 1)
+	jobTp := rows[0][3].(string)
+	require.True(t, strings.Contains(jobTp, "ingest"), jobTp)
+}
+
+func TestAddIndexIngestAdjustBackfillWorkerCountFail(t *testing.T) {
+	store := realtikvtest.CreateMockStoreAndSetup(t)
+	tk := testkit.NewTestKit(t, store)
+	tk.MustExec("drop database if exists addindexlit;")
+	tk.MustExec("create database addindexlit;")
+	tk.MustExec("use addindexlit;")
+	tk.MustExec(`set global tidb_ddl_enable_fast_reorg=on;`)
+	tk.MustExec("set @@global.tidb_ddl_reorg_worker_cnt = 20;")
+	tk.MustExec("create table t (a int primary key);")
+	var sb strings.Builder
+	sb.WriteString("insert into t values ")
+	for i := 0; i < 20; i++ {
+		sb.WriteString(fmt.Sprintf("(%d000)", i))
+		if i != 19 {
+			sb.WriteString(",")
+		}
+	}
+	tk.MustExec(sb.String())
+	tk.MustQuery("split table t between (0) and (20000) regions 20;").Check(testkit.Rows("19 1"))
+	tk.MustExec("alter table t add index idx(a);")
+	rows := tk.MustQuery("admin show ddl jobs 1;").Rows()
+	require.Len(t, rows, 1)
+	jobTp := rows[0][3].(string)
+	require.True(t, strings.Contains(jobTp, "ingest"), jobTp)
+}