*: read downstream table before CREATE TABLE to fix IF NOT EXISTS

loader/loader.go

@@ -1529,6 +1529,7 @@ func (l *Loader) cleanDumpFiles() {
 		var lastErr error
 		for f := range files {
 			if strings.HasSuffix(f, ".sql") {
+				// TODO: table structure files are not used now, but we plan to used them in future so not delete them
 				if strings.HasSuffix(f, "-schema-create.sql") || strings.HasSuffix(f, "-schema.sql") {
 					continue
 				}

syncer/syncer.go

@@ -1393,11 +1393,7 @@ func (s *Syncer) Run(ctx context.Context) (err error) {
 	if s.cfg.Mode == config.ModeAll && fresh {
 		delLoadTask = true
 		flushCheckpoint = true
-		err = s.loadTableStructureFromDump(ctx)
-		if err != nil {
-			tctx.L().Warn("error happened when load table structure from dump files", zap.Error(err))
-			cleanDumpFile = false
-		}
+		// TODO: loadTableStructureFromDump in future
 	} else {
 		cleanDumpFile = false
 	}
@@ -2525,6 +2521,7 @@ func (s *Syncer) trackDDL(usedSchema string, sql string, tableNames [][]*filter.
 		shouldSchemaExist            bool
 		shouldTableExistNum          int // tableNames[:shouldTableExistNum] should exist
 		shouldRefTableExistNum       int // tableNames[1:shouldTableExistNum] should exist, since first one is "caller table"
+		tryFetchDownstreamTable      bool // to make sure if not exists will execute correctly
 	)
 
 	switch node := stmt.(type) {
@@ -2548,6 +2545,7 @@ func (s *Syncer) trackDDL(usedSchema string, sql string, tableNames [][]*filter.
 		shouldSchemaExist = true
 		// for CREATE TABLE LIKE/AS, the reference tables should exist
 		shouldRefTableExistNum = len(srcTables)
+		tryFetchDownstreamTable = true
 	case *ast.DropTableStmt:
 		shouldExecDDLOnSchemaTracker = true
 		if err := s.checkpoint.DeleteTablePoint(ec.tctx, srcTable.Schema, srcTable.Name); err != nil {
@@ -2602,6 +2600,11 @@ func (s *Syncer) trackDDL(usedSchema string, sql string, tableNames [][]*filter.
 		}
 	}
 
+	if tryFetchDownstreamTable {
+		// ignore table not exists error, just try to fetch table from downstream.
+		_, _ = s.getTable(ec.tctx, srcTables[0].Schema, srcTables[0].Name, targetTables[0].Schema, targetTables[0].Name)
+	}
+
 	if shouldExecDDLOnSchemaTracker {
 		if err := s.schemaTracker.Exec(ec.tctx.Ctx, usedSchema, sql); err != nil {
 			ec.tctx.L().Error("cannot track DDL", zap.String("schema", usedSchema), zap.String("statement", sql), log.WrapStringerField("location", ec.currentLocation), log.ShortError(err))

tests/shardddl1_1/run.sh

@@ -252,11 +252,11 @@ function DM_027_CASE() {
 	run_sql_source1 "insert into ${shardddl1}.${tb2} values (4)"
 	run_sql_source1 "insert into ${shardddl1}.${tb3} values (5,6)"
 	# we now haven't checked table struct when create sharding table
-	# there should be a error message like "Unknown column 'val' in 'field list'", "unknown column val"
-	# but different TiDB version output different message. so we only roughly match here
+	# and we'll attach IF NOT EXISTS to every CREATE TABLE and fetch downstream table first, so downstream table strcuture
+	# is in use indeed. This leads to the error below.
 	run_dm_ctl_with_retry $WORK_DIR "127.0.0.1:$MASTER_PORT" \
 		"query-status test" \
-		"nknown column" 1 # ignore case for first letter
+		"Column count doesn't match value count: 1 (columns) vs 2 (values)" 1
 }
 
 function DM_027() {