*: add metadata lock when using the plan cache (#51897) (#52956)

close #51407

Author: ti-chi-bot

ddl/metadatalocktest/BUILD.bazel

@@ -7,7 +7,7 @@ go_test(
         "mdl_test.go",
     ],
     flaky = True,
-    shard_count = 34,
+    shard_count = 36,
     deps = [
         "//config",
         "//ddl",

ddl/metadatalocktest/mdl_test.go

@@ -872,6 +872,101 @@ func TestMDLPreparePlanCacheInvalid(t *testing.T) {
 	tk.MustQuery(`execute stmt_test_1 using @a;`).Check(testkit.Rows("1 <nil>", "2 <nil>", "3 <nil>", "4 <nil>"))
 }
 
+func TestMDLPreparePlanCacheExecute(t *testing.T) {
+	store, dom := testkit.CreateMockStoreAndDomain(t)
+
+	sv := server.CreateMockServer(t, store)
+
+	sv.SetDomain(dom)
+	dom.InfoSyncer().SetSessionManager(sv)
+	defer sv.Close()
+
+	conn1 := server.CreateMockConn(t, sv)
+	tk := testkit.NewTestKitWithSession(t, store, conn1.Context().Session)
+	conn2 := server.CreateMockConn(t, sv)
+	tkDDL := testkit.NewTestKitWithSession(t, store, conn2.Context().Session)
+	tk.MustExec("use test")
+	tk.MustExec("set global tidb_enable_metadata_lock=1")
+	tk.MustExec("create table t(a int);")
+	tk.MustExec("create table t2(a int);")
+	tk.MustExec("insert into t values(1), (2), (3), (4);")
+
+	tk.MustExec(`prepare stmt_test_1 from 'update t set a = ? where a = ?';`)
+	tk.MustExec(`set @a = 1, @b = 3;`)
+	tk.MustExec(`execute stmt_test_1 using @a, @b;`)
+
+	tk.MustExec("begin")
+
+	ch := make(chan struct{})
+
+	var wg sync.WaitGroup
+	wg.Add(1)
+	go func() {
+		<-ch
+		tkDDL.MustExec("alter table test.t add index idx(a);")
+		wg.Done()
+	}()
+
+	tk.MustQuery("select * from t2")
+	tk.MustExec(`set @a = 2, @b=4;`)
+	tk.MustExec(`execute stmt_test_1 using @a, @b;`)
+	tk.MustQuery("select @@last_plan_from_cache;").Check(testkit.Rows("1"))
+	// The plan is from cache, the metadata lock should be added to block the DDL.
+	ch <- struct{}{}
+
+	time.Sleep(5 * time.Second)
+
+	tk.MustExec("commit")
+
+	wg.Wait()
+
+	tk.MustExec("admin check table t")
+}
+
+func TestMDLPreparePlanCacheExecute2(t *testing.T) {
+	store, dom := testkit.CreateMockStoreAndDomain(t)
+
+	sv := server.CreateMockServer(t, store)
+
+	sv.SetDomain(dom)
+	dom.InfoSyncer().SetSessionManager(sv)
+	defer sv.Close()
+
+	conn1 := server.CreateMockConn(t, sv)
+	tk := testkit.NewTestKitWithSession(t, store, conn1.Context().Session)
+	conn2 := server.CreateMockConn(t, sv)
+	tkDDL := testkit.NewTestKitWithSession(t, store, conn2.Context().Session)
+	tk.MustExec("use test")
+	tk.MustExec("set global tidb_enable_metadata_lock=1")
+	tk.MustExec("create table t(a int);")
+	tk.MustExec("create table t2(a int);")
+	tk.MustExec("insert into t values(1), (2), (3), (4);")
+
+	tk.MustExec(`prepare stmt_test_1 from 'select * from t where a = ?';`)
+	tk.MustExec(`set @a = 1;`)
+	tk.MustExec(`execute stmt_test_1 using @a;`)
+
+	tk.MustExec("begin")
+	tk.MustQuery("select * from t2")
+
+	var wg sync.WaitGroup
+	wg.Add(1)
+	go func() {
+		tkDDL.MustExec("alter table test.t add index idx(a);")
+		wg.Done()
+	}()
+
+	wg.Wait()
+
+	tk.MustExec(`set @a = 2;`)
+	tk.MustExec(`execute stmt_test_1 using @a;`)
+	// The plan should not be from cache because the schema has changed.
+	tk.MustQuery("select @@last_plan_from_cache;").Check(testkit.Rows("0"))
+	tk.MustExec("commit")
+
+	tk.MustExec("admin check table t")
+}
+
 func TestMDLDisable2Enable(t *testing.T) {
 	store, dom := testkit.CreateMockStoreAndDomain(t)
 	sv := server.CreateMockServer(t, store)

domain/plan_replayer_dump.go

@@ -526,6 +526,7 @@ func dumpExplain(ctx sessionctx.Context, zw *zip.Writer, execStmts []ast.StmtNod
 	return nil
 }
 
+// extractTableNames extracts table names from the given stmts.
 func extractTableNames(ctx context.Context, sctx sessionctx.Context,
 	ExecStmts []ast.StmtNode, curDB model.CIStr) (map[tableNamePair]struct{}, error) {
 	tableExtractor := &tableNameExtractor{

executor/executor.go

@@ -1985,6 +1985,7 @@ func ResetContextOfStmt(ctx sessionctx.Context, s ast.StmtNode) (err error) {
 	vars.DiskTracker.ResetMaxConsumed()
 	vars.MemTracker.SessionID.Store(vars.ConnectionID)
 	vars.StmtCtx.TableStats = make(map[int64]interface{})
+	sc.MDLRelatedTableIDs = make(map[int64]int64)
 
 	isAnalyze := false
 	if execStmt, ok := s.(*ast.ExecuteStmt); ok {

executor/prepared.go

@@ -26,6 +26,7 @@ import (
 	"github.com/pingcap/tidb/parser/mysql"
 	plannercore "github.com/pingcap/tidb/planner/core"
 	"github.com/pingcap/tidb/sessionctx"
+	"github.com/pingcap/tidb/sessiontxn"
 	"github.com/pingcap/tidb/types"
 	"github.com/pingcap/tidb/util"
 	"github.com/pingcap/tidb/util/chunk"
@@ -117,7 +118,7 @@ func (e *PrepareExec) Next(ctx context.Context, req *chunk.Chunk) error {
 			return err
 		}
 	}
-	stmt, p, paramCnt, err := plannercore.GeneratePlanCacheStmtWithAST(ctx, e.ctx, stmt0)
+	stmt, p, paramCnt, err := plannercore.GeneratePlanCacheStmtWithAST(ctx, e.ctx, stmt0, sessiontxn.GetTxnManager(e.ctx).GetTxnInfoSchema())
 	if err != nil {
 		return err
 	}
@@ -211,7 +212,7 @@ func (e *DeallocateExec) Next(ctx context.Context, req *chunk.Chunk) error {
 	if e.ctx.GetSessionVars().EnablePreparedPlanCache {
 		bindSQL, _ := plannercore.GetBindSQL4PlanCache(e.ctx, preparedObj)
 		cacheKey, err := plannercore.NewPlanCacheKey(vars, preparedObj.StmtText, preparedObj.StmtDB, prepared.SchemaVersion,
-			0, bindSQL)
+			0, bindSQL, preparedObj.RelateVersion)
 		if err != nil {
 			return err
 		}

executor/seqtest/seq_executor_test.go

@@ -940,8 +940,7 @@ func TestBatchInsertDelete(t *testing.T) {
 		atomic.StoreUint64(&kv.TxnTotalSizeLimit, originLimit)
 	}()
 	// Set the limitation to a small value, make it easier to reach the limitation.
-	atomic.StoreUint64(&kv.TxnTotalSizeLimit, 5800)
-
+	atomic.StoreUint64(&kv.TxnTotalSizeLimit, 7000)
 	tk := testkit.NewTestKit(t, store)
 	tk.MustExec("use test")
 	tk.MustExec("drop table if exists batch_insert")

meta/meta.go

@@ -828,6 +828,8 @@ func (m *Meta) UpdateTable(dbID int64, tableInfo *model.TableInfo) error {
 		return errors.Trace(err)
 	}
 
+	tableInfo.Revision++
+
 	data, err := json.Marshal(tableInfo)
 	if err != nil {
 		return errors.Trace(err)

parser/model/model.go

@@ -567,6 +567,9 @@ type TableInfo struct {
 	ExchangePartitionInfo *ExchangePartitionInfo `json:"exchange_partition_info"`
 
 	TTLInfo *TTLInfo `json:"ttl_info"`
+
+	// Revision is per table schema's version, it will be increased when the schema changed.
+	Revision uint64 `json:"revision"`
 }
 
 // SepAutoInc decides whether _rowid and auto_increment id use separate allocator.

planner/core/plan_cache.go

@@ -76,8 +76,33 @@ func planCachePreprocess(ctx context.Context, sctx sessionctx.Context, isGeneral
 		vars.PreparedParams = append(vars.PreparedParams, val)
 	}
 
-	// step 3: check schema version
-	if stmtAst.SchemaVersion != is.SchemaMetaVersion() {
+	// step 3: add metadata lock and check each table's schema version
+	schemaNotMatch := false
+	for i := 0; i < len(stmt.dbName); i++ {
+		_, ok := is.TableByID(stmt.tbls[i].Meta().ID)
+		if !ok {
+			tblByName, err := is.TableByName(stmt.dbName[i], stmt.tbls[i].Meta().Name)
+			if err != nil {
+				return ErrSchemaChanged.GenWithStack("Schema change caused error: %s", err.Error())
+			}
+			delete(stmt.RelateVersion, stmt.tbls[i].Meta().ID)
+			stmt.tbls[i] = tblByName
+			stmt.RelateVersion[tblByName.Meta().ID] = tblByName.Meta().Revision
+		}
+		newTbl, err := tryLockMDLAndUpdateSchemaIfNecessary(sctx, stmt.dbName[i], stmt.tbls[i], is)
+		if err != nil {
+			schemaNotMatch = true
+			continue
+		}
+		if stmt.tbls[i].Meta().Revision != newTbl.Meta().Revision {
+			schemaNotMatch = true
+		}
+		stmt.tbls[i] = newTbl
+		stmt.RelateVersion[newTbl.Meta().ID] = newTbl.Meta().Revision
+	}
+
+	// step 4: check schema version
+	if schemaNotMatch || stmt.PreparedAst.SchemaVersion != is.SchemaMetaVersion() {
 		// In order to avoid some correctness issues, we have to clear the
 		// cached plan once the schema version is changed.
 		// Cached plan in prepared struct does NOT have a "cache key" with
@@ -99,7 +124,7 @@ func planCachePreprocess(ctx context.Context, sctx sessionctx.Context, isGeneral
 		stmtAst.SchemaVersion = is.SchemaMetaVersion()
 	}
 
-	// step 4: handle expiration
+	// step 5: handle expiration
 	// If the lastUpdateTime less than expiredTimeStamp4PC,
 	// it means other sessions have executed 'admin flush instance plan_cache'.
 	// So we need to clear the current session's plan cache.
@@ -110,6 +135,7 @@ func planCachePreprocess(ctx context.Context, sctx sessionctx.Context, isGeneral
 		stmtAst.CachedPlan = nil
 		vars.LastUpdateTime4PC = expiredTimeStamp4PC
 	}
+
 	return nil
 }
 
@@ -151,7 +177,7 @@ func GetPlanFromSessionPlanCache(ctx context.Context, sctx sessionctx.Context,
 			latestSchemaVersion = domain.GetDomain(sctx).InfoSchema().SchemaMetaVersion()
 		}
 		if cacheKey, err = NewPlanCacheKey(sctx.GetSessionVars(), stmt.StmtText,
-			stmt.StmtDB, stmtAst.SchemaVersion, latestSchemaVersion, bindSQL); err != nil {
+			stmt.StmtDB, stmtAst.SchemaVersion, latestSchemaVersion, bindSQL, stmt.RelateVersion); err != nil {
 			return nil, nil, err
 		}
 	}
@@ -293,7 +319,7 @@ func generateNewPlan(ctx context.Context, sctx sessionctx.Context, isGeneralPlan
 		if _, isolationReadContainTiFlash := sessVars.IsolationReadEngines[kv.TiFlash]; isolationReadContainTiFlash && !IsReadOnly(stmtAst.Stmt, sessVars) {
 			delete(sessVars.IsolationReadEngines, kv.TiFlash)
 			if cacheKey, err = NewPlanCacheKey(sessVars, stmt.StmtText, stmt.StmtDB,
-				stmtAst.SchemaVersion, latestSchemaVersion, bindSQL); err != nil {
+				stmtAst.SchemaVersion, latestSchemaVersion, bindSQL, stmt.RelateVersion); err != nil {
 				return nil, nil, err
 			}
 			sessVars.IsolationReadEngines[kv.TiFlash] = struct{}{}