ddl: fix bug of run ddl job error but actually take effect (#17341)

ddl/ddl_worker.go

@@ -474,6 +474,15 @@ func (w *worker) handleDDLJobQueue(d *ddlCtx) error {
 				err = w.finishDDLJob(t, job)
 				return errors.Trace(err)
 			}
+			if runJobErr != nil && !job.IsRollingback() && !job.IsRollbackDone() {
+				// If the running job meets an error
+				// and the job state is rolling back, it means that we have already handled this error.
+				// Some DDL jobs (such as adding indexes) may need to update the table info and the schema version,
+				// then shouldn't discard the KV modification.
+				// And the job state is rollback done, it means the job was already finished, also shouldn't discard too.
+				// Otherwise, we should discard the KV modification when running job.
+				txn.Discard()
+			}
 			err = w.updateDDLJob(t, job, runJobErr != nil)
 			if err = w.handleUpdateJobError(t, job, err); err != nil {
 				return errors.Trace(err)

ddl/serial_test.go

@@ -744,6 +744,27 @@ func (s *testSerialSuite) TestCancelJobByErrorCountLimit(c *C) {
 	c.Assert(err.Error(), Equals, "[ddl:8214]Cancelled DDL job")
 }
 
+func (s *testSerialSuite) TestTruncateTableUpdateSchemaVersionErr(c *C) {
+	tk := testkit.NewTestKit(c, s.store)
+	c.Assert(failpoint.Enable("github.com/pingcap/tidb/ddl/mockTruncateTableUpdateVersionError", `return(true)`), IsNil)
+	tk.MustExec("use test")
+	tk.MustExec("drop table if exists t")
+
+	limit := variable.GetDDLErrorCountLimit()
+	tk.MustExec("set @@global.tidb_ddl_error_count_limit = 5")
+	err := ddlutil.LoadDDLVars(tk.Se)
+	c.Assert(err, IsNil)
+	defer tk.MustExec(fmt.Sprintf("set @@global.tidb_ddl_error_count_limit = %d", limit))
+
+	tk.MustExec("create table t (a int)")
+	_, err = tk.Exec("truncate table t")
+	c.Assert(err, NotNil)
+	c.Assert(err.Error(), Equals, "[ddl:8214]Cancelled DDL job")
+	// Disable fail point.
+	c.Assert(failpoint.Disable("github.com/pingcap/tidb/ddl/mockTruncateTableUpdateVersionError"), IsNil)
+	tk.MustExec("truncate table t")
+}
+
 func (s *testSerialSuite) TestCanceledJobTakeTime(c *C) {
 	tk := testkit.NewTestKit(c, s.store)
 	tk.MustExec("use test")

ddl/table.go

@@ -478,6 +478,12 @@ func onTruncateTable(d *ddlCtx, t *meta.Meta, job *model.Job) (ver int64, _ erro
 		return ver, errors.Trace(err)
 	}
 
+	failpoint.Inject("mockTruncateTableUpdateVersionError", func(val failpoint.Value) {
+		if val.(bool) {
+			failpoint.Return(ver, errors.New("mock update version error"))
+		}
+	})
+
 	ver, err = updateSchemaVersion(t, job)
 	if err != nil {
 		return ver, errors.Trace(err)