ddl: cancel add index when can not find partition and fix GetPartitio…

…n function bug (#10475) (#10533)

ddl/db_test.go

@@ -4172,7 +4172,7 @@ func (s *testDBSuite) TestPartitionAddIndex(c *C) {
 
 	tk.MustExec("alter table partition_add_idx add index idx1 (hired)")
 	tk.MustExec("alter table partition_add_idx add index idx2 (id, hired)")
-	ctx := s.tk.Se.(sessionctx.Context)
+	ctx := tk.Se.(sessionctx.Context)
 	is := domain.GetDomain(ctx).InfoSchema()
 	t, err := is.TableByName(model.NewCIStr("test"), model.NewCIStr("partition_add_idx"))
 	c.Assert(err, IsNil)
@@ -4190,6 +4190,22 @@ func (s *testDBSuite) TestPartitionAddIndex(c *C) {
 
 	tk.MustExec("admin check table partition_add_idx")
 	tk.MustExec("drop table partition_add_idx")
+
+	// Test hash partition for pr 10475.
+	tk.MustExec("drop table if exists t1")
+	defer tk.MustExec("drop table if exists t1")
+	tk.MustExec("set @@session.tidb_enable_table_partition = '1';")
+	tk.MustExec("create table t1 (a int,b int,  primary key(a)) partition by hash(a) partitions 5;")
+	tk.MustExec("insert into t1 values (0,0),(1,1),(2,2),(3,3);")
+	tk.MustExec("alter table t1 add index idx(a)")
+	tk.MustExec("admin check table t1;")
+
+	// Test range partition for pr 10475.
+	tk.MustExec("drop table t1")
+	tk.MustExec("create table t1 (a int,b int,  primary key(a)) partition by range (a) (partition p0 values less than (10), partition p1 values less than (20));")
+	tk.MustExec("insert into t1 values (0,0);")
+	tk.MustExec("alter table t1 add index idx(a)")
+	tk.MustExec("admin check table t1;")
 }
 
 func (s *testDBSuite) TestAlterTableCharset(c *C) {

ddl/index.go

@@ -328,7 +328,16 @@ func (w *worker) onCreateIndex(d *ddlCtx, t *meta.Meta, job *model.Job) (ver int
 			return ver, errors.Trace(err)
 		}
 
-		err = w.runReorgJob(t, reorgInfo, d.lease, func() error {
+		err = w.runReorgJob(t, reorgInfo, d.lease, func() (addIndexErr error) {
+			defer func() {
+				r := recover()
+				if r != nil {
+					buf := util.GetStack()
+					logutil.Logger(ddlLogCtx).Error("[ddl] add table index panic", zap.Any("panic", r), zap.String("stack", string(buf)))
+					metrics.PanicCounter.WithLabelValues(metrics.LabelDDL).Inc()
+					addIndexErr = errCancelledDDLJob.GenWithStack("add table `%v` index `%v` panic", tblInfo.Name, indexInfo.Name)
+				}
+			}()
 			return w.addTableIndex(tbl, indexInfo, reorgInfo)
 		})
 		if err != nil {
@@ -1177,7 +1186,7 @@ func (w *worker) addTableIndex(t table.Table, idx *model.IndexInfo, reorgInfo *r
 		for !finish {
 			p := tbl.GetPartition(reorgInfo.PhysicalTableID)
 			if p == nil {
-				return errors.Errorf("Can not find partition id %d for table %d", reorgInfo.PhysicalTableID, t.Meta().ID)
+				return errCancelledDDLJob.GenWithStack("Can not find partition id %d for table %d", reorgInfo.PhysicalTableID, t.Meta().ID)
 			}
 			err = w.addPhysicalTableIndex(p, idx, reorgInfo)
 			if err != nil {

meta/meta.go

@@ -762,9 +762,9 @@ func (m *Meta) GetDDLReorgHandle(job *model.Job) (startHandle, endHandle, physic
 		err = errors.Trace(err)
 		return
 	}
-	// endHandle or physicalTableID may be 0, because older version TiDB (without table partition) doesn't store them.
+	// physicalTableID may be 0, because older version TiDB (without table partition) doesn't store them.
 	// update them to table's in this case.
-	if endHandle == 0 || physicalTableID == 0 {
+	if physicalTableID == 0 {
 		if job.ReorgMeta != nil {
 			endHandle = job.ReorgMeta.EndHandle
 		} else {

table/tables/partition.go

@@ -220,7 +220,13 @@ func (t *partitionedTable) locatePartition(ctx sessionctx.Context, pi *model.Par
 
 // GetPartition returns a Table, which is actually a partition.
 func (t *partitionedTable) GetPartition(pid int64) table.PhysicalTable {
-	return t.partitions[pid]
+	// Attention, can't simply use `return t.partitions[pid]` here.
+	// Because A nil of type *partition is a kind of `table.PhysicalTable`
+	p, ok := t.partitions[pid]
+	if !ok {
+		return nil
+	}
+	return p
 }
 
 // GetPartitionByRow returns a Table, which is actually a Partition.