planner: outer merge join cannot keep the prop of its inner child (#33359)

planner/core/exhaust_physical_plans.go

@@ -286,14 +286,16 @@ func (p *LogicalJoin) getEnforcedMergeJoin(prop *property.PhysicalProperty, sche
 		return nil
 	}
 	for _, item := range prop.SortItems {
-		isExist := false
+		isExist, hasLeftColInProp, hasRightColInProp := false, false, false
 		for joinKeyPos := 0; joinKeyPos < len(leftJoinKeys); joinKeyPos++ {
 			var key *expression.Column
 			if item.Col.Equal(p.ctx, leftJoinKeys[joinKeyPos]) {
 				key = leftJoinKeys[joinKeyPos]
+				hasLeftColInProp = true
 			}
 			if item.Col.Equal(p.ctx, rightJoinKeys[joinKeyPos]) {
 				key = rightJoinKeys[joinKeyPos]
+				hasRightColInProp = true
 			}
 			if key == nil {
 				continue
@@ -313,6 +315,13 @@ func (p *LogicalJoin) getEnforcedMergeJoin(prop *property.PhysicalProperty, sche
 		if !isExist {
 			return nil
 		}
+		// If the output wants the order of the inner side. We should reject it since we might add null-extend rows of that side.
+		if p.JoinType == LeftOuterJoin && hasRightColInProp {
+			return nil
+		}
+		if p.JoinType == RightOuterJoin && hasLeftColInProp {
+			return nil
+		}
 	}
 	// Generate the enforced sort merge join
 	leftKeys := getNewJoinKeysByOffsets(leftJoinKeys, offsets)

planner/core/integration_test.go

@@ -4939,3 +4939,22 @@ func (s *testIntegrationSerialSuite) TestIssue30271(c *C) {
 	tk.MustExec("set names utf8mb4 collate utf8mb4_general_ci;")
 	tk.MustQuery("select * from t where (a>'a' and b='a') or (b = 'A' and a < 'd') order by a,c;").Check(testkit.Rows("b a 1", "b A 2", "c a 3"))
 }
+
+func (s *testIntegrationSerialSuite) TestIssue33042(c *C) {
+	tk := testkit.NewTestKit(c, s.store)
+
+	tk.MustExec("use test")
+	tk.MustExec("create table t1(id int primary key, col1 int)")
+	tk.MustExec("create table t2(id int primary key, col1 int)")
+	tk.MustQuery("explain format='brief' SELECT /*+ merge_join(t1, t2)*/ * FROM (t1 LEFT JOIN t2 ON t1.col1=t2.id) order by t2.id;").Check(
+		testkit.Rows(
+			"Sort 12500.00 root  test.t2.id",
+			"└─MergeJoin 12500.00 root  left outer join, left key:test.t1.col1, right key:test.t2.id",
+			"  ├─TableReader(Build) 10000.00 root  data:TableFullScan",
+			"  │ └─TableFullScan 10000.00 cop[tikv] table:t2 keep order:true, stats:pseudo",
+			"  └─Sort(Probe) 10000.00 root  test.t1.col1",
+			"    └─TableReader 10000.00 root  data:TableFullScan",
+			"      └─TableFullScan 10000.00 cop[tikv] table:t1 keep order:false, stats:pseudo",
+		),
+	)
+}

planner/core/testdata/plan_suite_out.json

@@ -566,7 +566,7 @@
       },
       {
         "SQL": "select /*+ TIDB_SMJ(t1,t2,t3)*/ * from t t1 left outer join t t2 on t1.a = t2.a left outer join t t3 on t2.a = t3.a",
-        "Best": "MergeLeftOuterJoin{MergeLeftOuterJoin{TableReader(Table(t))->TableReader(Table(t))}(test.t.a,test.t.a)->TableReader(Table(t))}(test.t.a,test.t.a)"
+        "Best": "MergeLeftOuterJoin{MergeLeftOuterJoin{TableReader(Table(t))->TableReader(Table(t))}(test.t.a,test.t.a)->Sort->TableReader(Table(t))}(test.t.a,test.t.a)"
       },
       {
         "SQL": "select /*+ TIDB_SMJ(t1,t2,t3)*/ * from t t1 left outer join t t2 on t1.a = t2.a left outer join t t3 on t1.a = t3.a",