Support "Right Semi Join" plan shapes

Hash joins can support semijoin with the LHS input on the right, using
the existing logic for inner join, combined with the assurance that only
the first match for each inner tuple is considered, which can be
achieved by leveraging the HEAP_TUPLE_HAS_MATCH flag.  This can be very
useful in some cases since we may now have the option to hash the
smaller table instead of the larger.

Merge join could likely support "Right Semi Join" too.  However, the
benefit of swapping inputs tends to be small here, so we do not address
that in this patch.

Note that this patch also modifies a test query in join.sql to ensure it
continues testing as intended.  With this patch the original query would
result in a right-semi-join rather than semi-join, compromising its
original purpose of testing the fix for neqjoinsel's behavior for
semi-joins.

Author: Richard Guo
Reviewed-by: wenhui qiu, Alena Rybakina, Japin Li
Discussion: https://postgr.es/m/CAMbWs4_X1mN=ic+SxcyymUqFx9bB8pqSLTGJ-F=MHy4PW3eRXw@mail.gmail.com
(cherry picked from commit aa86129)

Author: Richard Guo

src/backend/commands/explain.c

@@ -2183,6 +2183,9 @@ ExplainNode(PlanState *planstate, List *ancestors,
 					case JOIN_LASJ_NOTIN:
 						jointype = "Left Anti Semi (Not-In)";
 						break;
+					case JOIN_RIGHT_SEMI:
+						jointype = "Right Semi";
+						break;
 					case JOIN_RIGHT_ANTI:
 						jointype = "Right Anti";
 						break;

src/backend/executor/nodeHashjoin.c

@@ -688,6 +688,14 @@ ExecHashJoinImpl(PlanState *pstate, bool parallel)
 					}
 				}
 
+				/*
+				 * In a right-semijoin, we only need the first match for each
+				 * inner tuple.
+				 */
+				if (node->js.jointype == JOIN_RIGHT_SEMI &&
+					HeapTupleHeaderHasMatch(HJTUPLE_MINTUPLE(node->hj_CurTuple)))
+					continue;
+
 				/*
 				 * We've got a match, but still need to test non-hashed quals.
 				 * ExecScanHashBucket already set up all the state needed to
@@ -704,10 +712,10 @@ ExecHashJoinImpl(PlanState *pstate, bool parallel)
 				{
 					node->hj_MatchedOuter = true;
 
-
 					/*
-					 * This is really only needed if HJ_FILL_INNER(node), but
-					 * we'll avoid the branch and just set it always.
+					 * This is really only needed if HJ_FILL_INNER(node) or if
+					 * we are in a right-semijoin, but we'll avoid the branch
+					 * and just set it always.
 					 */
 					if (!HeapTupleHeaderHasMatch(HJTUPLE_MINTUPLE(node->hj_CurTuple)))
 						HeapTupleHeaderSetMatch(HJTUPLE_MINTUPLE(node->hj_CurTuple));
@@ -1024,6 +1032,7 @@ ExecInitHashJoin(HashJoin *node, EState *estate, int eflags)
 	{
 		case JOIN_INNER:
 		case JOIN_SEMI:
+		case JOIN_RIGHT_SEMI:
 			break;
 		case JOIN_LEFT:
 		case JOIN_ANTI:

src/backend/optimizer/path/joinpath.c

@@ -338,8 +338,8 @@ add_paths_to_join_relation(PlannerInfo *root,
 	 * sorted. This includes both nestloops and mergejoins where the outer
 	 * path is already ordered.  Again, skip this if we can't mergejoin.
 	 * (That's okay because we know that nestloop can't handle
-	 * right/right-anti/full joins at all, so it wouldn't work in the
-	 * prohibited cases either.)
+	 * right/right-anti/right-semi/full joins at all, so it wouldn't work in
+	 * the prohibited cases either.)
 	 */
 	if (mergejoin_allowed)
 		match_unsorted_outer(root, joinrel, outerrel, innerrel,
@@ -1897,6 +1897,13 @@ match_unsorted_outer(PlannerInfo *root,
 	if (jointype == JOIN_DEDUP_SEMI || jointype == JOIN_DEDUP_SEMI_REVERSE)
 		jointype = JOIN_INNER;
 
+	/*
+	 * For now we do not support RIGHT_SEMI join in mergejoin or nestloop
+	 * join.
+	 */
+	if (jointype == JOIN_RIGHT_SEMI)
+		return;
+
 	/*
 	 * Nestloop only supports inner, left, semi, and anti joins.  Also, if we
 	 * are doing a right, right-anti or full mergejoin, we must use *all* the
@@ -2501,12 +2508,13 @@ hash_inner_and_outer(PlannerInfo *root,
 			 * total inner path will also be parallel-safe, but if not, we'll
 			 * have to search for the cheapest safe, unparameterized inner
 			 * path.  If doing JOIN_UNIQUE_INNER, we can't use any alternative
-			 * inner path.  If full, right, or right-anti join, we can't use
-			 * parallelism (building the hash table in each backend) because
-			 * no one process has all the match bits.
+			 * inner path.  If full, right, right-semi or right-anti join, we
+			 * can't use parallelism (building the hash table in each backend)
+			 * because no one process has all the match bits.
 			 */
 			if (save_jointype == JOIN_FULL ||
 				save_jointype == JOIN_RIGHT ||
+				save_jointype == JOIN_RIGHT_SEMI ||
 				save_jointype == JOIN_RIGHT_ANTI)
 				cheapest_safe_inner = NULL;
 			else if (cheapest_total_inner->parallel_safe)
@@ -2533,13 +2541,13 @@ hash_inner_and_outer(PlannerInfo *root,
  *	  Returns a list of RestrictInfo nodes for those clauses.
  *
  * *mergejoin_allowed is normally set to true, but it is set to false if
- * this is a right/right-anti/full join and there are nonmergejoinable join
- * clauses.  The executor's mergejoin machinery cannot handle such cases, so
- * we have to avoid generating a mergejoin plan.  (Note that this flag does
- * NOT consider whether there are actually any mergejoinable clauses.  This is
- * correct because in some cases we need to build a clauseless mergejoin.
- * Simply returning NIL is therefore not enough to distinguish safe from
- * unsafe cases.)
+ * this is a right-semi join, or this is a right/right-anti/full join and
+ * there are nonmergejoinable join clauses.  The executor's mergejoin
+ * machinery cannot handle such cases, so we have to avoid generating a
+ * mergejoin plan.  (Note that this flag does NOT consider whether there are
+ * actually any mergejoinable clauses.  This is correct because in some
+ * cases we need to build a clauseless mergejoin.  Simply returning NIL is
+ * therefore not enough to distinguish safe from unsafe cases.)
  *
  * We also mark each selected RestrictInfo to show which side is currently
  * being considered as outer.  These are transient markings that are only
@@ -2563,6 +2571,16 @@ select_mergejoin_clauses(PlannerInfo *root,
 	bool		have_nonmergeable_joinclause = false;
 	ListCell   *l;
 
+	/*
+	 * For now we do not support RIGHT_SEMI join in mergejoin: the benefit of
+	 * swapping inputs tends to be small here.
+	 */
+	if (jointype == JOIN_RIGHT_SEMI)
+	{
+		*mergejoin_allowed = false;
+		return NIL;
+	}
+
 	foreach(l, restrictlist)
 	{
 		RestrictInfo *restrictinfo = (RestrictInfo *) lfirst(l);

src/backend/optimizer/path/joinrels.c

@@ -1128,6 +1128,18 @@ populate_joinrel_with_paths(PlannerInfo *root, RelOptInfo *rel1,
 									 JOIN_SEMI, sjinfo,
 									 restrictlist);
 
+				/*
+				 * Also consider "Right Semi Join" plan shapes, with the inputs
+				 * swapped so that the LHS (rel1) becomes the hash/build side.
+				 * This lets us hash the smaller table.  Like JOIN_SEMI above,
+				 * this does not add a UniquePath, so (unlike the GPDB
+				 * JOIN_DEDUP_SEMI paths below) it is safe to consider
+				 * unconditionally.
+				 */
+				add_paths_to_joinrel(root, joinrel, rel2, rel1,
+									 JOIN_RIGHT_SEMI, sjinfo,
+									 restrictlist);
+
 				if (root->upd_del_replicated_table > 0 &&
 					(bms_is_member(root->upd_del_replicated_table, rel1->relids) ||
 					 bms_is_member(root->upd_del_replicated_table, rel2->relids)))

src/backend/optimizer/path/pathkeys.c

@@ -1425,6 +1425,9 @@ build_join_pathkeys(PlannerInfo *root,
 					JoinType jointype,
 					List *outer_pathkeys)
 {
+	/* RIGHT_SEMI should not come here */
+	Assert(jointype != JOIN_RIGHT_SEMI);
+
 	if (jointype == JOIN_FULL ||
 		jointype == JOIN_RIGHT ||
 		jointype == JOIN_RIGHT_ANTI)

src/backend/optimizer/prep/prepjointree.c

@@ -430,8 +430,8 @@ pull_up_sublinks_jointree_recurse(PlannerInfo *root, Node *jtnode,
 		 * point of the available_rels machinations is to ensure that we only
 		 * pull up quals for which that's okay.
 		 *
-		 * We don't expect to see any pre-existing JOIN_SEMI, JOIN_ANTI, or
-		 * JOIN_RIGHT_ANTI jointypes here.
+		 * We don't expect to see any pre-existing JOIN_SEMI, JOIN_ANTI,
+		 * JOIN_RIGHT_SEMI, or JOIN_RIGHT_ANTI jointypes here.
 		 */
 		switch (j->jointype)
 		{
@@ -3277,7 +3277,7 @@ reduce_outer_joins_pass2(Node *jtnode,
 				 * These could only have been introduced by pull_up_sublinks,
 				 * so there's no way that upper quals could refer to their
 				 * righthand sides, and no point in checking.  We don't expect
-				 * to see JOIN_RIGHT_ANTI yet.
+				 * to see JOIN_RIGHT_SEMI or JOIN_RIGHT_ANTI yet.
 				 */
 				break;
 			default:

src/include/nodes/nodes.h

@@ -1026,6 +1026,7 @@ typedef enum JoinType
 	JOIN_LASJ_NOTIN,			/* Left Anti Semi Join with Not-In semantics:
 									If any NULL values are produced by inner side,
 									return no join results. Otherwise, same as LASJ */
+	JOIN_RIGHT_SEMI,			/* 1 copy of each RHS row that has match(es) */
 	JOIN_RIGHT_ANTI,			/* 1 copy of each RHS row that has no match */
 
 	/*
@@ -1054,10 +1055,10 @@ typedef enum JoinType
 
 /*
  * OUTER joins are those for which pushed-down quals must behave differently
- * from the join's own quals.  This is in fact everything except INNER and
- * SEMI joins.  However, this macro must also exclude the JOIN_UNIQUE symbols
- * since those are temporary proxies for what will eventually be an INNER
- * join.
+ * from the join's own quals.  This is in fact everything except INNER, SEMI
+ * and RIGHT_SEMI joins.  However, this macro must also exclude the
+ * JOIN_UNIQUE symbols since those are temporary proxies for what will
+ * eventually be an INNER join.
  *
  * Note: semijoins are a hybrid case, but we choose to treat them as not
  * being outer joins.  This is okay principally because the SQL syntax makes

src/include/nodes/pathnodes.h

@@ -3275,9 +3275,9 @@ typedef struct PlaceHolderVar
  * min_lefthand and min_righthand for higher joins.)
  *
  * jointype is never JOIN_RIGHT; a RIGHT JOIN is handled by switching
- * the inputs to make it a LEFT JOIN.  It's never JOIN_RIGHT_ANTI either.
- * So the allowed values of jointype in a join_info_list member are only
- * LEFT, FULL, SEMI, or ANTI.
+ * the inputs to make it a LEFT JOIN.  It's never JOIN_RIGHT_SEMI or
+ * JOIN_RIGHT_ANTI either.  So the allowed values of jointype in a
+ * join_info_list member are only LEFT, FULL, SEMI, or ANTI.
  *
  * ojrelid is the RT index of the join RTE representing this outer join,
  * if there is one.  It is zero when jointype is INNER or SEMI, and can be

src/test/regress/sql/join.sql

@@ -214,10 +214,10 @@ SELECT *
 -- semijoin selectivity for <>
 --
 explain (costs off)
-select * from int4_tbl i4, tenk1 a
-where exists(select * from tenk1 b
-             where a.twothousand = b.twothousand and a.fivethous <> b.fivethous)
-      and i4.f1 = a.tenthous;
+select * from tenk1 a, tenk1 b
+where exists(select * from tenk1 c
+             where b.twothousand = c.twothousand and b.fivethous <> c.fivethous)
+      and a.tenthous = b.tenthous and a.tenthous < 5000;
 
 
 --