Skip to content

Commit c959f84

Browse files
tglsfdctglsfdc
committed
Fix Memoize to work with partitionwise joining.
A couple of places weren't up to speed for this. By sheer good luck, we didn't fail but just selected a non-memoized join plan, at least in the test case we have. Nonetheless, it's a bug, and I'm not quite sure that it couldn't have worse consequences in other examples. So back-patch to v14 where Memoize came in. Richard Guo Discussion: https://postgr.es/m/CAMbWs48GkNom272sfp0-WeD6_0HSR19BJ4H1c9ZKSfbVnJsvRg@mail.gmail.com
1 parent 74a600a commit c959f84

File tree

5 files changed

+74
-3
lines changed

5 files changed

+74
-3
lines changed

‎src/backend/optimizer/path/joinpath.c

Lines changed: 13 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -505,6 +505,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
505505
Path *outer_path, JoinType jointype,
506506
JoinPathExtraData *extra)
507507
{
508+
RelOptInfo *top_outerrel;
508509
List *param_exprs;
509510
List *hash_operators;
510511
ListCell *lc;
@@ -594,10 +595,21 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
594595
return NULL;
595596
}
596597

598+
/*
599+
* When considering a partitionwise join, we have clauses that reference
600+
* the outerrel's top parent not outerrel itself.
601+
*/
602+
if (outerrel->reloptkind == RELOPT_OTHER_MEMBER_REL)
603+
top_outerrel = find_base_rel(root, bms_singleton_member(outerrel->top_parent_relids));
604+
else if (outerrel->reloptkind == RELOPT_OTHER_JOINREL)
605+
top_outerrel = find_join_rel(root, outerrel->top_parent_relids);
606+
else
607+
top_outerrel = outerrel;
608+
597609
/* Check if we have hash ops for each parameter to the path */
598610
if (paraminfo_get_equal_hashops(root,
599611
inner_path->param_info,
600-
outerrel,
612+
top_outerrel,
601613
innerrel,
602614
&param_exprs,
603615
&hash_operators,

‎src/backend/optimizer/util/pathnode.c

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4193,6 +4193,7 @@ do { \
41934193

41944194
FLAT_COPY_PATH(mpath, path, MemoizePath);
41954195
REPARAMETERIZE_CHILD_PATH(mpath->subpath);
4196+
ADJUST_CHILD_ATTRS(mpath->param_exprs);
41964197
new_path = (Path *) mpath;
41974198
}
41984199
break;

‎src/include/nodes/pathnodes.h

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1513,8 +1513,8 @@ typedef struct MemoizePath
15131513
{
15141514
Path path;
15151515
Path *subpath; /* outerpath to cache tuples from */
1516-
List *hash_operators; /* hash operators for each key */
1517-
List *param_exprs; /* cache keys */
1516+
List *hash_operators; /* OIDs of hash equality ops for cache keys */
1517+
List *param_exprs; /* expressions that are cache keys */
15181518
bool singlerow; /* true if the cache entry is to be marked as
15191519
* complete after caching the first record. */
15201520
bool binary_mode; /* true when cache key should be compared bit

‎src/test/regress/expected/memoize.out

Lines changed: 39 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -197,6 +197,45 @@ SELECT * FROM strtest s1 INNER JOIN strtest s2 ON s1.t >= s2.t;', false);
197197
(8 rows)
198198

199199
DROP TABLE strtest;
200+
-- Ensure memoize works with partitionwise join
201+
SET enable_partitionwise_join TO on;
202+
CREATE TABLE prt (a int) PARTITION BY RANGE(a);
203+
CREATE TABLE prt_p1 PARTITION OF prt FOR VALUES FROM (0) TO (10);
204+
CREATE TABLE prt_p2 PARTITION OF prt FOR VALUES FROM (10) TO (20);
205+
INSERT INTO prt VALUES (0), (0), (0), (0);
206+
INSERT INTO prt VALUES (10), (10), (10), (10);
207+
CREATE INDEX iprt_p1_a ON prt_p1 (a);
208+
CREATE INDEX iprt_p2_a ON prt_p2 (a);
209+
ANALYZE prt;
210+
SELECT explain_memoize('
211+
SELECT * FROM prt t1 INNER JOIN prt t2 ON t1.a = t2.a;', false);
212+
explain_memoize
213+
------------------------------------------------------------------------------------------
214+
Append (actual rows=32 loops=N)
215+
-> Nested Loop (actual rows=16 loops=N)
216+
-> Index Only Scan using iprt_p1_a on prt_p1 t1_1 (actual rows=4 loops=N)
217+
Heap Fetches: N
218+
-> Memoize (actual rows=4 loops=N)
219+
Cache Key: t1_1.a
220+
Cache Mode: logical
221+
Hits: 3 Misses: 1 Evictions: Zero Overflows: 0 Memory Usage: NkB
222+
-> Index Only Scan using iprt_p1_a on prt_p1 t2_1 (actual rows=4 loops=N)
223+
Index Cond: (a = t1_1.a)
224+
Heap Fetches: N
225+
-> Nested Loop (actual rows=16 loops=N)
226+
-> Index Only Scan using iprt_p2_a on prt_p2 t1_2 (actual rows=4 loops=N)
227+
Heap Fetches: N
228+
-> Memoize (actual rows=4 loops=N)
229+
Cache Key: t1_2.a
230+
Cache Mode: logical
231+
Hits: 3 Misses: 1 Evictions: Zero Overflows: 0 Memory Usage: NkB
232+
-> Index Only Scan using iprt_p2_a on prt_p2 t2_2 (actual rows=4 loops=N)
233+
Index Cond: (a = t1_2.a)
234+
Heap Fetches: N
235+
(21 rows)
236+
237+
DROP TABLE prt;
238+
RESET enable_partitionwise_join;
200239
-- Exercise Memoize code that flushes the cache when a parameter changes which
201240
-- is not part of the cache key.
202241
-- Ensure we get a Memoize plan

‎src/test/regress/sql/memoize.sql

Lines changed: 19 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -104,6 +104,25 @@ SELECT * FROM strtest s1 INNER JOIN strtest s2 ON s1.t >= s2.t;', false);
104104

105105
DROP TABLE strtest;
106106

107+
-- Ensure memoize works with partitionwise join
108+
SET enable_partitionwise_join TO on;
109+
110+
CREATE TABLE prt (a int) PARTITION BY RANGE(a);
111+
CREATE TABLE prt_p1 PARTITION OF prt FOR VALUES FROM (0) TO (10);
112+
CREATE TABLE prt_p2 PARTITION OF prt FOR VALUES FROM (10) TO (20);
113+
INSERT INTO prt VALUES (0), (0), (0), (0);
114+
INSERT INTO prt VALUES (10), (10), (10), (10);
115+
CREATE INDEX iprt_p1_a ON prt_p1 (a);
116+
CREATE INDEX iprt_p2_a ON prt_p2 (a);
117+
ANALYZE prt;
118+
119+
SELECT explain_memoize('
120+
SELECT * FROM prt t1 INNER JOIN prt t2 ON t1.a = t2.a;', false);
121+
122+
DROP TABLE prt;
123+
124+
RESET enable_partitionwise_join;
125+
107126
-- Exercise Memoize code that flushes the cache when a parameter changes which
108127
-- is not part of the cache key.
109128

0 commit comments

Comments
 (0)