Skip to content

Commit 364d200

Browse files
Marina Polyakovakovdb75
Marina Polyakova
authored and
kovdb75
committed
Avoid making commutatively-duplicate clauses in EquivalenceClasses.
See the commit a5fc46414deb7cbcd4cec1275efac69b9ac10500 (Avoid making commutatively-duplicate clauses in EquivalenceClasses.) in PostgreSQL 16.
1 parent e32efa8 commit 364d200

File tree

4 files changed

+1038
-0
lines changed

4 files changed

+1038
-0
lines changed

‎expected/pathman_join_clause_4.out

Lines changed: 161 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,161 @@
1+
/*
2+
* Since 8edd0e794 (>= 12) Append nodes with single subplan are eliminated,
3+
* causing different output; pathman_gaps_1.out is the updated version.
4+
*/
5+
\set VERBOSITY terse
6+
SET search_path = 'public';
7+
CREATE SCHEMA pathman;
8+
CREATE EXTENSION pg_pathman SCHEMA pathman;
9+
CREATE SCHEMA test;
10+
/*
11+
* Test push down a join clause into child nodes of append
12+
*/
13+
/* create test tables */
14+
CREATE TABLE test.fk (
15+
id1 INT NOT NULL,
16+
id2 INT NOT NULL,
17+
start_key INT,
18+
end_key INT,
19+
PRIMARY KEY (id1, id2));
20+
CREATE TABLE test.mytbl (
21+
id1 INT NOT NULL,
22+
id2 INT NOT NULL,
23+
key INT NOT NULL,
24+
CONSTRAINT fk_fk FOREIGN KEY (id1, id2) REFERENCES test.fk(id1, id2),
25+
PRIMARY KEY (id1, key));
26+
SELECT pathman.create_hash_partitions('test.mytbl', 'id1', 8);
27+
create_hash_partitions
28+
------------------------
29+
8
30+
(1 row)
31+
32+
/* ...fill out with test data */
33+
INSERT INTO test.fk VALUES (1, 1);
34+
INSERT INTO test.mytbl VALUES (1, 1, 5), (1, 1, 6);
35+
/* gather statistics on test tables to have deterministic plans */
36+
ANALYZE;
37+
/* run test queries */
38+
EXPLAIN (COSTS OFF) /* test plan */
39+
SELECT m.tableoid::regclass, id1, id2, key, start_key, end_key
40+
FROM test.mytbl m JOIN test.fk USING(id1, id2)
41+
WHERE NOT key <@ int4range(6, end_key);
42+
QUERY PLAN
43+
-------------------------------------------------------------------------------------------------------
44+
Nested Loop
45+
-> Seq Scan on fk
46+
-> Custom Scan (RuntimeAppend)
47+
Prune by: (m.id1 = fk.id1)
48+
-> Seq Scan on mytbl_0 m
49+
Filter: ((id1 = fk.id1) AND (fk.id2 = id2) AND (NOT (key <@ int4range(6, fk.end_key))))
50+
-> Seq Scan on mytbl_1 m
51+
Filter: ((id1 = fk.id1) AND (fk.id2 = id2) AND (NOT (key <@ int4range(6, fk.end_key))))
52+
-> Seq Scan on mytbl_2 m
53+
Filter: ((id1 = fk.id1) AND (fk.id2 = id2) AND (NOT (key <@ int4range(6, fk.end_key))))
54+
-> Seq Scan on mytbl_3 m
55+
Filter: ((id1 = fk.id1) AND (fk.id2 = id2) AND (NOT (key <@ int4range(6, fk.end_key))))
56+
-> Seq Scan on mytbl_4 m
57+
Filter: ((id1 = fk.id1) AND (fk.id2 = id2) AND (NOT (key <@ int4range(6, fk.end_key))))
58+
-> Seq Scan on mytbl_5 m
59+
Filter: ((id1 = fk.id1) AND (fk.id2 = id2) AND (NOT (key <@ int4range(6, fk.end_key))))
60+
-> Seq Scan on mytbl_6 m
61+
Filter: ((id1 = fk.id1) AND (fk.id2 = id2) AND (NOT (key <@ int4range(6, fk.end_key))))
62+
-> Seq Scan on mytbl_7 m
63+
Filter: ((id1 = fk.id1) AND (fk.id2 = id2) AND (NOT (key <@ int4range(6, fk.end_key))))
64+
(20 rows)
65+
66+
/* test joint data */
67+
SELECT m.tableoid::regclass, id1, id2, key, start_key, end_key
68+
FROM test.mytbl m JOIN test.fk USING(id1, id2)
69+
WHERE NOT key <@ int4range(6, end_key);
70+
tableoid | id1 | id2 | key | start_key | end_key
71+
--------------+-----+-----+-----+-----------+---------
72+
test.mytbl_6 | 1 | 1 | 5 | |
73+
(1 row)
74+
75+
/*
76+
* Test case by @dimarick
77+
*/
78+
CREATE TABLE test.parent (
79+
id SERIAL NOT NULL,
80+
owner_id INTEGER NOT NULL
81+
);
82+
CREATE TABLE test.child (
83+
parent_id INTEGER NOT NULL,
84+
owner_id INTEGER NOT NULL
85+
);
86+
CREATE TABLE test.child_nopart (
87+
parent_id INTEGER NOT NULL,
88+
owner_id INTEGER NOT NULL
89+
);
90+
INSERT INTO test.parent (owner_id) VALUES (1), (2), (3), (3);
91+
INSERT INTO test.child (parent_id, owner_id) VALUES (1, 1), (2, 2), (3, 3), (5, 3);
92+
INSERT INTO test.child_nopart (parent_id, owner_id) VALUES (1, 1), (2, 2), (3, 3), (5, 3);
93+
SELECT pathman.create_hash_partitions('test.child', 'owner_id', 2);
94+
create_hash_partitions
95+
------------------------
96+
2
97+
(1 row)
98+
99+
/* gather statistics on test tables to have deterministic plans */
100+
ANALYZE;
101+
/* Query #1 */
102+
EXPLAIN (COSTS OFF) SELECT * FROM test.parent
103+
LEFT JOIN test.child ON test.child.parent_id = test.parent.id AND
104+
test.child.owner_id = test.parent.owner_id
105+
WHERE test.parent.owner_id = 3 and test.parent.id IN (3, 4);
106+
QUERY PLAN
107+
-----------------------------------------------------------------------------------------------------
108+
Nested Loop Left Join
109+
-> Seq Scan on parent
110+
Filter: ((id = ANY ('{3,4}'::integer[])) AND (owner_id = 3))
111+
-> Custom Scan (RuntimeAppend)
112+
Prune by: ((child.owner_id = 3) AND (child.owner_id = parent.owner_id))
113+
-> Seq Scan on child_1 child
114+
Filter: ((owner_id = 3) AND (owner_id = parent.owner_id) AND (parent_id = parent.id))
115+
(7 rows)
116+
117+
SELECT * FROM test.parent
118+
LEFT JOIN test.child ON test.child.parent_id = test.parent.id AND
119+
test.child.owner_id = test.parent.owner_id
120+
WHERE test.parent.owner_id = 3 and test.parent.id IN (3, 4);
121+
id | owner_id | parent_id | owner_id
122+
----+----------+-----------+----------
123+
3 | 3 | 3 | 3
124+
4 | 3 | |
125+
(2 rows)
126+
127+
/* Query #2 */
128+
EXPLAIN (COSTS OFF) SELECT * FROM test.parent
129+
LEFT JOIN test.child ON test.child.parent_id = test.parent.id AND
130+
test.child.owner_id = 3
131+
WHERE test.parent.owner_id = 3 and test.parent.id IN (3, 4);
132+
QUERY PLAN
133+
----------------------------------------------------------------------
134+
Nested Loop Left Join
135+
Join Filter: (child.parent_id = parent.id)
136+
-> Seq Scan on parent
137+
Filter: ((id = ANY ('{3,4}'::integer[])) AND (owner_id = 3))
138+
-> Seq Scan on child_1 child
139+
Filter: (owner_id = 3)
140+
(6 rows)
141+
142+
SELECT * FROM test.parent
143+
LEFT JOIN test.child ON test.child.parent_id = test.parent.id AND
144+
test.child.owner_id = 3
145+
WHERE test.parent.owner_id = 3 and test.parent.id IN (3, 4);
146+
id | owner_id | parent_id | owner_id
147+
----+----------+-----------+----------
148+
3 | 3 | 3 | 3
149+
4 | 3 | |
150+
(2 rows)
151+
152+
DROP TABLE test.child CASCADE;
153+
NOTICE: drop cascades to 2 other objects
154+
DROP TABLE test.child_nopart CASCADE;
155+
DROP TABLE test.mytbl CASCADE;
156+
NOTICE: drop cascades to 8 other objects
157+
DROP TABLE test.fk CASCADE;
158+
DROP TABLE test.parent CASCADE;
159+
DROP SCHEMA test;
160+
DROP EXTENSION pg_pathman CASCADE;
161+
DROP SCHEMA pathman;

‎expected/pathman_lateral_4.out

Lines changed: 128 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,128 @@
1+
/*
2+
* Sometimes join selectivity improvements patches in pgpro force nested loop
3+
* members swap -- in pathman_lateral_1.out and pathman_lateral_3.out
4+
*
5+
* Since 55a1954da16 and 6ef77cf46e8 (>= 13) output of EXPLAIN was changed,
6+
* now it includes aliases for inherited tables.
7+
*/
8+
\set VERBOSITY terse
9+
SET search_path = 'public';
10+
CREATE EXTENSION pg_pathman;
11+
CREATE SCHEMA test_lateral;
12+
/* create table partitioned by HASH */
13+
create table test_lateral.data(id int8 not null);
14+
select create_hash_partitions('test_lateral.data', 'id', 10);
15+
create_hash_partitions
16+
------------------------
17+
10
18+
(1 row)
19+
20+
insert into test_lateral.data select generate_series(1, 10000);
21+
VACUUM ANALYZE;
22+
set enable_hashjoin = off;
23+
set enable_mergejoin = off;
24+
/* all credits go to Ivan Frolkov */
25+
explain (costs off)
26+
select * from
27+
test_lateral.data as t1,
28+
lateral(select * from test_lateral.data as t2 where t2.id > t1.id) t2,
29+
lateral(select * from test_lateral.data as t3 where t3.id = t2.id + t1.id) t3
30+
where t1.id between 1 and 100 and
31+
t2.id between 2 and 299 and
32+
t1.id > t2.id and
33+
exists(select * from test_lateral.data t
34+
where t1.id = t2.id and t.id = t3.id);
35+
QUERY PLAN
36+
--------------------------------------------------------------------------------------------
37+
Nested Loop
38+
-> Nested Loop
39+
Join Filter: ((t2.id + t1.id) = t.id)
40+
-> HashAggregate
41+
Group Key: t.id
42+
-> Append
43+
-> Seq Scan on data_0 t_1
44+
-> Seq Scan on data_1 t_2
45+
-> Seq Scan on data_2 t_3
46+
-> Seq Scan on data_3 t_4
47+
-> Seq Scan on data_4 t_5
48+
-> Seq Scan on data_5 t_6
49+
-> Seq Scan on data_6 t_7
50+
-> Seq Scan on data_7 t_8
51+
-> Seq Scan on data_8 t_9
52+
-> Seq Scan on data_9 t_10
53+
-> Materialize
54+
-> Nested Loop
55+
Join Filter: ((t2.id > t1.id) AND (t1.id > t2.id) AND (t1.id = t2.id))
56+
-> Append
57+
-> Seq Scan on data_0 t2_1
58+
Filter: ((id >= 2) AND (id <= 299))
59+
-> Seq Scan on data_1 t2_2
60+
Filter: ((id >= 2) AND (id <= 299))
61+
-> Seq Scan on data_2 t2_3
62+
Filter: ((id >= 2) AND (id <= 299))
63+
-> Seq Scan on data_3 t2_4
64+
Filter: ((id >= 2) AND (id <= 299))
65+
-> Seq Scan on data_4 t2_5
66+
Filter: ((id >= 2) AND (id <= 299))
67+
-> Seq Scan on data_5 t2_6
68+
Filter: ((id >= 2) AND (id <= 299))
69+
-> Seq Scan on data_6 t2_7
70+
Filter: ((id >= 2) AND (id <= 299))
71+
-> Seq Scan on data_7 t2_8
72+
Filter: ((id >= 2) AND (id <= 299))
73+
-> Seq Scan on data_8 t2_9
74+
Filter: ((id >= 2) AND (id <= 299))
75+
-> Seq Scan on data_9 t2_10
76+
Filter: ((id >= 2) AND (id <= 299))
77+
-> Materialize
78+
-> Append
79+
-> Seq Scan on data_0 t1_1
80+
Filter: ((id >= 1) AND (id <= 100))
81+
-> Seq Scan on data_1 t1_2
82+
Filter: ((id >= 1) AND (id <= 100))
83+
-> Seq Scan on data_2 t1_3
84+
Filter: ((id >= 1) AND (id <= 100))
85+
-> Seq Scan on data_3 t1_4
86+
Filter: ((id >= 1) AND (id <= 100))
87+
-> Seq Scan on data_4 t1_5
88+
Filter: ((id >= 1) AND (id <= 100))
89+
-> Seq Scan on data_5 t1_6
90+
Filter: ((id >= 1) AND (id <= 100))
91+
-> Seq Scan on data_6 t1_7
92+
Filter: ((id >= 1) AND (id <= 100))
93+
-> Seq Scan on data_7 t1_8
94+
Filter: ((id >= 1) AND (id <= 100))
95+
-> Seq Scan on data_8 t1_9
96+
Filter: ((id >= 1) AND (id <= 100))
97+
-> Seq Scan on data_9 t1_10
98+
Filter: ((id >= 1) AND (id <= 100))
99+
-> Custom Scan (RuntimeAppend)
100+
Prune by: (t3.id = t.id)
101+
-> Seq Scan on data_0 t3
102+
Filter: (t.id = id)
103+
-> Seq Scan on data_1 t3
104+
Filter: (t.id = id)
105+
-> Seq Scan on data_2 t3
106+
Filter: (t.id = id)
107+
-> Seq Scan on data_3 t3
108+
Filter: (t.id = id)
109+
-> Seq Scan on data_4 t3
110+
Filter: (t.id = id)
111+
-> Seq Scan on data_5 t3
112+
Filter: (t.id = id)
113+
-> Seq Scan on data_6 t3
114+
Filter: (t.id = id)
115+
-> Seq Scan on data_7 t3
116+
Filter: (t.id = id)
117+
-> Seq Scan on data_8 t3
118+
Filter: (t.id = id)
119+
-> Seq Scan on data_9 t3
120+
Filter: (t.id = id)
121+
(84 rows)
122+
123+
set enable_hashjoin = on;
124+
set enable_mergejoin = on;
125+
DROP TABLE test_lateral.data CASCADE;
126+
NOTICE: drop cascades to 10 other objects
127+
DROP SCHEMA test_lateral;
128+
DROP EXTENSION pg_pathman;

0 commit comments

Comments
 (0)