1.强制对临时对象使用COMMIT而不是2PC
postgres=# SET enforce_two_phase_commit TO off; SET
2.创建测试表并插入测试数据
postgres=# CREATE TEMP TABLE tab_test_plancache(q1 int8, q2 int8);
CREATE TABLE
postgres=# INSERT INTO tab_test_plancache VALUES(' 123 ',' 456');
INSERT 0 1
postgres=# INSERT INTO tab_test_plancache VALUES('123 ','4567890123456789');
INSERT 0 1
postgres=# INSERT INTO tab_test_plancache VALUES('4567890123456789','123');
INSERT 0 1
postgres=# INSERT INTO tab_test_plancache VALUES(+4567890123456789,'4567890123456789');
INSERT 0 1
postgres=# INSERT INTO tab_test_plancache VALUES('+4567890123456789','-4567890123456789');
INSERT 0 1
3.创建并使用缓存的计划
postgres=# PREPARE prepstmt AS SELECT * FROM tab_test_plancache ORDER BY q1, q2; PREPARE postgres=# EXECUTE prepstmt; q1 | q2 ------------------+------------------- 123 | 456 123 | 4567890123456789 4567890123456789 | -4567890123456789 4567890123456789 | 123 4567890123456789 | 4567890123456789 (5 rows)
并且包含一个带有绑定变量的
postgres=# PREPARE prepstmt2(bigint) AS SELECT * FROM tab_test_plancache WHERE q1 = $1 ORDER BY q1, q2; PREPARE postgres=# EXECUTE prepstmt2(123); q1 | q2 -----+------------------ 123 | 456 123 | 4567890123456789 (2 rows)
4.删除临时表,查看现象
postgres=# DROP TABLE tab_test_plancache; DROP TABLE postgres=# EXECUTE prepstmt; ERROR: relation "tab_test_plancache" does not exist on dn_6001_6002 postgres=# EXECUTE prepstmt2(123); ERROR: relation "tab_test_plancache" does not exist on dn_6001_6002
重建临时表
postgres=# select * from tab_test_plancache; q1 | q2 ------------------+------------------- 123 | 456 123 | 4567890123456789 4567890123456789 | -4567890123456789 4567890123456789 | 123 4567890123456789 | 4567890123456789 (5 rows) postgres=# EXECUTE prepstmt; q1 | q2 ------------------+------------------- 123 | 456 123 | 4567890123456789 4567890123456789 | -4567890123456789 4567890123456789 | 123 4567890123456789 | 4567890123456789 (5 rows) postgres=# EXECUTE prepstmt2(123); q1 | q2 -----+------------------ 123 | 456 123 | 4567890123456789 (2 rows)
这表明原始计划是纯文本的,不依赖于OID
5.prepared statements应该防止在输出的tupdesc中更改, 因为clients可能不希望这种情况瞬间改变
postgres=# ALTER TABLE tab_test_plancache ADD COLUMN q3 bigint; ALTER TABLE postgres=# EXECUTE prepstmt; ERROR: cached plan must not change result type postgres=# EXECUTE prepstmt2(123); ERROR: cached plan must not change result type
例子里增加了一列,但是报出了缓存的计划不能更改结果类型,可以通过还原原来表的结构解决
postgres=# ALTER TABLE tab_test_plancache ADD COLUMN q3 bigint; ALTER TABLE postgres=# select * from tab_test_plancache; q1 | q2 | q3 ------------------+-------------------+---- 123 | 456 | 123 | 4567890123456789 | 4567890123456789 | -4567890123456789 | 4567890123456789 | 123 | 4567890123456789 | 4567890123456789 | (5 rows) postgres=# ALTER TABLE tab_test_plancache DROP COLUMN q3; ALTER TABLE postgres=# EXECUTE prepstmt; q1 | q2 ------------------+------------------- 123 | 456 123 | 4567890123456789 4567890123456789 | -4567890123456789 4567890123456789 | 123 4567890123456789 | 4567890123456789 (5 rows) postgres=# EXECUTE prepstmt2(123); q1 | q2 -----+------------------ 123 | 456 123 | 4567890123456789 (2 rows)
6.检查使用视图的有效性
如果尝试使用一个视图的话,这个视图不会直接用于生成的计划中,但也是有效的
postgres=# CREATE TEMP VIEW pcacheview AS postgres-# SELECT * FROM tab_test_plancache; CREATE VIEW postgres=# PREPARE vprep AS SELECT * FROM pcacheview ORDER BY q1, q2; PREPARE postgres=# EXECUTE vprep; q1 | q2 ------------------+------------------- 123 | 456 123 | 4567890123456789 4567890123456789 | -4567890123456789 4567890123456789 | 123 4567890123456789 | 4567890123456789 (5 rows) postgres=# CREATE OR REPLACE TEMP VIEW pcacheview AS SELECT q1, q2+1 AS q2 FROM tab_test_plancache ORDER BY q1, q2; CREATE VIEW postgres=# EXECUTE vprep; q1 | q2 ------------------+------------------- 123 | 457 123 | 4567890123456790 4567890123456789 | -4567890123456788 4567890123456789 | 124 4567890123456789 | 4567890123456790 (5 rows)
7.检查基本 SPI plan 是否有效
postgres=# create function cache_test(int) returns int as $$ postgres$# declare total int; postgres$# begin postgres$# create table t1_plancache(f1 int); postgres$# insert into t1_plancache values($1); postgres$# insert into t1_plancache values(11); postgres$# insert into t1_plancache values(12); postgres$# insert into t1_plancache values(13); postgres$# select sum(f1) into total from t1_plancache; postgres$# drop table t1_plancache; postgres$# return total; postgres$# end postgres$# $$ language plpgsql; CREATE FUNCTION postgres=# select cache_test(1); cache_test ------------ 37 (1 row) postgres=# select cache_test(2); cache_test ------------ 38 (1 row) postgres=# select cache_test(3); cache_test ------------ 39 (1 row)
8.检查plpgsql“简单表达式”的有效性
postgres=# create temp view v1 as postgres-# select 2+2 as f1; CREATE VIEW postgres=# create function cache_test_2() returns int as $$ postgres$# begin postgres$# return f1 from v1; postgres$# end$$ language plpgsql; CREATE FUNCTION postgres=# select cache_test_2(); cache_test_2 -------------- 4 (1 row) postgres=# create or replace temp view v1 as postgres-# select 2+2+4 as f1; CREATE VIEW postgres=# select cache_test_2(); cache_test_2 -------------- 8 (1 row)
9.检查缓存执行计划使用与search_path影响
可以看到,两个schema下都有同一张表,修改了search_path后,缓存执行计划执行的是search_path下的表,所以缓存执行计划会受search_path影响。
postgres=# create schema s1 postgres-# create table abc (f1 int); CREATE SCHEMA postgres=# create schema s2 postgres-# create table abc (f1 int); CREATE SCHEMA postgres=# insert into s1.abc values(123); INSERT 0 1 postgres=# insert into s2.abc values(456); INSERT 0 1 postgres=# set search_path = s1; SET postgres=# prepare p1 as select f1 from abc; PREPARE postgres=# execute p1; f1 ----- 123 (1 row) postgres=# set search_path = s2; SET postgres=# select f1 from abc; f1 ----- 456 (1 row) postgres=# execute p1; f1 ----- 456 (1 row) postgres=# alter table s1.abc add column f2 float8; ALTER TABLE postgres=# execute p1; f1 ----- 456 (1 row) postgres=# drop schema s1 cascade; NOTICE: drop cascades to table s1.abc DROP SCHEMA postgres=# drop schema s2 cascade; NOTICE: drop cascades to table abc DROP SCHEMA postgres=# reset search_path; RESET
10.检查regclass常量是否有效
postgres=# create sequence seq;
CREATE SEQUENCE
postgres=# prepare p2 as select nextval('seq');
PREPARE
postgres=# execute p2;
## nextval
```
1
```
(1 row)
postgres=# drop sequence seq;
DROP SEQUENCE
postgres=# create sequence seq;
CREATE SEQUENCE
postgres=# execute p2;
## nextval
```
1
```
(1 row)
11.检查DDL,然后立即重新使用SPI plan
postgres=# create function cachebug() returns void as $$ postgres$# declare r int; postgres$# begin postgres$# drop table if exists temptable cascade; postgres$# create temp table temptable as select * from generate_series(1,3) as f1; postgres$# create temp view vv as select * from temptable; postgres$# for r in select * from vv order by 1 loop postgres$# raise notice '%', r; postgres$# end loop; postgres$# end$$ language plpgsql; CREATE FUNCTION postgres=# select cachebug(); NOTICE: table "temptable" does not exist, skipping CONTEXT: SQL statement "drop table if exists temptable cascade" PL/pgSQL function cachebug() line 4 at SQL statement referenced column: cachebug NOTICE: 1 CONTEXT: referenced column: cachebug NOTICE: 2 CONTEXT: referenced column: cachebug NOTICE: 3 CONTEXT: referenced column: cachebug cachebug ---------- (1 row) postgres=# select cachebug(); NOTICE: drop cascades to view vv CONTEXT: SQL statement "drop table if exists temptable cascade" PL/pgSQL function cachebug() line 4 at SQL statement referenced column: cachebug NOTICE: 1 CONTEXT: referenced column: cachebug NOTICE: 2 CONTEXT: referenced column: cachebug NOTICE: 3 CONTEXT: referenced column: cachebug cachebug ---------- (1 row)