postgresql获取真正的execution plan

xuefuhe2026-02-04 9:48

analyze会真正执行，除了select语句，最好放在一个事务中运行，然后rollback.

BEGIN;

EXPLAIN ANALYZE UPDATE ...;

ROLLBACK;

常用方式如下：

EXPLAIN (ANALYZE, BUFFERS, TIMING) SELECT ...;

sql 复制代码

ccc=# select * from t1;
 id | name 
----+------
  1 | aaa
  2 | bbb
  3 | ccc
  4 | ddd
(4 rows)

ccc=# begin;
BEGIN
ccc=*# EXPLAIN (ANALYZE, BUFFERS, TIMING) update t1 set name='yyy' where id=3;
                                              QUERY PLAN                                               
-------------------------------------------------------------------------------------------------------
 Update on t1  (cost=0.00..21.25 rows=0 width=0) (actual time=0.113..0.114 rows=0.00 loops=1)
   Buffers: shared hit=3 dirtied=1
   ->  Seq Scan on t1  (cost=0.00..21.25 rows=4 width=64) (actual time=0.012..0.014 rows=1.00 loops=1)
         Filter: (id = 3)
         Rows Removed by Filter: 3
         Buffers: shared hit=1
 Planning:
   Buffers: shared hit=3 read=6
 Planning Time: 2.737 ms
 Execution Time: 2.241 ms
(10 rows)

ccc=*# rollback;
ROLLBACK
ccc=# select * from t1;
 id | name 
----+------
  1 | aaa
  2 | bbb
  3 | ccc
  4 | ddd
(4 rows)

ccc=#

调试执行计划时可以临时enable/disable一些选项，详细列表见：

https://www.postgresql.org/docs/current/runtime-config-query.html

-- Plan with nested loop allowed (default)

EXPLAIN (ANALYZE, FORMAT JSON)

SELECT ... ;

-- Compare with nested loop discouraged

BEGIN;

SET LOCAL enable_nestloop = off;

EXPLAIN (ANALYZE, FORMAT JSON)

SELECT ... ;

COMMIT;