vit=# select count(*) from evtags;
count
---------
4496914
vit=# explain select tag from evtags where evid in (1002, 1023);
QUERY PLAN
---------------------------------------------------------------------------------
Index Only Scan using evtags_pkey on evtags (cost=0.00..15.64 rows=12 width=7)
Index Cond: (evid = ANY ('{1002,1023}'::integer[]))
到目前为止,这似乎完全没问题。接下来,我想使用另一个表中的 ID,而不是在查询中指定它们。
vit=# select count(*) from zzz;
count
-------
49738
开始了...
vit=# explain select tag from evtags where evid in (select evid from zzz);
QUERY PLAN
-----------------------------------------------------------------------
Hash Semi Join (cost=1535.11..142452.47 rows=291712 width=7)
Hash Cond: (evtags.evid = zzz.evid)
-> Seq Scan on evtags (cost=0.00..69283.14 rows=4496914 width=11)
-> Hash (cost=718.38..718.38 rows=49738 width=4)
-> Seq Scan on zzz (cost=0.00..718.38 rows=49738 width=4)
为什么要对更大的表进行索引扫描,这样做的正确方法是什么?
编辑
我重新创建了我的zzz
表,现在由于某种原因更好:
vit=# explain analyze select tag from evtags where evid in (select evid from zzz);
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------
Nested Loop (cost=708.00..2699.17 rows=2248457 width=7) (actual time=28.935..805.923 rows=244353 loops=1)
-> HashAggregate (cost=708.00..710.00 rows=200 width=4) (actual time=28.893..54.461 rows=38822 loops=1)
-> Seq Scan on zzz (cost=0.00..601.80 rows=42480 width=4) (actual time=0.032..10.985 rows=40000 loops=1)
-> Index Only Scan using evtags_pkey on evtags (cost=0.00..9.89 rows=6 width=11) (actual time=0.015..0.017 rows=6 loops=38822)
Index Cond: (evid = zzz.evid)
Heap Fetches: 0
Total runtime: 825.651 ms
但经过几次处决后,它变为
vit=# explain analyze select tag from evtags where evid in (select evid from zzz);
QUERY PLAN
---------------------------------------------------------------------------------------------------------------------------------------------------
Merge Semi Join (cost=4184.11..127258.48 rows=235512 width=7) (actual time=38.269..1461.755 rows=244353 loops=1)
Merge Cond: (evtags.evid = zzz.evid)
-> Index Only Scan using evtags_pkey on evtags (cost=0.00..136736.89 rows=4496914 width=11) (actual time=0.038..899.647 rows=3630070 loops=1)
Heap Fetches: 0
-> Materialize (cost=4184.04..4384.04 rows=40000 width=4) (actual time=38.212..61.038 rows=40000 loops=1)
-> Sort (cost=4184.04..4284.04 rows=40000 width=4) (actual time=38.208..51.104 rows=40000 loops=1)
Sort Key: zzz.evid
Sort Method: external sort Disk: 552kB
-> Seq Scan on zzz (cost=0.00..577.00 rows=40000 width=4) (actual time=0.018..8.833 rows=40000 loops=1)
Total runtime: 1484.293 ms
...实际上更慢。有没有办法暗示它是一个“正确的”执行计划?
这些操作的重点是我想对我的数据子集执行多次查询,并想使用单独的临时表来保存我想要处理的记录的 ID。