Question

我遇到了＆＃34; sub select＆＃34;查询查询：

select
 f.timestamp::date as date,
   user_id,
   activity_type,
   f.container_id as group_id,
   (
      select
          string_agg(distinct("userId"), ',') as group_owners
        from
          jusers_groups_copy g
        where
          g.place_id = f.container_id
          and state like 'owner'
    ) as group_owners
 from
   fact_activity f
 where
   f.container_type like '700'
   and f.timestamp::date < to_date('2016-09-05', 'YYYY-MM-DD')
 group by
  date, user_id, activity_type, group_id
 order by
  date, user_id, activity_type, group_id

实际上，string_add内部需要20秒才能渲染。我使用pgAdmin来解释查询，他给了我这条消息：

"Group  (cost=7029.62..651968.20 rows=17843 width=27) (actual time=431.017..4513.973 rows=11483 loops=1)"
"  Buffers: shared hit=139498 read=411, temp read=255 written=255"
"  ->  Sort  (cost=7029.62..7074.90 rows=18111 width=27) (actual time=430.630..667.098 rows=54660 loops=1)"
"        Sort Key: ((f."timestamp")::date), f.user_id, f.activity_type, f.container_id"
"        Sort Method: external merge  Disk: 2008kB"
"        Buffers: shared hit=1702 read=411, temp read=255 written=255"
"        ->  Seq Scan on fact_activity f  (cost=0.00..5748.76 rows=18111 width=27) (actual time=0.107..188.827 rows=54660 loops=1)"
"              Filter: ((container_type ~~ '700'::text) AND (("timestamp")::date < to_date('2016-09-05'::text, 'YYYY-MM-DD'::text)))"
"              Rows Removed by Filter: 125414"
"              Buffers: shared hit=1691 read=411"
"  SubPlan 1"
"    ->  Aggregate  (cost=36.12..36.13 rows=1 width=5) (actual time=0.315..0.318 rows=1 loops=11483)"
"          Buffers: shared hit=137796"
"          ->  Seq Scan on users_groups_copy g  (cost=0.00..36.09 rows=11 width=5) (actual time=0.041..0.266 rows=13 loops=11483)"
"                Filter: ((state ~~ 'owner'::text) AND (place_id = f.container_id))"
"                Rows Removed by Filter: 1593"
"                Buffers: shared hit=137796"
"Total runtime: 4536.074 ms"

此外，我尝试加入表格但请求速度更慢，如下所示：

select
 f.timestamp::date as date,
   user_id,
   activity_type,
   f.container_id as group_id,
   string_agg(distinct("userId"), ',') as group_owners
 from
   fact_activity f
 join jusers_groups_copy g
 on g.place_id = f.container_id
 where
   f.container_type like '700'
   and f.timestamp::date < to_date('2016-09-05', 'YYYY-MM-DD')
   and g.state like 'owner'
 group by
  date, user_id, activity_type, group_id
 order by
  date, user_id, activity_type, group_id

最后，这个数据库中有任何索引，这是为什么请求缓慢？

我想知道如何改进此请求。

提前致谢

Answer 1

在不更改查询的情况下，最大的性能提升将是子选择中加速子选择的表上的索引：

CREATE INDEX nice_name ON jusers_groups_copy(place_id, state text_pattern_ops);

但我会将查询重写为连接。这样，您可能会获得比嵌套循环更高效的内容，具体取决于您的数据。

而不是

SELECT f.somecol,
   (SELECT g.othercol
    FROM jusers_groups_copy g
    WHERE g.place_id = f.container_id
      AND g.state LIKE 'owner')
FROM fact_activity f
WHERE ...;

你应该写

SELECT f.somecol, g.othercol
FROM fact_activity f
   JOIN jusers_groups_copy g
      ON g.place_id = f.container_id
WHERE g.state LIKE 'owner'
  AND ...;

根据所选的连接类型，上面的索引（对于嵌套循环）或不同的索引可以使查询更快。

Answer 2

我想你需要在/data/postgresql.conf中更改一些配置使用以下网站

pgtune

我认为最重要的参数是“work_mem”

慢查询postgreSQL

2 个答案: