Question

嘿伙计们，尝试优化此查询以解决重复的用户问题：

SELECT userid, 'ismaster' AS name, 'false' AS propvalue FROM user 
WHERE userid NOT IN (SELECT userid FROM userprop WHERE name = 'ismaster');

问题在于NOT IN之后的选择是120.000记录并且它将永远消失。

使用注释中建议的解释前缀返回：

                                    QUERY PLAN

--------------------------------------------------------------------------------
--
 Seq Scan on user  (cost=5559.38..122738966.99 rows=61597 width=8)
   Filter: (NOT (SubPlan 1))
   SubPlan 1
     ->  Materialize  (cost=5559.38..7248.33 rows=121395 width=8)
           ->  Seq Scan on userprop  (cost=0.00..4962.99 rows=121395 width=8
)
                 Filter: ((name)::text = 'ismaster'::text)
(6 rows)

有什么建议吗？

Answer 1

您是否在userid上添加了索引？

或尝试其他变体：

SELECT userid, 'ismaster' AS name, 'false' AS propvalue FROM user 
WHERE NOT EXISTS 
 (SELECT * FROM userprop 
 WHERE userpop.userid = user.userid 
   AND name = 'ismaster');

Answer 2

名称列是否已编入索引？名称值的选择性如何？此外，只要您希望某人建议对查询进行更改，您就可以提供查询计划，即使是看似简单的查询也是如此。这样我们就知道规划师正在做什么。

Answer 3

根据this answer，使用LEFT JOIN ... IS NULL可能比NOT EXISTS更快或更慢，具体取决于RDBMS，但它们在PostGres上是等效的。

SELECT u.userid, 'ismaster' AS name, 'false' AS propvalue FROM user u
LEFT JOIN userprop up ON u.userid = up.userid AND up.name <> 'ismaster'
WHERE up.userid IS NULL

Postgres查询优化

3 个答案: