MySQL in查询优化

https://blog.csdn.net/gua___gua/article/details/47401621

MySQL in查询优化<一>

原创 2015年08月10日 17:57:56

• 5137

开发说他写了个SQL特别慢，让看看。

1. select * from t_channel where id_ in(select distinct cdbh from sjkk_gcjl where jgsj>'2015-01-02 08:00:00' and jgsj<'2015-01-02 12:00:00');
2. ......
3. 30min+

然后我查询内部SQL,只需要3s+

1. mysql> select distinct cdbh from sjkk_gcjl where jgsj>'2015-01-02 08:00:00' and jgsj<'2015-01-02 12:00:00';
2. .....
3. 1755 rows in set (3.30 sec)
4. mysql> select count(*) from t_channel;
5. ....
6. 12062 rows in set (0.70 sec)

开发写的SQL为啥那么慢呢？看看执行计划

1. explain extended select * from t_channel where id_ in(select distinct cdbh from sjkk_gcjl where jgsj>'2015-01-02 08:00:00' and jgsj<'2015-01-02 12:00:00');
2. +----+--------------------+-----------+-------+---------------+----------+---------+------+--------+----------+------------------------------+
3. | id | select_type        | table     | type  | possible_keys | key      | key_len | ref  | rows   | filtered | Extra                        |
4. +----+--------------------+-----------+-------+---------------+----------+---------+------+--------+----------+------------------------------+
5. |  1 | PRIMARY            | t_channel | ALL   | NULL          | NULL     | NULL    | NULL |  12062 |   100.00 | Using where                  |
6. |  2 | DEPENDENT SUBQUERY | sjkk_gcjl | range | idx_jgsj      | idx_jgsj | 8       | NULL | 731868 |   100.00 | Using where; Using temporary |
7. +----+--------------------+-----------+-------+---------------+----------+---------+------+--------+----------+------------------------------

看看数据库转换后的语句

1. mysql> show warnings;
2. SELECT
3. `shanghai_full`.`t_channel`.`ID_` AS `ID_`,
4. `shanghai_full`.`t_channel`.`Code_` AS `Code_`,
5. ...... --这里会列出所有字段
6. FROM
7. `shanghai_full`.`t_channel`
8. WHERE
9. < in_optimizer > (
10. `shanghai_full`.`t_channel`.`ID_` ,< EXISTS > (
11. SELECT DISTINCT
12. 1
13. FROM
14. `shanghai_full`.`sjkk_gcjl`
15. WHERE
16. (
17. (
18. `shanghai_full`.`sjkk_gcjl`.`jgsj` > '2015-01-02 08:00:00'
19. )
20. AND (
21. `shanghai_full`.`sjkk_gcjl`.`jgsj` < '2015-01-02 12:00:00'
22. )
23. AND (
24. < CACHE > (
25. `shanghai_full`.`t_channel`.`ID_`
26. ) = `shanghai_full`.`sjkk_gcjl`.`cdbh`
27. )
28. )
29. )
30. );

可见，经过mysql优化器后，in 给转换成exists的方式（mysql认为换成exists更快，呵呵）。

慢的原因：走了exists，查询把外表t_channel做为主表进行全表扫描，每扫描一行数据，然后对子查询进行查询看这条数据是否符合条件。

特别说明：mysql版本是5.5。在5.6中mysql做了改进，将in的这种查询转换为了join。

优化方式一（将in查询转换为连接查询）

1. select t1.* from t_channel t1,(select distinct cdbh from sjkk_gcjl where jgsj>'2015-01-02 08:00:00' and jgsj<'2015-01-02 12:00:00') t2 where t1.id_=t2.cdbh;
2. ......
3. 1264 rows in set (3.30 sec)
4. mysql> explain extended select t1.* from t_channel t1,(select distinct cdbh from sjkk_gcjl where jgsj>'2015-01-02 08:00:00' and jgsj<'2015-01-02 12:00:00') t2 where t1.id_www.255055.cn/ =t2.cdbh;
5. +----+-------------+------------+--------+---------------+----------+---------+---------+--------+----------+------------------------------+
6. | id | select_type | table      | type   | possible_keys | key      | key_len | ref     | rows   | filtered | Extra                        |
7. +----+-------------+------------+--------+---------------+----------+---------+---------+--------+----------+------------------------------+
8. |  1 | PRIMARY     | <derived2> | ALL    | NULL          | NULL     | NULL    | NULL    |   1755 |   100.00 |                              |
9. |  1 | PRIMARY     | t1    www.boshenyl.cn      | eq_ref | PRIMARY,ID_   | PRIMARY  | 74      | t2.cdbh |      1 |   100.00 | Using where                  |
10. |  2 | DERIVED     | sjkk_gcjl  | range  | idx_jgsj      | idx_jgsj | 8       | NULL    | 731868 |   100.00 | Using where; Using temporary |
11. +----+-------------+------------+--------+---------------+----------+---------+---------+--------+----------+------------------------------+
12. mysql> show warnings;
13. SELECT
14. `shanghai_full`.`t1`.`ID_` AS `ID_`,
15. ........ ---这里会列出所有字段FROM
16. `shanghai_full`.`t_channel` `t1`
17. JOIN (
18. SELECT DISTINCT
19. `shanghai_full`.`sjkk_gcjl`.`cdbh` AS `www.cnzhaotai.com cdbh`
20. FROM
21. `shanghai_full`.www.fengshen157.com `sjkk_gcjl`
22. WHERE
23. (
24. (
25. `shanghai_full`.`sjkk_gcjl`.`jgsj` > '2015-01-02 08:00:00'
26. )
27. AND (
28. `shanghai_full`.`sjkk_gcjl`.`jgsj` < '2015-01-02 12:00:00'
29. )
30. )
31. ) `t2`
32. WHERE
33. (
34. `shanghai_full`.`t1`.`ID_` = www.taohuayuan178.com `t2`.`cdbh`
35. );

优化方式二（使用memory引擎的临时表）

1. mysql> create temporary table tmp_www.yibaoyule1.com channel  engine=memory (select distinct cdbh from sjkk_gcjl where jgsj>'2015-01-02 08:00:00' and jgsj<'2015-01-02 12:00:00');
2. Query OK, 1755 rows affected (9.00 sec)
3. Records: 1755  Duplicates: 0  Warnings: 0
4. mysql> select * from t_channel where id_ in(select * from tmp_channel);
5. .....
6. 1264 rows in set (0.26 sec)
7. mysql> explain extended select * from t_channel where id_ in(select * from tmp_channel);
8. +----+--------------------+-------------+------+---------------+------+---------+------+------+----------+-------------+
9. | id | select_type        | table       | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
10. +----+--------------------+-------------+------+---------------+------+---------+------+------+----------+-------------+
11. |  1 | PRIMARY            | t_channel   | ALL  | NULL          | NULL | NULL    | NULL | 3224 |   100.00 | Using where |
12. |  2 | DEPENDENT SUBQUERY | tmp_channel | ALL  | NULL          | NULL | NULL    | NULL |   20 |   100.00 | Using where |
13. +----+--------------------+-------------+------+---------------+------+---------+------+------+----------+-------------+
14. mysql> show warnings;
15. | Note  | 1003 | select `vmc_jiaqi`.`t_channel`.`ID_` AS `ID_`,`vmc_jiaqi`.`t_channel`.`Code_` AS `Code_`,`vmc_jiaqi`.`t_channel`.`HostId_` AS `HostId_`,`vmc_jiaqi`.`t_channel`.`RoadMonitorStationId_` AS `RoadMonitorStationId_`,`vmc_jiaqi`.`t_channel`.`ChannelNo_` AS `ChannelNo_`,`vmc_jiaqi`.`t_channel`.`Name_` AS `Name_`,`vmc_jiaqi`.`t_channel`.`ChannelType_` AS `ChannelType_`,`vmc_jiaqi`.`t_channel`.`Ext_` AS `Ext_`,`vmc_jiaqi`.`t_channel`.`DeviceAddress_` AS `DeviceAddress_`,`vmc_jiaqi`.`t_channel`.`MinSpeed_`www.mhylpt.comS `MinSpeed_`,`vmc_jiaqi`.`t_channel`.`MaxSpeed_` AS `MaxSpeed_`,`vmc_jiaqi`.`t_channel`.`LimitMinRatio_` AS `LimitMinRatio_`,`vmc_jiaqi`.`t_channel`.`LimitMaxRatio_` AS `LimitMaxRatio_`,`vmc_jiaqi`.`t_channel`.`Direction_` AS `Direction_`,`vmc_jiaqi`.`t_channel`.`JcDirection_` AS `JcDirection_`,`vmc_jiaqi`.`t_channel`.`LaneNum_` AS `LaneNum_`,`vmc_jiaqi`.`t_channel`.`FrameDropNum_` AS `FrameDropNum_`,`vmc_jiaqi`.`t_channel`.`RecogizeRectLeft_` AS `RecogizeRectLeft_`,`vmc_jiaqi`.`t_channel`.`RecogizeRectTop_` AS `RecogizeRectTop_`,`vmc_jiaqi`.`t_channel`.`RecogizeRectWidth_` AS `RecogizeRectWidth_`,`vmc_jiaqi`.`t_channel`.`RecogizeRectHeight_` AS `RecogizeRectHeight_`,`vmc_jiaqi`.`t_channel`.`RmpServerIp_` AS `RmpServerIp_`,`vmc_jiaqi`.`t_channel`.`RmpServerPort_` AS `RmpServerPort_`,`vmc_jiaqi`.`t_channel`.`VirtualGateServerId_` AS `VirtualGateServerId_`,`vmc_jiaqi`.`t_channel`.`LaneType_` AS `LaneType_`,`vmc_jiaqi`.`t_channel`.`DeviceType_` AS `DeviceType_`,`vmc_jiaqi`.`t_channel`.`ManufacturerId_` AS `ManufacturerId_`,`vmc_jiaqi`.`t_channel`.`IsLocalSavePicture_` AS `IsLocalSavePicture_`,`vmc_jiaqi`.`t_channel`.`OrderNO_` AS `OrderNO_`,`vmc_jiaqi`.`t_channel`.`channelStatus` AS `channelStatus` from `vmc_jiaqi`.`t_channel` where <in_optimizer>(`vmc_jiaqi`.`t_channel`.`ID_`,<exists>(select 1 from `vmc_jiaqi`.`tmp_channel` where (<cache>(`vmc_jiaqi`.`t_channel`.`ID_`) = `vmc_jiaqi`.`tmp_channel`.`cdbh`)))

注意：第二种方式还是使用了exists的执行方式，所以这种方式没有第一种方式好，在特定的条件下可能会有用处，