-name tails join spa csdn duplicate 查詢 tracking 全表掃描

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

MySQL in查詢優化<一>

原創 2015年08月10日 17:57:56

• 5137

開發說他寫了個SQL特別慢，讓看看。

[html] view plain copy

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+

然後我查詢內部SQL,只需要3s+

[html] view plain copy

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)
5. mysql> select count(*) from t_channel;
6. ....
7. 12062 rows in set (0.70 sec)

開發寫的SQL為啥那麽慢呢？看看執行計劃

[html] view plain copy

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. +----+--------------------+-----------+-------+---------------+----------+---------+------+--------+----------+------------------------------

看看數據庫轉換後的語句
[html] view plain copy

1. mysql> show warnings;
2. SELECT
3. `shanghai_full`.`t_channel`.`ID_` AS `ID_`,
4. `shanghai_full`.`t_channel`.`Code_` AS `Code_`,
5. ...... --這裏會列出所有字段
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. );

可見，經過mysql優化器後，in 給轉換成exists的方式（mysql認為換成exists更快，呵呵）。

慢的原因：走了exists，查詢把外表t_channel做為主表進行全表掃描，每掃描一行數據，然後對子查詢進行查詢看這條數據是否符合條件。

特別說明：mysql版本是5.5。在5.6中mysql做了改進，將in的這種查詢轉換為了join。

優化方式一（將in查詢轉換為連接查詢）

[html] view plain copy

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)
5. 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;
6. +----+-------------+------------+--------+---------------+----------+---------+---------+--------+----------+------------------------------+
7. | id | select_type | table | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
8. +----+-------------+------------+--------+---------------+----------+---------+---------+--------+----------+------------------------------+
9. | 1 | PRIMARY | <derived2> | ALL | NULL | NULL | NULL | NULL | 1755 | 100.00 | |
10. | 1 | PRIMARY | t1 www.boshenyl.cn | eq_ref | PRIMARY,ID_ | PRIMARY | 74 | t2.cdbh | 1 | 100.00 | Using where |
11. | 2 | DERIVED | sjkk_gcjl | range | idx_jgsj | idx_jgsj | 8 | NULL | 731868 | 100.00 | Using where; Using temporary |
12. +----+-------------+------------+--------+---------------+----------+---------+---------+--------+----------+------------------------------+
14. mysql> show warnings;
16. SELECT
17. `shanghai_full`.`t1`.`ID_` AS `ID_`,
18. ........ ---這裏會列出所有字段FROM
19. `shanghai_full`.`t_channel` `t1`
20. JOIN (
21. SELECT DISTINCT
22. `shanghai_full`.`sjkk_gcjl`.`cdbh` AS `www.cnzhaotai.com cdbh`
23. FROM
24. `shanghai_full`.www.fengshen157.com `sjkk_gcjl`
25. WHERE
26. (
27. (
28. `shanghai_full`.`sjkk_gcjl`.`jgsj` > ‘2015-01-02 08:00:00‘
29. )
30. AND (
31. `shanghai_full`.`sjkk_gcjl`.`jgsj` < ‘2015-01-02 12:00:00‘
32. )
33. )
34. ) `t2`
35. WHERE
36. (
37. `shanghai_full`.`t1`.`ID_` = www.taohuayuan178.com `t2`.`cdbh`
38. );

優化方式二（使用memory引擎的臨時表）

[html] view plain copy

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)
8. mysql> explain extended select * from t_channel where id_ in(select * from tmp_channel);
9. +----+--------------------+-------------+------+---------------+------+---------+------+------+----------+-------------+
10. | id | select_type | table | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
11. +----+--------------------+-------------+------+---------------+------+---------+------+------+----------+-------------+
12. | 1 | PRIMARY | t_channel | ALL | NULL | NULL | NULL | NULL | 3224 | 100.00 | Using where |
13. | 2 | DEPENDENT SUBQUERY | tmp_channel | ALL | NULL | NULL | NULL | NULL | 20 | 100.00 | Using where |
14. +----+--------------------+-------------+------+---------------+------+---------+------+------+----------+-------------+
16. mysql> show warnings;
17. | 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`)))

註意：第二種方式還是使用了exists的執行方式，所以這種方式沒有第一種方式好，在特定的條件下可能會有用處，

MySQL in查詢優化