MySQL 子查询优化案例

开发人员给了一个sql ，结构如下delete from B where ID in (select NID from H where guid='xxx');

内部sql满足条件的结果集只有一条，但是整个删除操作执行了将近1分钟，如果是将结果集放在括号里或者将in改为= ，执行的速度可以实现毫秒级别

但是如果内部查询结果集多于一行，采用第一种方案的话需要更改程序，后来又试了一种更改为join，速度也是极快。

测试表，t1.id上有索引，t2.id无索引

mysql> select * from t1;         mysql> select * from t2;    
+------+------+----------+       +------+---------+          
| id   | name | class_id |       | id   | name    |          
+------+------+----------+       +------+---------+          
|    1 | aa   |     NULL |       |    2 | myname2 |          
|    2 | aa   |     NULL |       |    6 | myname5 |          
|    3 | dd   |     NULL |       +------+---------+          
|    6 | cc   |     NULL |       2 rows in set (0.01 sec)    
+------+------+----------+                                   
4 rows in set (0.00 sec)                                     

使用子查询及改为join后的执行计划

mysql> explain delete from t1 where id in (select id from t2 where name='aa');
+----+--------------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
| id | select_type        | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+--------------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
|  1 | DELETE             | t1    | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    4 |   100.00 | Using where |
|  2 | DEPENDENT SUBQUERY | t2    | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    2 |    50.00 | Using where |
+----+--------------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
2 rows in set (0.00 sec)     

mysql> explain delete t1.* from t1 inner join t2 where t1.id=t2.id and t2.name='aa';
+----+-------------+-------+------------+------+---------------+--------+---------+-------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key    | key_len | ref   | rows | filtered | Extra       |
+----+-------------+-------+------------+------+---------------+--------+---------+-------+------+----------+-------------+
|  1 | SIMPLE      | t2    | NULL       | ALL  | NULL          | NULL   | NULL    | NULL  |    2 |    50.00 | Using where |
|  1 | DELETE      | t1    | NULL       | ref  | idx_id        | idx_id | 5       | const |    1 |   100.00 | Using where |
+----+-------------+-------+------------+------+---------------+--------+---------+-------+------+----------+-------------+
2 rows in set (0.01 sec)  

对于子查询的执行计划可以看出先对t1进行全表扫描，然后执行select id from t2 where name='aa' and t1.id=t2.id ，如果有值则删除t.* where id=t1.id

而对于改为join的sql来说，优化器会很智能的选取小表来作为驱动表，然后再走索引删除t1.* , 而对于子查询官方文档解释为由外向内执行

为了更加直观的看两种方式的执行过程，打开回话级别的profiling

mysql> show profiles;
+----------+------------+------------------------------------------------------------------------------+
| Query_ID | Duration   | Query                                                                        |
+----------+------------+------------------------------------------------------------------------------+
|        3 | 0.00137075 | delete from t1 where id in (select id from t2 where name='aa')               |
|        4 | 0.00211725 | explain delete t1.* from t1 inner join t2 where t1.id=t2.id and t2.name='aa' |
|        5 | 0.00132050 | delete t1.* from t1 inner join t2 where t1.id=t2.id and t2.name='aa'         |
+----------+------------+------------------------------------------------------------------------------+

mysql> show profile for query 3                mysql> show profile for query 5
    -> ;                                           -> ;
+----------------------+----------+            +--------------------------------+----------+
| Status               | Duration |            | Status                         | Duration |
+----------------------+----------+            +--------------------------------+----------+
| starting             | 0.000388 |            | starting                       | 0.000360 |
| checking permissions | 0.000026 |            | checking permissions           | 0.000013 |
| checking permissions | 0.000008 |            | checking permissions           | 0.000007 |
| Opening tables       | 0.000105 |            | checking permissions           | 0.000004 |
| init                 | 0.000152 |            | init                           | 0.000005 |
| System lock          | 0.000083 |            | Opening tables                 | 0.000048 |
| updating             | 0.000084 |            | init                           | 0.000048 |
| optimizing           | 0.000031 |            | deleting from main table       | 0.000022 |
| statistics           | 0.000083 |            | System lock                    | 0.000028 |
| preparing            | 0.000052 |            | optimizing                     | 0.000043 |
| executing            | 0.000013 |            | statistics                     | 0.000144 |
| Sending data         | 0.000114 |            | preparing                      | 0.000144 |
| executing            | 0.000009 |            | executing                      | 0.000009 |
| Sending data         | 0.000017 |            | Sending data                   | 0.000246 |
| executing            | 0.000005 |            | deleting from reference tables | 0.000073 |
| Sending data         | 0.000019 |            | end                            | 0.000012 |
| executing            | 0.000006 |            | end                            | 0.000010 |
| Sending data         | 0.000018 |            | query end                      | 0.000016 |
| end                  | 0.000019 |            | closing tables                 | 0.000015 |
| query end            | 0.000020 |            | freeing items                  | 0.000037 |
| closing tables       | 0.000021 |            | cleaning up                    | 0.000039 |
| freeing items        | 0.000054 |            +--------------------------------+----------+
| cleaning up          | 0.000046 |            21 rows in set, 1 warning (0.00 sec)
+----------------------+----------+
23 rows in set, 1 warning (0.01 sec)         

我第一眼关注的是两条语句senting data的次数，子查询对应的sending data是4次，子查询先对外部表进行全表扫描，结果集是4行，然后进行循环遍历拿出每一行与内部查询进行关联，共执行了4次内部查询，并且每次都对内部查询的结果集做一下判断是否有值，如果有值则再进行删除

小小的记录一下，在优化器的探索之路上慢慢爬