MySQL的子查询及相关优化学习教程

一、子查询
1、where型子查询
（把内层查询结果当作外层查询的比较条件）

#不用order by 来查询最新的商品
select goods_id,goods_name from goods where goods_id = (select max(goods_id) from goods);

#取出每个栏目下最新的产品(goods_id唯一)
select cat_id,goods_id,goods_name from goods where goods_id in(select max(goods_id) from goods group by cat_id); 

2、from型子查询
(把内层的查询结果供外层再次查询)
#用子查询查出挂科两门及以上的同学的平均成绩
思路：

#先查出哪些同学挂科两门以上
select name,count(*) as gk from stu where score < 60 having gk >=2;
#以上查询结果，我们只要名字就可以了，所以再取一次名字
select name from (select name,count(*) as gk from stu having gk >=2) as t;
#找出这些同学了，那么再计算他们的平均分
select name,avg(score) from stu where name in (select name from (select name,count(*) as gk from stu having gk >=2) as t) group by name;

3、exists型子查询
（把外层查询结果拿到内层，看内层的查询是否成立）

#查询哪些栏目下有商品，栏目表category,商品表goods
select cat_id,cat_name from category where exists(select * from goods where goods.cat_id = category.cat_id);

二、优化
从句式的形式看，子查询分为特殊格式子查询和非特殊格式子查询，特殊格式的子查询中又包括in、all、any、some、exists等类型的子查询，对于有的类型的子查询，mysql有的支持优化，有的不支持，具体情况如下。

 

示例一，mysql不支持对exists类型的子查询的优化：

exists类型的相关子查询，查询执行计划如下：

mysql> explain extended select * from t1 where exists (select 1 from t2 where t1.a1= t2.a2 and t2.a2>10);

+----+--------------------+-------+------+------+-------------+
| id | select_type    | table | type | key | extra    |

+----+--------------------+-------+------+------+-------------+

| 1 | primary      | t1  | all | null | using where |

| 2 | dependent subquery | t2  | all | null | using where |

+----+--------------------+-------+------+------+-------------+

2 rows in set, 2 warnings (0.00 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

  `test`.`t1`.`b1` as `b1`

from `test`.`t1`

where exists(/* select#2 */

  select 1

  from `test`.`t2`

  where ((`test`.`t1`.`a1` = `test`.`t2`.`a2`) and (`test`.`t2`.`a2` > 10))

)

从查询执行计划看，子查询存在，mysql没有进一步做子查询的优化工作。

另外的一个exists类型的相关子查询，查询执行计划如下：

mysql> explain extended select * from t1 where exists (select 1 from t2 where t1.b1= t2.b2 and t1.a1=10);

+----+--------------------+-------+------+------+-------------+

| id | select_type    | table | type | key | extra    |

+----+--------------------+-------+------+------+-------------+

| 1 | primary      | t1  | all | null | using where |

| 2 | dependent subquery | t2  | all | null | using where |

+----+--------------------+-------+------+------+-------------+

2 rows in set, 3 warnings (0.02 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

  `test`.`t1`.`b1` as `b1`

from `test`.`t1`

where exists(/* select#2 */

  select 1

  from `test`.`t2`

  where ((`test`.`t1`.`b1` = `test`.`t2`.`b2`) and (`test`.`t1`.`a1` = 10))

)

从查询执行计划看，子查询存在，mysql没有进一步做子查询的优化工作。

 

示例二，mysql不支持对not exists类型的子查询的优化：

not exists类型的相关子查询，查询执行计划如下：

mysql> explain extended select * from t1 where not exists (select 1 from t2 where t1.a1= t2.a2 and t2.a2>10);

+----+--------------------+-------+------+------+-------------+

| id | select_type    | table | type | key | extra    |

+----+--------------------+-------+------+------+-------------+

| 1 | primary      | t1  | all | null | using where |

| 2 | dependent subquery | t2  | all | null | using where |

+----+--------------------+-------+------+------+-------------+

2 rows in set, 2 warnings (0.00 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

  `test`.`t1`.`b1` as `b1`

from `test`.`t1`

where (not(exists(

  /* select#2 */ select 1

  from `test`.`t2`

  where ((`test`.`t1`.`a1` = `test`.`t2`.`a2`) and (`test`.`t2`.`a2` > 10))))

)

从查询执行计划看，子查询存在，mysql没有进一步做子查询的优化工作。

 

另外的一个not exists类型的相关子查询，查询执行计划如下：

mysql> explain extended select * from t1 where not exists (select 1 from t2 where t1.b1= t2.b2 and t1.a1=10);

+----+--------------------+-------+------+------+-------------+

| id | select_type    | table | type | key | extra    |

+----+--------------------+-------+------+------+-------------+

| 1 | primary      | t1  | all | null | using where |

| 2 | dependent subquery | t2  | all | null | using where |

+----+--------------------+-------+------+------+-------------+

2 rows in set, 3 warnings (0.00 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

  `test`.`t1`.`b1` as `b1`

from `test`.`t1`

where (not(exists(

  /* select#2 */ select 1

  from `test`.`t2`

  where ((`test`.`t1`.`b1` = `test`.`t2`.`b2`) and (`test`.`t1`.`a1` = 10))))

)

从查询执行计划看，子查询存在，mysql没有进一步做子查询的优化工作。

 

示例三，mysql支持对in类型的子查询的优化，按也有不支持的情况存在：

in非相关子查询，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 in (select a2 from t2 where t2.a2>10);

+----+--------------+-------------+------+------+----------------------------------------------------+

| id | select_type | table    | type | key | extra  |

+----+--------------+-------------+------+------+----------------------------------------------------+

| 1 | simple    | <subquery2> | all | null | null  |

| 1 | simple    | t1     | all | null | using where; using join buffer (block nested loop) |

| 2 | materialized | t2     | all | null | using where  |

+----+--------------+-------------+------+------+----------------------------------------------------+

3 rows in set, 1 warning (0.00 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

  `test`.`t1`.`b1` as `b1`

from `test`.`t1` semi join (`test`.`t2`)

where ((`test`.`t1`.`a1` = `<subquery2>`.`a2`) and (`test`.`t2`.`a2` > 10))

从查询执行计划看，表t2被物化后，与表t1执行了半连接（semi join）。尽管有“subquery2”这样的内容看起来是子查询，但是表t2已经被上拉到表t1层执行了半连接，所以mysql支持in子查询优化为半连接操作。

 

另外一个in非相关子查询，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 in (select a2 from t2 where t2.a2=10);

+----+--------------+-------------+------+------+----------------------------------------------------+

| id | select_type | table    | type | key | extra  |

+----+--------------+-------------+------+------+----------------------------------------------------+

| 1 | simple    | <subquery2> | all | null | using where  |

| 1 | simple    | t1     | all | null | using where; using join buffer (block nested loop) |

| 2 | materialized | t2     | all | null | using where  |

+----+--------------+-------------+------+------+----------------------------------------------------+

3 rows in set, 1 warning (0.02 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

  `test`.`t1`.`b1` as `b1`

from `test`.`t1` semi join (`test`.`t2`)

where ((`<subquery2>`.`a2` = 10) and (`test`.`t1`.`a1` = 10) and (`test`.`t2`.`a2` = 10))

从查询执行计划看，子查询不存在，表t1和t2直接做了块嵌套循环半连接（block nested loop），把子查询上拉到父查询中用嵌套循环半连接完成in操作。另外，由于子查询上拉，使得增加连接条件“a1=a2”，而原先的条件“a2=10”可以利用常量传递优化技术，使得“a1=a2=10”，所以查询执行计划中，两个索引扫描的条件分别为：a1 = 10、a2 = 10。

 

另外一个in非相关子查询，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 in (select a2 from t2 where t1.a1=10);

+----+-------------+-------+------+------------------------------------------------------------------+

| id | select_type | table | type | extra      |

+----+-------------+-------+------+------------------------------------------------------------------+

| 1 | simple   | t2  | all | using where; start temporary      |

| 1 | simple   | t1  | all | using where; end temporary; using join buffer (block nested loop)|

+----+-------------+-------+------+------------------------------------------------------------------+

2 rows in set, 2 warnings (0.00 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

  `test`.`t1`.`b1` as `b1`

from `test`.`t1` semi join (`test`.`t2`)

where ((`test`.`t2`.`a2` = 10) and (`test`.`t1`.`a1` = 10))

从查询执行计划看，子子查询不存在，表t1和t2直接做了块嵌套循环连接（block nested loop），但属于半连接操作（semi join），把子查询上拉到父查询中用嵌套循环半连接完成in操作。

 

示例四，mysql支持对not in类型的子查询的优化

not in非相关子查询，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 not in (select a2 from t2 where t2.a2>10);

+----+-------------+-------+------+------+-------------+

| id | select_type | table | type | key | extra    |

+----+-------------+-------+------+------+-------------+

| 1 | primary   | t1  | all | null | using where |

| 2 | subquery  | t2  | all | null | using where |

+----+-------------+-------+------+------+-------------+

2 rows in set, 1 warning (0.02 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

`test`.`t1`.`b1` as `b1`

from `test`.`t1`

where (not(<in_optimizer>(

  `test`.`t1`.`a1`,`test`.`t1`.`a1` in (

    <materialize> (/* select#2 */

      select `test`.`t2`.`a2`

      from `test`.`t2`

      where (`test`.`t2`.`a2` > 10)

      having 1

    ),

    <primary_index_lookup>(

      `test`.`t1`.`a1` in <temporary table> on <auto_key>

      where ((`test`.`t1`.`a1` = `materialized-subquery`.`a2`))

    )

   )

  ))

)

从查询执行计划看，表t2做了子查询（subquery）。而子查询被物化（materialize）。所以，mysql对于not in子查询采用了物化的优化方式，但不支持子查询的消除。

 

另外一个not in非相关子查询，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 not in (select a2 from t2 where t2.a2=10);

+----+-------------+-------+------+------+-------------+

| id | select_type | table | type | key | extra    |

+----+-------------+-------+------+------+-------------+

| 1 | primary   | t1  | all | null | using where |

| 2 | subquery  | t2  | all | null | using where |

+----+-------------+-------+------+------+-------------+

2 rows in set, 1 warning (0.00 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,`test`.`t1`.`b1` as `b1`

from `test`.`t1`

where (not(<in_optimizer>(

  `test`.`t1`.`a1`,`test`.`t1`.`a1` in (

    <materialize> (/* select#2 */

      select `test`.`t2`.`a2`

      from `test`.`t2`

      where (`test`.`t2`.`a2` = 10)

      having 1

    ),

    <primary_index_lookup>(

      `test`.`t1`.`a1` in <temporary table> on <auto_key>

      where ((`test`.`t1`.`a1` = `materialized-subquery`.`a2`))

    )

  )

  ))

)

从查询执行计划看，表t2做了子查询（subquery）。而子查询被物化（materialize）。所以，mysql对于not in子查询采用了物化的优化方式，但不支持子查询的消除。

 

示例五，mysql支持对all类型的子查询的优化：

不相关的all子查询，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 >all (select a2 from t2 where t2.a2>10);

+----+-------------+-------+------+------+-------------+

| id | select_type | table | type | key | extra    |

+----+-------------+-------+------+------+-------------+

| 1 | primary   | t1  | all | null | using where |

| 2 | subquery  | t2  | all | null | using where |

+----+-------------+-------+------+------+-------------+

2 rows in set, 1 warning (0.00 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,`test`.`t1`.`b1` as `b1`

from `test`.`t1`

where <not>((`test`.`t1`.`a1` <= <max>(

  /* select#2 */

  select `test`.`t2`.`a2`

  from `test`.`t2`

  where (`test`.`t2`.`a2` > 10)

  )

))

从查询执行计划看，出现了子查询（subquery），但是，子查询被“<= <max>”操作符限制，而子查询中的被查询列a2上存在唯一索引，所以可以利用索引求最值，所以mysql支持“>all”式的子查询优化，子查询只被执行一次即可求得最大值。

 

不相关的all子查询，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 =all (select a2 from t2 where t2.a2=10);

+----+--------------------+-------+------+------+-------------+

| id | select_type    | table | type | key | extra    |

+----+--------------------+-------+------+------+-------------+

| 1 | primary      | t1  | all | null | using where |

| 2 | dependent subquery | t2  | all | null | using where |

+----+--------------------+-------+------+------+-------------+

2 rows in set, 1 warning (0.00 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,`test`.`t1`.`b1` as `b1`

from `test`.`t1`

where <not>(<in_optimizer>(

  `test`.`t1`.`a1`,<exists>(

    /* select#2 */ select 1 from `test`.`t2`

    where ((`test`.`t2`.`a2` = 10) and

      <if>(outer_field_is_not_null,

        ((<cache>(`test`.`t1`.`a1`) <> 10) or <cache>(isnull(10))),

        true

      )

    )

    having <if>(outer_field_is_not_null, <is_not_null_test>(`test`.`t2`.`a2`), true)

  )

))

从查询执行计划看，出现了子查询（subquery），但是被查询优化器处理后的语句中包含“exists”，这表明mysql对于“=all”式的子查询优化用“exists strategy”方式优化，所以mysql支持“=all”式的子查询优化。

 

不相关的all子查询，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 <all (select a2 from t2 where t2.a2=10);

+----+-------------+-------+------+------+-------------+

| id | select_type | table | type | key | extra    |

+----+-------------+-------+------+------+-------------+

| 1 | primary   | t1  | all | null | using where |

| 2 | subquery  | t2  | all | null | using where |

+----+-------------+-------+------+------+-------------+

2 rows in set, 1 warning (0.00 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,`test`.`t1`.`b1` as `b1`

from `test`.`t1`

where <not>((`test`.`t1`.`a1` >= <min>

  (/* select#2 */

    select `test`.`t2`.`a2`

    from `test`.`t2`

    where (`test`.`t2`.`a2` = 10)

  )

))

从查询执行计划看，出现了子查询（subquery），但是，子查询被“>= <min>”操作符限制，而子查询中的被查询列a2上存在唯一索引，所以可以利用索引求最值，所以mysql支持“<all”式的子查询优化，子查询只被执行一次即可求得最小值。

 

示例六，mysql支持对some类型的子查询的优化：

使用了“>some”式子的子查询被优化，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 >some (select a2 from t2 where t2.a2>10);

+----+-------------+-------+------+------+-------------+

| id | select_type | table | type | key | extra    |

+----+-------------+-------+------+------+-------------+

| 1 | primary   | t1  | all | null | using where |

| 2 | subquery  | t2  | all | null | using where |

+----+-------------+-------+------+------+-------------+

2 rows in set, 1 warning (0.05 sec)

被查询优化器处理后的语句为：

 /* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

   `test`.`t1`.`b1` as `b1`

from `test`.`t1`

where <nop>((`test`.`t1`.`a1` > (

  /* select#2 */

  select min(`test`.`t2`.`a2`)

  from `test`.`t2`

  where (`test`.`t2`.`a2` > 10)

)))

从查询执行计划看，出现了子查询（subquery），但是，子查询被“min”函数限制，而子查询中的被查询列a2上存在唯一索引，所以可以利用索引求最值，所以mysql支持“>some”式的子查询优化，子查询只被执行一次即可求得最大值。

 

使用了“=some”式子的子查询被优化，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 =some (select a2 from t2 where t2.a2=10);

+----+--------------+-------------+------+------+----------------------------------------------------+

| id | select_type | table    | type | key | extra  |

+----+--------------+-------------+------+------+----------------------------------------------------+

| 1 | simple    | <subquery2> | all | null | using where  |

| 1 | simple    | t1     | all | null | using where; using join buffer (block nested loop) |

| 2 | materialized | t2     | all | null | using where  |

+----+--------------+-------------+------+------+----------------------------------------------------+

3 rows in set, 1 warning (0.01 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

`test`.`t1`.`b1` as `b1`

from `test`.`t1` semi join (`test`.`t2`)

where ((`<subquery2>`.`a2` = 10) and (`test`.`t1`.`a1` = 10) and (`test`.`t2`.`a2` = 10))

从查询执行计划看，没有出现了子查询，表t2被物化，与表t1进行了半连接。

 

使用了“<some”式子的子查询被优化，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 <some (select a2 from t2 where t2.a2=10);

+----+-------------+-------+------+------+-------------+

| id | select_type | table | type | key | extra    |

+----+-------------+-------+------+------+-------------+

| 1 | primary   | t1  | all | null | using where |

| 2 | subquery  | t2  | all | null | using where |

+----+-------------+-------+------+------+-------------+

2 rows in set, 1 warning (0.00 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

  `test`.`t1`.`b1` as `b1`

from `test`.`t1`

where <nop>(

  (

    `test`.`t1`.`a1` < (/* select#2 */

      select max(`test`.`t2`.`a2`)

      from `test`.`t2`

      where (`test`.`t2`.`a2` = 10)

    )

  )

)

从查询执行计划看，出现了子查询（subquery），但是，子查询被“max”函数限制，而子查询中的被查询列a2上存在唯一索引，所以可以利用索引求最值，所以mysql支持“<some”式的子查询优化，子查询只被执行一次即可求得最大值。

 

示例七，mysql支持对any类型的子查询的优化：

使用了“>any”式子的子查询被优化，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 >any (select a2 from t2 where t2.a2>10);

+----+-------------+-------+------+------+-------------+

| id | select_type | table | type | key | extra    |

+----+-------------+-------+------+------+-------------+

| 1 | primary   | t1  | all | null | using where |

| 2 | subquery  | t2  | all | null | using where |

+----+-------------+-------+------+------+-------------+

2 rows in set, 1 warning (0.00 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

  `test`.`t1`.`b1` as `b1`

from `test`.`t1`

where <nop>(

  (

    `test`.`t1`.`a1` > (/* select#2 */

      select min(`test`.`t2`.`a2`)

      from `test`.`t2`

      where (`test`.`t2`.`a2` > 10)

    )

  )

)

从查询执行计划看，出现了子查询（subquery），但是，子查询被“min”函数限制，而子查询中的被查询列a2上存在唯一索引，所以可以利用索引求最值，所以mysql支持“>any”式的子查询优化，子查询只被执行一次即可求得最小值。

 

使用了“=any”式子的子查询被优化，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 =any (select a2 from t2 where t2.a2>10);

+----+--------------+-------------+------+------+----------------------------------------------------+

| id | select_type | table    | type | key | extra  |

+----+--------------+-------------+------+------+----------------------------------------------------+

| 1 | simple    | <subquery2> | all | null | null  |

| 1 | simple    | t1     | all | null | using where; using join buffer (block nested loop) |

| 2 | materialized | t2     | all | null | using where  |

+----+--------------+-------------+------+------+----------------------------------------------------+

3 rows in set, 1 warning (0.02 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

  `test`.`t1`.`b1` as `b1`

from `test`.`t1` semi join (`test`.`t2`)

where ((`test`.`t1`.`a1` = `<subquery2>`.`a2`) and (`test`.`t2`.`a2` > 10))

从查询执行计划看，没有出现了子查询，表t2被物化，与表t1进行了半连接。

 

使用了“<any”式子的子查询被优化，查询执行计划如下：

mysql> explain extended select * from t1 where t1.a1 <any (select a2 from t2 where t2.a2>10);

+----+-------------+-------+------+------+-------------+

| id | select_type | table | type | key | extra    |

+----+-------------+-------+------+------+-------------+

| 1 | primary   | t1  | all | null | using where |

| 2 | subquery  | t2  | all | null | using where |

+----+-------------+-------+------+------+-------------+

2 rows in set, 1 warning (0.00 sec)

被查询优化器处理后的语句为：

/* select#1 */ select `test`.`t1`.`id1` as `id1`,`test`.`t1`.`a1` as `a1`,

  `test`.`t1`.`b1` as `b1`

from `test`.`t1`

where <nop>(

  (

    `test`.`t1`.`a1` < (/* select#2 */

      select max(`test`.`t2`.`a2`)

      from `test`.`t2`

      where (`test`.`t2`.`a2` > 10)

    )

  )

)

从查询执行计划看，出现了子查询（subquery），但是，子查询被“max”函数限制，而子查询中的被查询列a2上存在唯一索引，所以可以利用索引求最值，所以mysql支持“<any”式的子查询优化，子查询只被执行一次即可求得最大值。