MYSQL --- 基于开销的执行计划探讨

今天在和网友Mablevi探讨连接查询使用索引的情况，在改变连接顺序后，用EXPLAIN查看执行计划，都是小表在前面，大表在后面。我们知道MYSQL执行连接查询的时候，使用的叫“内循环”的技术，即依次取出外层循环的表（驱动表）的记录，然后到内层循环表（被驱动表）去查找符合条件的记录。驱动表是依次循环，全表扫描的，被驱动表可以通过索引查找，提高效率。

 

同时我们知道，MYSQL的执行计划是会去计算开销的，一个SQL查询可能会产生好几百种执行计划，根据开销最小原则，EXPLAIN出来的，就是经过挑选后产生的最优开销。下面我们来实际操作验证一下MYSQL的执行技术是否真的够“聪明”。

 

先看看表记录数：

mysql> select count(*) from terminal;
+----------+
| count(*) |
+----------+
| 1579 |
+----------+
1 row in set (0.06 sec)

mysql> select count(*) from area;
+----------+
| count(*) |
+----------+
| 3350 |
+----------+
1 row in set (0.09 sec)

mysql>

在看看索引情况

mysql> show index from terminal;
+----------+------------+---------------+--------------+-------------+----------
-+-------------+----------+--------+------+------------+---------+
| Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation
 | Cardinality | Sub_part | Packed | Null | Index_type | Comment |
+----------+------------+---------------+--------------+-------------+----------
-+-------------+----------+--------+------+------------+---------+
| terminal | 0 | PRIMARY | 1 | terminal_id | A
 | 1565 | NULL | NULL | | BTREE | |
| terminal | 1 | index_version | 1 | equ_version | A
 | 5 | NULL | NULL | YES | BTREE | |
+----------+------------+---------------+--------------+-------------+----------
-+-------------+----------+--------+------+------------+---------+
2 rows in set (0.08 sec)

mysql> show index from area;
+-------+------------+----------+--------------+-------------+-----------+------
-------+----------+--------+------+------------+---------+
| Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardi
nality | Sub_part | Packed | Null | Index_type | Comment |
+-------+------------+----------+--------------+-------------+-----------+------
-------+----------+--------+------+------------+---------+
| area | 0 | PRIMARY | 1 | area_id | A |
  3811 | NULL | NULL | | BTREE | |
+-------+------------+----------+--------------+-------------+-----------+------
-------+----------+--------+------+------------+---------+
1 row in set (0.09 sec)

mysql>

terminal的字段area_center_id是area表的外键。

1：正常情况下，用小表驱动大表查询看看执行的开销

mysql> select sql_no_cache count(*) from terminal inner join area on terminal.area_center_id=area.area_id;
+----------+
| count(*) |
+----------+
| 1577 |
+----------+
1 row in set (0.13 sec)

mysql> explain select sql_no_cache count(*) from terminal inner join area on terminal.area_center_id=area.area_id;
+----+-------------+----------+--------+---------------+---------+---------+-----------------------------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+----------+--------+---------------+---------+---------+-----------------------------+------+-------------+
| 1 | SIMPLE | terminal | ALL | NULL | NULL | NULL | NULL | 1565 | |
| 1 | SIMPLE | area | eq_ref | PRIMARY | PRIMARY | 4 | cpc.terminal.area_center_id | 1 | Using index |
+----+-------------+----------+--------+---------------+---------+---------+-----------------------------+------+-------------+
2 rows in set (0.09 sec)

mysql> show status like '%last%';
+-----------------+-------------+
| Variable_name | Value |
+-----------------+-------------+
| Last_query_cost | 1906.999000 |
+-----------------+-------------+
1 row in set (0.11 sec)

mysql>

很明显，首先查询的是TERMINAL表。在这个执行计划下，一共开销1906.999次随机IO读取。

现在换过来，用大表驱动小表，看看执行计划和开销

mysql> explain select sql_no_cache count(*) from area inner join terminal on terminal.area_center_id=area.area_id;
+----+-------------+----------+--------+---------------+---------+---------+-----------------------------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+----------+--------+---------------+---------+---------+-----------------------------+------+-------------+
| 1 | SIMPLE | terminal | ALL | NULL | NULL | NULL | NULL | 1565 | |
| 1 | SIMPLE | area | eq_ref | PRIMARY | PRIMARY | 4 | cpc.terminal.area_center_id | 1 | Using index |
+----+-------------+----------+--------+---------------+---------+---------+-----------------------------+------+-------------+
2 rows in set (0.47 sec)

执行计划还是自动选择先读小表，再读大表。

我们再强制执行计划安装查询语句定义的顺序读取表，看看开销

mysql> select sql_no_cache count(*) from area straight_join terminal on terminal.area_center_id=area.area_id;
+----------+
| count(*) |
+----------+
| 1577 |
+----------+
1 row in set (0.38 sec)

mysql> show status like '%last%';
+-----------------+----------------+
| Variable_name | Value |
+-----------------+----------------+
| Last_query_cost | 1192884.999000 |
+-----------------+----------------+
1 row in set (0.06 sec)


mysql> explain select sql_no_cache count(*) from area straight_join terminal on terminal.area_center_id=area.area_id;
+----+-------------+----------+-------+---------------+---------+---------+------+------+--------------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+----------+-------+---------------+---------+---------+------+------+--------------------------------+
| 1 | SIMPLE | area | index | PRIMARY | PRIMARY | 4 | NULL | 3811 | Using index |
| 1 | SIMPLE | terminal | ALL | NULL | NULL | NULL | NULL | 1565 | Using where; Using join buffer |
+----+-------------+----------+-------+---------------+---------+---------+------+------+--------------------------------+
2 rows in set (0.06 sec)

mysql>

执行开销猛增到1192884.999000。是原计划开销的625倍。

这也难怪MYSQL会无视我们的查询语句，而选择自己更优的执行计划了。