SQL优化：使用explain

前文说了EXPLAIN的输出的含义，本文实战一下。

Database Schema

DROP DATABASE dbTest;
CREATE DATABASE dbTest;
USE dbTest;
CREATE TABLE t1
(
c_primary_key               INT,
c_unique_key                CHAR(64),
c_unique_not_null_key       CHAR(64) NOT NULL,
c_key                       CHAR(64),
c_multi_key_part1           CHAR(64),
c_multi_key_part2           CHAR(64),
c_int_value                 INT,
c_str_value                 CHAR(64),
PRIMARY KEY(c_primary_key),
UNIQUE KEY(c_unique_key),
UNIQUE KEY(c_unique_not_null_key),
KEY(c_multi_key_part1, c_multi_key_part2),
KEY(c_key)
)ENGINE=InnoDB;
CREATE TABLE t2
(
c_primary_key               INT,
c_unique_key                CHAR(64),
c_unique_not_null_key       CHAR(64) NOT NULL,
c_key                       CHAR(64),
c_multi_key_part1           CHAR(64),
c_multi_key_part2           CHAR(64),
c_int_value                 INT,
c_str_value                 CHAR(64),
c_t1_primary_key            INT,
PRIMARY KEY(c_primary_key),
UNIQUE KEY(c_unique_key),
UNIQUE KEY(c_unique_not_null_key),
KEY(c_multi_key_part1, c_multi_key_part2),
KEY(c_key),
UNIQUE KEY(c_t1_primary_key)
)ENGINE=InnoDB;

Join类型

const

1. 使用primary key查找一条记录，满足const条件。

   mysql> EXPLAIN SELECT * FROM dbTest.t1 WHERE c_primary_key=1;
   +----+-------------+-------+-------+---------------+---------+---------+-------+------+-------+
   | id | select_type | table | type  | possible_keys | key     | key_len | ref   | rows | Extra |
   +----+-------------+-------+-------+---------------+---------+---------+-------+------+-------+
   |  1 | SIMPLE      | t1    | const | PRIMARY       | PRIMARY | 4       | const |    1 | NULL  |
   +----+-------------+-------+-------+---------------+---------+---------+-------+------+-------+
   

2. 使用unique key查找一条非NULL记录，满足const条件。

   mysql> EXPLAIN SELECT * FROM dbTest.t1 WHERE c_unique_key='4cb15758c8e311e5b46f06af68695f49';
   +----+-------------+-------+-------+---------------+--------------+---------+-------+------+-------+
   | id | select_type | table | type  | possible_keys | key          | key_len | ref   | rows | Extra |
   +----+-------------+-------+-------+---------------+--------------+---------+-------+------+-------+
   |  1 | SIMPLE      | t1    | const | c_unique_key  | c_unique_key | 65      | const |    1 | NULL  |
   +----+-------------+-------+-------+---------------+--------------+---------+-------+------+-------+
   

3. 使用unique key查找NULL记录，这种情况下NULL记录可能有多条，所以不满足const条件。而是ref条件，ref意味着可能得到匹配的结果不唯一，即可能存在多条。那么对于unique key，可以为NULL，那么我们再来看：

   mysql> EXPLAIN SELECT * FROM dbTest.t1 WHERE c_unique_key is NULL;
   +----+-------------+-------+------+---------------+--------------+---------+-------+------+-----------------------+
   | id | select_type | table | type | possible_keys | key          | key_len | ref   | rows | Extra                 |
   +----+-------------+-------+------+---------------+--------------+---------+-------+------+-----------------------+
   |  1 | SIMPLE      | t1    | ref  | c_unique_key  | c_unique_key | 65      | const |    1 | Using index condition |
   +----+-------------+-------+------+---------------+--------------+---------+-------+------+-----------------------+
   

4. 使用非NULL的unique key来查询，跟primary key类似，都是唯一的，所以满足const条件。

   mysql> EXPLAIN SELECT * FROM dbTest.t1 WHERE c_unique_not_null_key='00047412c96511e5844906af68695f49' limit 1;
   +----+-------------+-------+-------+-----------------------+-----------------------+---------+-------+------+-------+
   | id | select_type | table | type  | possible_keys         | key                   | key_len | ref   | rows | Extra |
   +----+-------------+-------+-------+-----------------------+-----------------------+---------+-------+------+-------+
   |  1 | SIMPLE      | t1    | const | c_unique_not_null_key | c_unique_not_null_key | 64      | const |    1 | NULL  |
   +----+-------------+-------+-------+-----------------------+-----------------------+---------+-------+------+-------+
   

5. 使用非唯一的index列查询，可能存在多条记录，所以是ref而不是const。

   mysql> EXPLAIN SELECT * FROM dbTest.t1 WHERE c_key='4cb15758c8e311e5b46f06af68695f49';
   +----+-------------+-------+------+---------------+-------+---------+-------+------+-----------------------+
   | id | select_type | table | type | possible_keys | key   | key_len | ref   | rows | Extra                 |
   +----+-------------+-------+------+---------------+-------+---------+-------+------+-----------------------+
   |  1 | SIMPLE      | t1    | ref  | c_key         | c_key | 65      | const |    1 | Using index condition |
   +----+-------------+-------+------+---------------+-------+---------+-------+------+-----------------------+
   

   const至多有一条记录满足条件！

eq_ref

前文说，eq_ref类型对于之前表的每一个行组合，只从该表中读取一条记录。只有一条记录匹配要求，索引必须是primary key或者unique key（非NULL）。

1. 使用primary key进行表关联

   mysql> EXPLAIN SELECT * FROM dbTest.t1,dbTest.t2 WHERE t1.c_primary_key=t2.c_key;
   +----+-------------+-------+--------+---------------+---------+---------+-----------------+------+-------------+
   | id | select_type | table | type   | possible_keys | key     | key_len | ref             | rows | Extra       |
   +----+-------------+-------+--------+---------------+---------+---------+-----------------+------+-------------+
   |  1 | SIMPLE      | t2    | ALL    | c_key         | NULL    | NULL    | NULL            | 8042 | Using where |
   |  1 | SIMPLE      | t1    | eq_ref | PRIMARY       | PRIMARY | 4       | dbTest.t2.c_key |    1 | Using where |
   +----+-------------+-------+--------+---------------+---------+---------+-----------------+------+-------------+
   

   可以看到，对于t2表中的每一行，t1中都有唯一的一行（至多一行）进行匹配，所以最终匹配为eq_ref。

2. 使用NOT NULL的unique key（唯一的一行）进行关联

   mysql> EXPLAIN SELECT * FROM dbTest.t1,dbTest.t2 WHERE t1.c_unique_not_null_key=t2.c_unique_not_null_key;
   +----+-------------+-------+--------+-----------------------+-----------------------+---------+---------------------------------+------+-------+
   | id | select_type | table | type   | possible_keys         | key                   | key_len | ref                             | rows | Extra |
   +----+-------------+-------+--------+-----------------------+-----------------------+---------+---------------------------------+------+-------+
   |  1 | SIMPLE      | t1    | ALL    | c_unique_not_null_key | NULL                  | NULL    | NULL                            | 9307 | NULL  |
   |  1 | SIMPLE      | t2    | eq_ref | c_unique_not_null_key | c_unique_not_null_key | 64      | dbTest.t1.c_unique_not_null_key |    1 | NULL  |
   +----+-------------+-------+--------+-----------------------+-----------------------+---------+---------------------------------+------+-------+
   

   eq_ref至多有一条记录满足条件！

ref

对于之前表的每一个组合，匹配到索引值的所有记录将被读取。例如匹配那些左侧前缀的key（multi-part key），或者非primary key，或者非unique index（匹配值不是NULL），或者unique index（但是匹配值是NULL）。

1. 使用index列匹配多行记录

   mysql> EXPLAIN SELECT * FROM t1 WHERE t1.c_key='58a95fbac96511e5844906';
   +----+-------------+-------+------+---------------+-------+---------+-------+------+-----------------------+
   | id | select_type | table | type | possible_keys | key   | key_len | ref   | rows | Extra                 |
   +----+-------------+-------+------+---------------+-------+---------+-------+------+-----------------------+
   |  1 | SIMPLE      | t1    | ref  | c_key         | c_key | 65      | const |    1 | Using index condition |
   +----+-------------+-------+------+---------------+-------+---------+-------+------+-----------------------+
   

2. 多表关联时使用index列匹配多行记录

   mysql> EXPLAIN SELECT * FROM t1,t2 WHERE t1.c_unique_key=t2.c_key;
   +----+-------------+-------+------+---------------+-------+---------+------------------------+------+-------------+
   | id | select_type | table | type | possible_keys | key   | key_len | ref                    | rows | Extra       |
   +----+-------------+-------+------+---------------+-------+---------+------------------------+------+-------------+
   |  1 | SIMPLE      | t1    | ALL  | c_unique_key  | NULL  | NULL    | NULL                   | 9307 | Using where |
   |  1 | SIMPLE      | t2    | ref  | c_key         | c_key | 65      | dbTest.t1.c_unique_key |    1 | NULL        |
   +----+-------------+-------+------+---------------+-------+---------+------------------------+------+-------------+
   

3. 匹配左侧前缀（t1.c_multi_key_part1）的多行记录

   mysql> EXPLAIN SELECT * FROM t1,t2 WHERE t1.c_multi_key_part1=t2.c_key;
   +----+-------------+-------+------+-------------------+-------------------+---------+-----------------+------+-------------+
   | id | select_type | table | type | possible_keys     | key               | key_len | ref             | rows | Extra       |
   +----+-------------+-------+------+-------------------+-------------------+---------+-----------------+------+-------------+
   |  1 | SIMPLE      | t2    | ALL  | c_key             | NULL              | NULL    | NULL            | 4904 | Using where |
   |  1 | SIMPLE      | t1    | ref  | c_multi_key_part1 | c_multi_key_part1 | 65      | dbTest.t2.c_key |    1 | NULL        |
   +----+-------------+-------+------+-------------------+-------------------+---------+-----------------+------+-------------+
   

4. 使用unique index查找NULL的记录

   mysql> EXPLAIN SELECT * FROM t1 WHERE t1.c_unique_key is NULL;
   +----+-------------+-------+------+---------------+--------------+---------+-------+------+-----------------------+
   | id | select_type | table | type | possible_keys | key          | key_len | ref   | rows | Extra                 |
   +----+-------------+-------+------+---------------+--------------+---------+-------+------+-----------------------+
   |  1 | SIMPLE      | t1    | ref  | c_unique_key  | c_unique_key | 65      | const |    1 | Using index condition |
   +----+-------------+-------+------+---------------+--------------+---------+-------+------+-----------------------+
   

   ref可以匹配多行记录！

ref_or_null

从这个关键词可以看出，ref或者NULL，既在ref的基础上加上NULL的搜索。以下例子对应于ref的记录。

1. 使用index列匹配多行记录，或者NULL记录

   mysql> EXPLAIN SELECT * FROM t1 WHERE t1.c_key='58a95fbac96511e5844906' or t1.c_key is NULL;
   +----+-------------+-------+-------------+---------------+-------+---------+-------+------+-----------------------+
   | id | select_type | table | type        | possible_keys | key   | key_len | ref   | rows | Extra                 |
   +----+-------------+-------+-------------+---------------+-------+---------+-------+------+-----------------------+
   |  1 | SIMPLE      | t1    | ref_or_null | c_key         | c_key | 65      | const |    2 | Using index condition |
   +----+-------------+-------+-------------+---------------+-------+---------+-------+------+-----------------------+
   

2. 其余不再给输出，参照ref，SQL如下。

   mysql> EXPLAIN SELECT * FROM t1,t2 WHERE t1.c_unique_key=t2.c_key or t1.c_unique_key is NULL;
   mysql> EXPLAIN SELECT * FROM t1,t2 WHERE t1.c_multi_key_part1=t2.c_key or t1.c_multi_key_part1 is NULL;
   

   ref_or_null = ref + NULL记录，所以是多行

range

对于一个给定的range，使用index来获取记录。range可以使用=，<>，>，>=，BETWEEN，IN()等操作符。

1. BETWEEN操作符

   mysql> EXPLAIN SELECT * FROM t1 WHERE t1.c_primary_key BETWEEN 10 AND 20;
   +----+-------------+-------+-------+---------------+---------+---------+------+------+-------------+
   | id | select_type | table | type  | possible_keys | key     | key_len | ref  | rows | Extra       |
   +----+-------------+-------+-------+---------------+---------+---------+------+------+-------------+
   |  1 | SIMPLE      | t1    | range | PRIMARY       | PRIMARY | 4       | NULL |   10 | Using where |
   +----+-------------+-------+-------+---------------+---------+---------+------+------+-------------+
   

2. IN操作符

   mysql> EXPLAIN SELECT * FROM t1 WHERE t1.c_primary_key IN (10, 20);
   +----+-------------+-------+-------+---------------+---------+---------+------+------+-------------+
   | id | select_type | table | type  | possible_keys | key     | key_len | ref  | rows | Extra       |
   +----+-------------+-------+-------+---------------+---------+---------+------+------+-------------+
   |  1 | SIMPLE      | t1    | range | PRIMARY       | PRIMARY | 4       | NULL |    2 | Using where |
   +----+-------------+-------+-------+---------------+---------+---------+------+------+-------------+
   

等等，不再赘述。

range使用index，多条记录！

index

该类型跟ALL类型，但是不同之处在于搜索的是index数据，因为index数据比较小，所以效率肯定比ALL要高。分为两种情况：

1. 索引数据足够满足要求，即索引数据包括了查询的所有数据（在InnoDB下，索引数据数据存储的内容，包括 索引列+主键列，如下，因为c_key索引数据包括了c_key列值+c_primary_key列值，所以只需遍历索引即可）

   mysql> EXPLAIN SELECT c_primary_key,c_key FROM t1;
   +----+-------------+-------+-------+---------------+-------+---------+------+------+-------------+
   | id | select_type | table | type  | possible_keys | key   | key_len | ref  | rows | Extra       |
   +----+-------------+-------+-------+---------------+-------+---------+------+------+-------------+
   |  1 | SIMPLE      | t1    | index | NULL          | c_key | 65      | NULL | 4915 | Using index |
   +----+-------------+-------+-------+---------------+-------+---------+------+------+-------------+
   

2. 需要根据某个索引列的顺序进行查询，如下。第一条EXPLAIN从t1中select出所有的c_primary_key，这是不需要用到c_key（当然用c_key是也可以的）。但是第二条EXPLAIN由于需要按照c_key的进行排序（而c_key index就是有序的），所以只需要c_key index存储的顺序读取出来即可达到排序的功能，所以使用c_key就起到了排序的作用。

   mysql> EXPLAIN SELECT c_primary_key FROM t1;
   +----+-------------+-------+-------+---------------+-----------------------+---------+------+------+-------------+
   | id | select_type | table | type  | possible_keys | key                   | key_len | ref  | rows | Extra       |
   +----+-------------+-------+-------+---------------+-----------------------+---------+------+------+-------------+
   |  1 | SIMPLE      | t1    | index | NULL          | c_unique_not_null_key | 64      | NULL | 4915 | Using index |
   +----+-------------+-------+-------+---------------+-----------------------+---------+------+------+-------------+
   mysql> EXPLAIN SELECT c_primary_key FROM t1 order by c_key;
   +----+-------------+-------+-------+---------------+-------+---------+------+------+-------------+
   | id | select_type | table | type  | possible_keys | key   | key_len | ref  | rows | Extra       |
   +----+-------------+-------+-------+---------------+-------+---------+------+------+-------------+
   |  1 | SIMPLE      | t1    | index | NULL          | c_key | 65      | NULL | 4915 | Using index |
   +----+-------------+-------+-------+---------------+-------+---------+------+------+-------------+
   

   index这种情况也就比ALL要好一点，这种SQL需要重点review，以防带来灾难！

ALL

终于到了最差的情况，全表扫描。即没有合适的索引数据，所以只能一行一行的扫描数据了，沦落至此，可想而知效果极差。例如：

1. 查询符合条件的记录，如下，由于c_str_value列没有索引，导致只能进行全表扫描。优化方法：可以在c_str_value列上加上索引。

   mysql> EXPLAIN SELECT * FROM t1 where t1.c_str_value='1111';
   +----+-------------+-------+------+---------------+------+---------+------+------+-------------+
   | id | select_type | table | type | possible_keys | key  | key_len | ref  | rows | Extra       |
   +----+-------------+-------+------+---------------+------+---------+------+------+-------------+
   |  1 | SIMPLE      | t1    | ALL  | NULL          | NULL | NULL    | NULL | 4915 | Using where |
   +----+-------------+-------+------+---------------+------+---------+------+------+-------------+
   

2. 查询一个表的所有记录，如下。优化方法：在满足业务需求的情况下，把查询的所有列改成某几列，这样若是某个索引数据满足条件的话，可以不用遍历全表，而仅仅遍历索引数据即可。

   mysql> EXPLAIN SELECT * FROM t1;
   +----+-------------+-------+------+---------------+------+---------+------+------+-------+
   | id | select_type | table | type | possible_keys | key  | key_len | ref  | rows | Extra |
   +----+-------------+-------+------+---------------+------+---------+------+------+-------+
   |  1 | SIMPLE      | t1    | ALL  | NULL          | NULL | NULL    | NULL | 4915 | NULL  |
   +----+-------------+-------+------+---------------+------+---------+------+------+-------+
   

   最慢的查询，必须得review这些SQL，数据量大的情况下，必然带来灾难！

总结

通过以上，我们可以看到效率排序为： const < eq_ref < ref < ref_or_null(range) < index < ALL，通常index和ALL是需要重点注意的。让我们的嗅觉灵敏起来吧。:)

本着理论指导实践的原则，以上用实例对理论做了实践，难免出错，敬请指正。

遗留问题：

1. unique index如何保存NULL的索引？这个key允许多条NULL记录存在么？

2. primary index可以为NULL么？
3. index列中如何存储NULL数据？