MySQL的JOIN（四）：JOIN优化实践之快速匹配

这篇博文讲述如何优化扫描速度。我们通过MySQL的JOIN（二）：JOIN原理得知了两张表的JOIN操作就是不断从驱动表中取出记录，然后查找出被驱动表中与之匹配的记录并连接。这个过程的实质就是查询操作，想要优化查询操作，建索引是最常用的方式。那索引怎么建呢？我们来讨论下，首先插入测试数据。

    CREATE TABLE t1 (
        id INT PRIMARY KEY AUTO_INCREMENT,
        type INT
    );
    SELECT COUNT(*) FROM t1;
    +----------+
    | COUNT(*) |
    +----------+
    |   110000 |
    +----------+
    CREATE TABLE t2 (
        id INT PRIMARY KEY AUTO_INCREMENT,
        type INT
    );
    SELECT COUNT(*) FROM t2;
    +----------+
    | COUNT(*) |
    +----------+
    |      100 |
    +----------+

左连接

左连接中，左表是驱动表，右表是被驱动表。想要快速查找被驱动表中匹配的记录，所以我们可以在右表建索引，从而提高连接性能。

    -- 首先两个表都没建索引
    EXPLAIN SELECT * FROM t1 LEFT JOIN t2 ON t1.type=t2.type;
    +----+-------+------+------+--------+----------------------------------------------------+
    | id | table | type | key  | rows   | Extra                                              |
    +----+-------+------+------+--------+----------------------------------------------------+
    |  1 | t1    | ALL  | NULL | 110428 | NULL                                               |
    |  1 | t2    | ALL  | NULL |    100 | Using where; Using join buffer (Block Nested Loop) |
    +----+-------+------+------+--------+----------------------------------------------------+
    -- 尝试在左表建立索引，改进不大
    CREATE INDEX idx_type ON t1(type);
    EXPLAIN SELECT * FROM t1 LEFT JOIN t2 ON t1.type=t2.type;
    +----+-------+-------+----------+--------+----------------------------------------------------+
    | id | table | type  | key      | rows   | Extra                                              |
    +----+-------+-------+----------+--------+----------------------------------------------------+
    |  1 | t1    | index | idx_type | 110428 | Using index                                        |
    |  1 | t2    | ALL   | NULL     |    100 | Using where; Using join buffer (Block Nested Loop) |
    +----+-------+-------+----------+--------+----------------------------------------------------+

    -- 尝试在右表建立索引，效果拔群，Using index!!!
    DROP INDEX idx_type ON t1;
    CREATE INDEX idx_type ON t2(type);
    EXPLAIN SELECT * FROM t1 LEFT JOIN t2 ON t1.type=t2.type;
    +----+-------+------+---------------+----------+--------+-------------+
    | id | table | type | possible_keys | key      | rows   | Extra       |
    +----+-------+------+---------------+----------+--------+-------------+
    |  1 | t1    | ALL  | NULL          | NULL     | 110428 | NULL        |
    |  1 | t2    | ref  | idx_type      | idx_type |      1 | Using index |
    +----+-------+------+---------------+----------+--------+-------------+

右连接

右连接中，右表是驱动表，左表是被驱动表，想要快速查找被驱动表中匹配的记录，所以我们可以在左表建索引，从而提高连接性能。

    DROP INDEX idx_type ON t2;
    -- 两个表都没有索引
    EXPLAIN SELECT * FROM t1 RIGHT JOIN t2 ON t1.type=t2.type;
    +----+-------+------+------+--------+----------------------------------------------------+
    | id | table | type | key  | rows   | Extra                                              |
    +----+-------+------+------+--------+----------------------------------------------------+
    |  1 | t2    | ALL  | NULL |    100 | NULL                                               |
    |  1 | t1    | ALL  | NULL | 110428 | Using where; Using join buffer (Block Nested Loop) |
    +----+-------+------+------+--------+----------------------------------------------------+

    -- 在右边建立索引，改进不大
    CREATE INDEX idx_type ON t2(type);
    EXPLAIN SELECT * FROM t1 RIGHT JOIN t2 ON t1.type=t2.type;
    +----+-------+-------+---------------+----------+--------+----------------------------------------------------+
    | id | table | type  | possible_keys | key      | rows   | Extra                                              |
    +----+-------+-------+---------------+----------+--------+----------------------------------------------------+
    |  1 | t2    | index | NULL          | idx_type |    100 | Using index                                        |
    |  1 | t1    | ALL   | NULL          | NULL     | 110428 | Using where; Using join buffer (Block Nested Loop) |
    +----+-------+-------+---------------+----------+--------+----------------------------------------------------+

    -- 尝试在左边建立索引，效果拔群！
    DROP INDEX idx_type ON t2;
    CREATE INDEX idx_type ON t1(type);
    EXPLAIN SELECT * FROM t1 RIGHT JOIN t2 ON t1.type=t2.type;
    +----+-------+------+---------------+--------------+------+-------------+
    | id | table | type | possible_keys | ref          | rows | Extra       |
    +----+-------+------+---------------+--------------+------+-------------+
    |  1 | t2    | ALL  | NULL          | NULL         |  100 | NULL        |
    |  1 | t1    | ref  | idx_type      | test.t2.type |    5 | Using index |
    +----+-------+------+---------------+--------------+------+-------------+

内连接

我们知道，MySQL Optimizer会对内连接做优化，不管谁内连接谁，都是用小表驱动大表，所以如果要优化内连接，可以在大表上建立索引，以提高连接性能。

另外注意一点，在小表上建立索引时，MySQL Optimizer会认为用大表驱动小表效率更快，转而用大表驱动小表。

对内连接小表驱动大表的优化策略不清楚的话，可以看MySQL的JOIN（三）：JOIN优化实践之内循环的次数

    DROP INDEX idx_type ON t1;
    -- 两个表都没有索引，t2驱动t1
    EXPLAIN SELECT * FROM t1 INNER JOIN t2 ON t1.type=t2.type;
    +----+-------+------+------+--------+----------------------------------------------------+
    | id | table | type | key  | rows   | Extra                                              |
    +----+-------+------+------+--------+----------------------------------------------------+
    |  1 | t2    | ALL  | NULL |    100 | NULL                                               |
    |  1 | t1    | ALL  | NULL | 110428 | Using where; Using join buffer (Block Nested Loop) |
    +----+-------+------+------+--------+----------------------------------------------------+
    -- 在t2表上建立索引，MySQL的Optimizer发现后，用大表驱动了小表
    CREATE INDEX idx_type ON t2(type);
    EXPLAIN SELECT * FROM t1 INNER JOIN t2 ON t1.type=t2.type;
    +----+-------+------+----------+--------+-------------+
    | id | table | type | key      | rows   | Extra       |
    +----+-------+------+----------+--------+-------------+
    |  1 | t1    | ALL  | NULL     | 110428 | Using where |
    |  1 | t2    | ref  | idx_type |      1 | Using index |
    +----+-------+------+----------+--------+-------------+

    -- 在t1表上建立索引，再加上t1是大表，符合“小表驱动大表”的原则，性能比上面的语句要好
    DROP INDEX idx_type ON t2;
    CREATE INDEX idx_type ON t1(type);
    EXPLAIN SELECT * FROM t1 INNER JOIN t2 ON t1.type=t2.type;
    +----+-------+------+---------------+----------+------+-------------+
    | id | table | type | possible_keys | key      | rows | Extra       |
    +----+-------+------+---------------+----------+------+-------------+
    |  1 | t2    | ALL  | NULL          | NULL     |  100 | Using where |
    |  1 | t1    | ref  | idx_type      | idx_type |    5 | Using index |
    +----+-------+------+---------------+----------+------+-------------+

三表连接

上面都是两表连接，三表连接也是一样的，找出驱动表和被驱动表，在被驱动表上建立索引，即可提高连接性能。

总结

想要从快速匹配的角度优化JOIN，首先就是找出谁是驱动表，谁是被驱动表，然后在被驱动表上建立索引即可。