先看一下arena_match_index的表结构,大家注意表的索引结构
CREATE TABLE `arena_match_index` (
`tid` int(10) unsigned NOT NULL DEFAULT '0',
`mid` int(10) unsigned NOT NULL DEFAULT '0',
`group` int(10) unsigned NOT NULL DEFAULT '0',
`round` tinyint(3) unsigned NOT NULL DEFAULT '0',
`day` date NOT NULL DEFAULT '0000-00-00',
`begintime` datetime NOT NULL DEFAULT '0000-00-00 00:00:00',
UNIQUE KEY `tm` (`tid`,`mid`),
KEY `mid` (`mid`),
KEY `begintime` (`begintime`),
KEY `dg` (`day`,`group`),
KEY `td` (`tid`,`day`)
) ENGINE=MyISAM DEFAULT CHARSET=utf8
接着看下面的sql:
SELECT round FROM arena_match_index WHERE `day` = '2010-12-31' AND `group` = 18 AND `begintime` < '2010-12-31 12:14:28' order by begintime LIMIT 1;
mysql> explain SELECT round FROM arena_match_index WHERE `day` = '2010-12-31' AND `group` = 18 AND `begintime` < '2010-12-31 12:14:28' order by begintime LIMIT 1;
+----+-------------+-------------------+-------+---------------+-----------+---------+------+--------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------------------+-------+---------------+-----------+---------+------+--------+-------------+
| 1 | SIMPLE | arena_match_index | range | begintime,dg | begintime | 8 | NULL | 226480 | Using where |
+----+-------------+-------------------+-------+---------------+-----------+---------+------+--------+-------------+
explain的结果显示使用`begintime`索引要扫描22w条记录,这样的查询性能是非常糟糕的,实际的执行情况也是初次执行(还未有缓存数据时)时需要30秒以上的时间。
实际上这个查询使用`dg`联合索引的性能更好,因为同一天同一个小组内也就几十场比赛,因此应该优先使用`dg`索引定位到匹配的数据集合再进行排序,那么如何告诉mysql使用指定索引呢?使用use index语句:
mysql> explain SELECT round FROM arena_match_index use index (dg) WHERE `day` = '2010-12-31' AND `group` = 18 AND `begintime` < '2010-12-31 12:14:28' order by begintime LIMIT 1;
+----+-------------+-------------------+------+---------------+------+---------+-------------+------+-----------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------------------+------+---------------+------+---------+-------------+------+-----------------------------+
| 1 | SIMPLE | arena_match_index | ref | dg | dg | 7 | const,const | 757 | Using where; Using filesort |
+----+-------------+-------------------+------+---------------+------+---------+-------------+------+-----------------------------+
explain结果显示使用`dg`联合索引只需要扫描757条数据,性能直接提升了上百倍,实际的执行情况也是几乎立即就返回了查询结果。
在最初的查询语句中只要把order by begintime去掉,mysql就会使用`dg`索引了,再次印证了order by会影响mysql的索引选择策略!
mysql> explain SELECT round FROM arena_match_index WHERE `day` = '2010-12-31' AND `group` = 18 AND `begintime` < '2010-12-31 12:14:28' LIMIT 1;
+----+-------------+-------------------+------+---------------+------+---------+-------------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------------------+------+---------------+------+---------+-------------+------+-------------+
| 1 | SIMPLE | arena_match_index | ref | begintime,dg | dg | 7 | const,const | 717 | Using where |
+----+-------------+-------------------+------+---------------+------+---------+-------------+------+-------------+
通过上面的例子说mysql有时候也并不聪明,并非总能做出最优选择,还是需要我们开发者对它进行“调教”!