通常,我们会采用ORDER BY LIMIT start, offset 的方式来进行分页查询。例如下面这个SQL:
SELECT * FROM `t1` WHERE ftype=1 ORDER BY id DESC LIMIT 100, 10;
或者像下面这个不带任何条件的分页SQL:
SELECT * FROM `t1` ORDER BY id DESC LIMIT 100, 10;
一般而言,分页SQL的耗时随着 start 值的增加而急剧增加,我们来看下面这2个不同起始值的分页SQL执行耗时:
[email protected]> SELECT * FROM `t1` WHERE ftype=1 ORDER BY id DESC LIMIT 500, 10; … 10 rows in set (0.05 sec) [email protected]> SELECT * FROM `t1` WHERE ftype=6 ORDER BY id DESC LIMIT 935500, 10; … 10 rows in set (2.39 sec)
可以看到,随着分页数量的增加,SQL查询耗时也有数十倍增加,显然不科学。今天我们就来分析下,如何能优化这个分页方案。 一般滴,想要优化分页的终极方案就是:没有分页,哈哈哈~~~,不要说我讲废话,确实如此,可以把分页算法交给Sphinx、Lucence等第三方解决方案,没必要让MySQL来做它不擅长的事情。 当然了,有小伙伴说,用第三方太麻烦了,我们就想用MySQL来做这个分页,咋办呢?莫急,且待我们慢慢分析,先看下表DDL、数据量、查询SQL的执行计划等信息:
[email protected]> SHOW CREATE TABLE `t1`; CREATE TABLE `t1` ( `id` int(10) unsigned NOT NULL AUTO_INCREMENT, ... `ftype` tinyint(3) unsigned NOT NULL, ... PRIMARY KEY (`id`) ) ENGINE=InnoDB DEFAULT CHARSET=utf8; [email protected]> select count(*) from t1; +----------+ | count(*) | +----------+ | 994584 | +----------+ [email protected]> EXPLAIN SELECT * FROM `t1` WHERE ftype=1 ORDER BY id DESC LIMIT 500, 10\G *************************** 1. row *************************** id: 1 select_type: SIMPLE table: t1 type: index possible_keys: NULL key: PRIMARY key_len: 4 ref: NULL rows: 510 Extra: Using where [email protected]> EXPLAIN SELECT * FROM `t1` WHERE ftype=1 ORDER BY id DESC LIMIT 935500, 10\G *************************** 1. row *************************** id: 1 select_type: SIMPLE table: t1 type: index possible_keys: NULL key: PRIMARY key_len: 4 ref: NULL rows: 935510 Extra: Using where
可以看到,虽然通过主键索引进行扫描了,但第二个SQL需要扫描的记录数太大了,而且需要先扫描约935510条记录,然后再根据排序结果取10条记录,这肯定是非常慢了。 针对这种情况,我们的优化思路就比较清晰了,有两点:
1、尽可能从索引中直接获取数据,避免或减少直接扫描行数据的频率 2、尽可能减少扫描的记录数,也就是先确定起始的范围,再往后取N条记录即可
据此,我们有两种相应的改写方法:子查询、表连接,即下面这样的:
#采用子查询的方式优化,在子查询里先从索引获取到最大id,然后倒序排,再取10行结果集 #注意这里采用了2次倒序排,因此在取LIMIT的start值时,比原来的值加了10,即935510,否则结果将和原来的不一致 [email protected]> EXPLAIN SELECT * FROM (SELECT * FROM `t1` WHERE id > ( SELECT id FROM `t1` WHERE ftype=1 ORDER BY id DESC LIMIT 935510, 1) LIMIT 10) t ORDER BY id DESC\G *************************** 1. row *************************** id: 1 select_type: PRIMARY table: <derived2> type: ALL possible_keys: NULL key: NULL key_len: NULL ref: NULL rows: 10 Extra: Using filesort *************************** 2. row *************************** id: 2 select_type: DERIVED table: t1 type: ALL possible_keys: PRIMARY key: NULL key_len: NULL ref: NULL rows: 973192 Extra: Using where *************************** 3. row *************************** id: 3 select_type: SUBQUERY table: t1 type: index possible_keys: NULL key: PRIMARY key_len: 4 ref: NULL rows: 935511 Extra: Using where #采用INNER JOIN优化,JOIN子句里也优先从索引获取ID列表,然后直接关联查询获得最终结果,这里不需要加10 [email protected]> EXPLAIN SELECT * FROM `t1` INNER JOIN ( SELECT id FROM `t1` WHERE ftype=1 ORDER BY id DESC LIMIT 935500,10) t2 USING (id)\G *************************** 1. row *************************** id: 1 select_type: PRIMARY table: <derived2> type: ALL possible_keys: NULL key: NULL key_len: NULL ref: NULL rows: 935510 Extra: NULL *************************** 2. row *************************** id: 1 select_type: PRIMARY table: t1 type: eq_ref possible_keys: PRIMARY key: PRIMARY key_len: 4 ref: t2.id rows: 1 Extra: NULL *************************** 3. row *************************** id: 2 select_type: DERIVED table: t1 type: index possible_keys: NULL key: PRIMARY key_len: 4 ref: NULL rows: 973192 Extra: Using where
然后我们来对比下这2个优化后的新SQL执行时间:
[email protected]> SELECT * FROM (SELECT * FROM `t1` WHERE id > ( SELECT id FROM `t1` WHERE ftype=1 ORDER BY id DESC LIMIT 935510, 1) LIMIT 10) T ORDER BY id DESC; ... rows in set (1.86 sec) #采用子查询优化,从profiling的结果来看,相比原来的那个SQL快了:28.2% [email protected]> SELECT * FROM `t1` INNER JOIN ( SELECT id FROM `t1` WHERE ftype=1 ORDER BY id DESC LIMIT 935500,10) t2 USING (id); ... 10 rows in set (1.83 sec) #采用INNER JOIN优化,从profiling的结果来看,相比原来的那个SQL快了:30.8%
我们再来看一个不带过滤条件的分页SQL对比:
#原始SQL [email protected]> EXPLAIN SELECT * FROM `t1` ORDER BY id DESC LIMIT 935500, 10\G *************************** 1. row *************************** id: 1 select_type: SIMPLE table: t1 type: index possible_keys: NULL key: PRIMARY key_len: 4 ref: NULL rows: 935510 Extra: NULL [email protected]> SELECT * FROM `t1` ORDER BY id DESC LIMIT 935500, 10; ... 10 rows in set (2.22 sec) #采用子查询优化 [email protected]> EXPLAIN SELECT * FROM (SELECT * FROM `t1` WHERE id > ( SELECT id FROM `t1` ORDER BY id DESC LIMIT 935510, 1) LIMIT 10) t ORDER BY id DESC; *************************** 1. row *************************** id: 1 select_type: PRIMARY table: <derived2> type: ALL possible_keys: NULL key: NULL key_len: NULL ref: NULL rows: 10 Extra: Using filesort *************************** 2. row *************************** id: 2 select_type: DERIVED table: t1 type: ALL possible_keys: PRIMARY key: NULL key_len: NULL ref: NULL rows: 973192 Extra: Using where *************************** 3. row *************************** id: 3 select_type: SUBQUERY table: t1 type: index possible_keys: NULL key: PRIMARY key_len: 4 ref: NULL rows: 935511 Extra: Using index [email protected]> SELECT * FROM (SELECT * FROM `t1` WHERE id > ( SELECT id FROM `t1` ORDER BY id DESC LIMIT 935510, 1) LIMIT 10) t ORDER BY id DESC; … 10 rows in set (2.01 sec) #采用子查询优化,从profiling的结果来看,相比原来的那个SQL快了:10.6% #采用INNER JOIN优化 [email protected]> EXPLAIN SELECT * FROM `t1` INNER JOIN ( SELECT id FROM `t1`ORDER BY id DESC LIMIT 935500,10) t2 USING (id)\G *************************** 1. row *************************** id: 1 select_type: PRIMARY table: type: ALL possible_keys: NULL key: NULL key_len: NULL ref: NULL rows: 935510 Extra: NULL *************************** 2. row *************************** id: 1 select_type: PRIMARY table: t1 type: eq_ref possible_keys: PRIMARY key: PRIMARY key_len: 4 ref: t1.id rows: 1 Extra: NULL *************************** 3. row *************************** id: 2 select_type: DERIVED table: t1 type: index possible_keys: NULL key: PRIMARY key_len: 4 ref: NULL rows: 973192 Extra: Using index [email protected]> SELECT * FROM `t1` INNER JOIN ( SELECT id FROM `t1`ORDER BY id DESC LIMIT 935500,10) t2 USING (id); … 10 rows in set (1.70 sec) #采用INNER JOIN优化,从profiling的结果来看,相比原来的那个SQL快了:30.2%
至此,我们看到采用子查询或者INNER JOIN进行优化后,都有大幅度的提升,这个方法也同样适用于较小的分页,虽然LIMIT开始的 start 位置小了很多,SQL执行时间也快了很多,但采用这种方法后,带WHERE条件的分页分别能提高查询效率:24.9%、156.5%,不带WHERE条件的分页分别提高查询效率:554.5%、11.7%,各位可以自行进行测试验证。单从提升比例说,还是挺可观的,确保这些优化方法可以适用于各种分页模式,就可以从一开始就是用。 我们来看下各种场景相应的提升比例是多少:
大分页,带WHERE | 大分页,不带WHERE | 大分页平均提升比例 | 小分页,带WHERE | 小分页,不带WHERE | 总体平均提升比例 | |
子查询优化 | 28.20% | 10.60% | 19.40% | 24.90% | 554.40% | 154.53% |
INNER JOIN优化 | 30.80% | 30.20% | 30.50% | 156.50% | 11.70% | 57.30% |
结论:这样看就和明显了,尤其是针对大分页的情况,因此我们优先推荐使用INNER JOIN方式优化分页算法。
上述每次测试都重启mysqld实例,并且加了SQL_NO_CACHE,以保证每次都是直接数据文件或索引文件中读取。如果数据经过预热后,查询效率会一定程度提升,但但上述相应的效率提升比例还是基本一致的。
2014/07/28后记更新:
其实如果是不带任何条件的分页,就没必要用这么麻烦的方法了,可以采用对主键采用范围检索的方法,例如参考这篇: Advance for MySQL Pagination
转自: http://imysql.com/2014/07/26/mysql-optimization-case-paging-optimize.shtml