mysql查询时，offset过大影响性能的原因与优化方法
mysql查询使用select命令，配合limit，offset参数可以读取指定范围的记录。本文将介绍mysql查询时，offset过大影响性能的原因及优化方法。
准备测试数据表及数据
1.创建表
<span style="color:#000000"><code><span style="color:#000088 !important">CREATE</span> <span style="color:#000088 !important">TABLE</span> <span style="color:#009900 !important">`member`</span> (
<span style="color:#009900 !important">`id`</span> <span style="color:#000088 !important">int</span>(<span style="color:#006666 !important">10</span>) unsigned <span style="color:#000088 !important">NOT</span> <span style="color:#000088 !important">NULL</span> AUTO_INCREMENT,
<span style="color:#009900 !important">`name`</span> <span style="color:#000088 !important">varchar</span>(<span style="color:#006666 !important">10</span>) <span style="color:#000088 !important">NOT</span> <span style="color:#000088 !important">NULL</span> COMMENT <span style="color:#009900 !important">'姓名'</span>,
<span style="color:#009900 !important">`gender`</span> tinyint(<span style="color:#006666 !important">3</span>) unsigned <span style="color:#000088 !important">NOT</span> <span style="color:#000088 !important">NULL</span> COMMENT <span style="color:#009900 !important">'性别'</span>,
<span style="color:#000088 !important">PRIMARY</span> <span style="color:#000088 !important">KEY</span> (<span style="color:#009900 !important">`id`</span>),
<span style="color:#000088 !important">KEY</span> <span style="color:#009900 !important">`gender`</span> (<span style="color:#009900 !important">`gender`</span>)
) ENGINE=InnoDB <span style="color:#000088 !important">DEFAULT</span> CHARSET=utf8;</code></span>
2.插入1000000条记录
<span style="color:#000000"><code><?php
$pdo = new PDO("mysql:host=localhost;dbname=user","root",'<em>');
</em>for($i=0; $i<1000000; $i++){
$name = substr(md5(time().mt_rand(000,999)),0,10);
$gender = mt_rand(1,2);
$sqlstr = "insert into member(name,gender) values('".$name."','".$gender."')";
$stmt = $pdo->prepare($sqlstr);
$stmt->execute();
}
?>
<span style="color:#009900 !important">mysql> select count(*) from member;
+----------+</span>
<span style="color:#009900 !important">| count(*) |
+----------+</span>
<span style="color:#009900 !important">| 1000000 |
+----------+</span>
1 row in set (0.23 sec)</code></span>
3.当前数据库版本
<span style="color:#000000"><code><span style="color:#009900 !important">mysql> select version();
+-----------+</span>
<span style="color:#009900 !important">| version() |
+-----------+</span>
<span style="color:#009900 !important">| 5.6.24 |
+-----------+</span>
1 row in set (0.01 sec)</code></span>
分析offset过大影响性能的原因
1.offset较小的情况
<span style="color:#000000"><code><span style="color:#009900 !important">mysql> select * from member where gender=1 limit 10,1;
+----+------------+--------+</span>
<span style="color:#009900 !important">| id | name | gender |
+----+------------+--------+</span>
<span style="color:#009900 !important">| 26 | 509e279687 | 1 |
+----+------------+--------+</span>
1 row in set (0.00 sec)
<span style="color:#009900 !important">mysql> select * from member where gender=1 limit 100,1;
+-----+------------+--------+</span>
<span style="color:#009900 !important">| id | name | gender |
+-----+------------+--------+</span>
<span style="color:#009900 !important">| 211 | 07c4cbca3a | 1 |
+-----+------------+--------+</span>
1 row in set (0.00 sec)
<span style="color:#009900 !important">mysql> select * from member where gender=1 limit 1000,1;
+------+------------+--------+</span>
<span style="color:#009900 !important">| id | name | gender |
+------+------------+--------+</span>
<span style="color:#009900 !important">| 1975 | e95b8b6ca1 | 1 |
+------+------------+--------+</span>
1 row in set (0.00 sec)</code></span>当offset较小时，查询速度很快，效率较高。
2.offset较大的情况
<span style="color:#000000"><code><span style="color:#009900 !important">mysql> select * from member where gender=1 limit 100000,1;
+--------+------------+--------+</span>
<span style="color:#009900 !important">| id | name | gender |
+--------+------------+--------+</span>
<span style="color:#009900 !important">| 199798 | 540db8c5bc | 1 |
+--------+------------+--------+</span>
1 row in set (0.12 sec)
<span style="color:#009900 !important">mysql> select * from member where gender=1 limit 200000,1;
+--------+------------+--------+</span>
<span style="color:#009900 !important">| id | name | gender |
+--------+------------+--------+</span>
<span style="color:#009900 !important">| 399649 | 0b21fec4c6 | 1 |
+--------+------------+--------+</span>
1 row in set (0.23 sec)
<span style="color:#009900 !important">mysql> select * from member where gender=1 limit 300000,1;
+--------+------------+--------+</span>
<span style="color:#009900 !important">| id | name | gender |
+--------+------------+--------+</span>
<span style="color:#009900 !important">| 599465 | f48375bdb8 | 1 |
+--------+------------+--------+</span>
1 row in set (0.31 sec)</code></span>当offset很大时，会出现效率问题，随着offset的增大，执行效率下降。
分析影响性能原因
<span style="color:#000000"><code><span style="color:#000088 !important">select</span> * <span style="color:#000088 !important">from</span> member <span style="color:#000088 !important">where</span> gender=<span style="color:#006666 !important">1</span> limit <span style="color:#006666 !important">300000</span>,<span style="color:#006666 !important">1</span>;</code></span>- 1
因为数据表是InnoDB，根据InnoDB索引的结构，查询过程为：
- 通过二级索引查到主键值（找出所有gender=1的id)。
- 再根据查到的主键值通过主键索引找到相应的数据块（根据id找出对应的数据块内容）。
- 根据offset的值，查询300001次主键索引的数据，最后将之前的300000条丢弃，取出最后1条。
不过既然二级索引已经找到主键值，为什么还需要先用主键索引找到数据块，再根据offset的值做偏移处理呢？
如果在找到主键索引后，先执行offset偏移处理，跳过300000条，再通过第300001条记录的主键索引去读取数据块，这样就能提高效率了。
如果我们只查询出主键，看看有什么不同
<span style="color:#000000"><code><span style="color:#009900 !important">mysql> select id from member where gender=1 limit 300000,1;
+--------+</span>
<span style="color:#009900 !important">| id |
+--------+</span>
<span style="color:#009900 !important">| 599465 |
+--------+</span>
1 row in set (0.09 sec)</code></span>很明显，如果只查询主键，执行效率对比查询全部字段，有很大的提升。
推测
只查询主键的情况
因为二级索引已经找到主键值，而查询只需要读取主键，因此mysql会先执行offset偏移操作，再根据后面的主键索引读取数据块。
需要查询所有字段的情况
因为二级索引只找到主键值，但其他字段的值需要读取数据块才能获取。因此mysql会先读出数据块内容，再执行offset偏移操作，最后丢弃前面需要跳过的数据，返回后面的数据。
证实
InnoDB中有buffer pool，存放最近访问过的数据页，包括数据页和索引页。
为了测试，先把mysql重启，重启后查看buffer pool的内容。
<span style="color:#000000"><code>mysql<span style="color:#4f4f4f !important">></span> <span style="color:#000088 !important">select</span> index_name,count(<span style="color:#4f4f4f !important">*</span>) from information_schema<span style="color:#4f4f4f !important">.</span>INNODB_BUFFER_PAGE <span style="color:#000088 !important">where</span> INDEX_NAME <span style="color:#000088 !important">in</span>(<span style="color:#009900 !important">'primary'</span>,<span style="color:#009900 !important">'gender'</span>) <span style="color:#006666 !important">and</span> TABLE_NAME like <span style="color:#009900 !important">'%member%'</span> <span style="color:#000088 !important">group</span> <span style="color:#000088 !important">by</span> index_name;
Empty <span style="color:#4f4f4f !important">set</span> (<span style="color:#006666 !important">0.04</span> sec)</code></span>可以看到，重启后，没有访问过任何的数据页。
查询所有字段，再查看buffer pool的内容
<span style="color:#000000"><code><span style="color:#009900 !important">mysql> select * from member where gender=1 limit 300000,1;
+--------+------------+--------+</span>
<span style="color:#009900 !important">| id | name | gender |
+--------+------------+--------+</span>
<span style="color:#009900 !important">| 599465 | f48375bdb8 | 1 |
+--------+------------+--------+</span>
1 row in set (0.38 sec)
<span style="color:#009900 !important">mysql> select index_name,count(*) from information_schema.INNODB_BUFFER_PAGE where INDEX_NAME in('primary','gender') and TABLE_NAME like '%member%' group by index_name;
+------------+----------+</span>
<span style="color:#009900 !important">| index_name | count(*) |
+------------+----------+</span>
| gender | 261 |
<span style="color:#009900 !important">| PRIMARY | 1385 |
+------------+----------+</span>
2 rows in set (0.06 sec)</code></span>可以看出，此时buffer pool中关于member表有1385个数据页，261个索引页。
重启mysql清空buffer pool，继续测试只查询主键
<span style="color:#000000"><code><span style="color:#009900 !important">mysql> select id from member where gender=1 limit 300000,1;
+--------+</span>
<span style="color:#009900 !important">| id |
+--------+</span>
<span style="color:#009900 !important">| 599465 |
+--------+</span>
1 row in set (0.08 sec)
<span style="color:#009900 !important">mysql> select index_name,count(*) from information_schema.INNODB_BUFFER_PAGE where INDEX_NAME in('primary','gender') and TABLE_NAME like '%member%' group by index_name;
+------------+----------+</span>
<span style="color:#009900 !important">| index_name | count(*) |
+------------+----------+</span>
| gender | 263 |
<span style="color:#009900 !important">| PRIMARY | 13 |
+------------+----------+</span>
2 rows in set (0.04 sec)</code></span>可以看出，此时buffer pool中关于member表只有13个数据页，263个索引页。因此减少了多次通过主键索引访问数据块的I/O操作，提高执行效率。
因此可以证实，mysql查询时，offset过大影响性能的原因是多次通过主键索引访问数据块的I/O操作。（注意，只有InnoDB有这个问题，而MYISAM索引结构与InnoDB不同，二级索引都是直接指向数据块的，因此没有此问题 ）。
InnoDB与MyISAM引擎索引结构对比图
优化方法
根据上面的分析，我们知道查询所有字段会导致主键索引多次访问数据块造成的I/O操作。
因此我们先查出偏移后的主键，再根据主键索引查询数据块的所有内容即可优化。
<span style="color:#000000"><code><span style="color:#009900 !important">mysql> select a.* from member as a inner join (select id from member where gender=1 limit 300000,1) as b on a.id=b.id;
+--------+------------+--------+</span>
<span style="color:#009900 !important">| id | name | gender |
+--------+------------+--------+</span>
<span style="color:#009900 !important">| 599465 | f48375bdb8 | 1 |
+--------+------------+--------+</span>
1 row in set (0.08 sec)</code></span>