作者：蒋乐兴 MySQL DBA，擅长 python 和 SQL，目前维护着 github 的两个开源项目：mysqltools 、dbmc 以及独立博客：https://www.sqlpy.com。 本文来源：原创投稿 *爱可生开源社区出品，原创内容未经授权不得随意使用，转载请联系小编并注明来源。
MRR 要解决的问题
MRR 是 MySQL 针对特定查询的一种优化手段。假设一个查询有二级索引可用，读完二级索引后要回表才能查到那些不在当前二级索引上的列值，由于二级索引上引用的主键值不一定是有序的，因此就有可能造成大量的随机 IO，如果回表前把主键值给它排一下序，那么在回表的时候就可以用顺序 IO 取代原本的随机 IO。
环境准备
为了实验我们要准备一下表结构和数据。
-- 创建表
mysql> show create table t;
+----------------------------------------------------------------------+
| Table | Create Table |
+----------------------------------------------------------------------+
| t | CREATE TABLE `t` (
`id` int NOT NULL AUTO_INCREMENT,
`i0` int NOT NULL,
`i1` int NOT NULL,
`i2` int NOT NULL,
`i3` int NOT NULL,
`c0` varchar(128) NOT NULL,
`c1` varchar(128) NOT NULL,
`f0` float NOT NULL,
`f1` float NOT NULL,
PRIMARY KEY (`id`),
KEY `idx_i0` (`i0`)
) ENGINE=InnoDB
+----------------------------------------------------------------------+
1 row in set (0.00 sec)
-- 造数据
mysql> select count(*) from t;
+----------+
| count(*) |
+----------+
| 1120000 |
+----------+
1 row in set (0.77 sec)
--
update t set i0 = id % 100;
MRR 的优化效果
- 有 MRR 优化（Using MRR）时 SQL 的耗时情况。
mysql> explain select i0,i3 from t where i0 between 1 and 2;
+----+-------------+-------+------------+-------+---------------+--------+---------+------+-------+----------+----------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+-------+---------------+--------+---------+------+-------+----------+----------------------------------+
| 1 | SIMPLE | t | NULL | range | idx_i0 | idx_i0 | 4 | NULL | 43968 | 100.00 | Using index condition; Using MRR |
+----+-------------+-------+------------+-------+---------------+--------+---------+------+-------+----------+----------------------------------+
1 row in set, 1 warning (0.00 sec)
mysql> select i0,i3 from t where i0 between 1 and 2;
22400 rows in set (0.80 sec)
- 关闭 MRR 优化。
set optimizer_switch = "index_merge=on,index_merge_union=on,index_merge_sort_union=on,index_merge_intersection=on,engine_condition_pushdown=on,index_condition_pushdown=on,mrr=off,mrr_cost_based=on,block_nested_loop=on,batched_key_access=off,materialization=on,semijoin=on,loosescan=on,firstmatch=on,duplicateweedout=on,subquery_materialization_cost_based=on,use_index_extensions=on,condition_fanout_filter=on,derived_merge=on,use_invisible_indexes=off,skip_scan=on,hash_join=on";
mysql> explain select i0,i3 from t where i0 between 1 and 2;
+----+-------------+-------+------------+-------+---------------+--------+---------+------+-------+----------+-----------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+-------+---------------+--------+---------+------+-------+----------+-----------------------+
| 1 | SIMPLE | t | NULL | range | idx_i0 | idx_i0 | 4 | NULL | 43968 | 100.00 | Using index condition |
+----+-------------+-------+------------+-------+---------------+--------+---------+------+-------+----------+-----------------------+
1 row in set, 1 warning (0.00 sec)
mysql> select i0,i3 from t where i0 between 1 and 2;
22400 rows in set (2.56 sec)
结论 就刚才的测试场景开启 MRR 优化可以得到 3 倍的性能提升。
MRR 的优化器参数调整
如果想关闭 MRR 优化的话，就要把优化器开关 mrr 设置为 off。
默认只有在优化器认为 MRR 可以带来优化的情况下才会走 MRR，如果你想不管什么时候能走 MRR 的都走 MRR 的话，你要把 mrr_cost_based 设置为 off，不过最好不要这么干，因为这确实是一个坑，MRR 不一定什么时候都好，全表扫描有时候会更加快，如果在这种场景下走 MRR 就完成了。
开启 MRR 关闭基于开销的优化。
-- mrr=on,mrr_cost_based=off
set optimizer_switch = "index_merge=on,index_merge_union=on,index_merge_sort_union=on,index_merge_intersection=on,engine_condition_pushdown=on,index_condition_pushdown=on,mrr=on,mrr_cost_based=off,block_nested_loop=on,batched_key_access=off,materialization=on,semijoin=on,loosescan=on,firstmatch=on,duplicateweedout=on,subquery_materialization_cost_based=on,use_index_extensions=on,condition_fanout_filter=on,derived_merge=on,use_invisible_indexes=off,skip_scan=on,hash_join=on";
mysql> explain select i0,i3 from t where i0 between 1 and 10;
+----+-------------+-------+------------+-------+---------------+--------+---------+------+--------+----------+----------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+-------+---------------+--------+---------+------+--------+----------+----------------------------------+
| 1 | SIMPLE | t | NULL | range | idx_i0 | idx_i0 | 4 | NULL | 218492 | 100.00 | Using index condition; Using MRR |
+----+-------------+-------+------------+-------+---------------+--------+---------+------+--------+----------+----------------------------------+
1 row in set, 1 warning (0.00 sec)
select i0,i3 from t where i0 between 1 and 10;
112000 rows in set (4.86 sec)
开启 MRR 开启基于开销的优化。
-- mrr=on,mrr_cost_based=on
set optimizer_switch = "index_merge=on,index_merge_union=on,index_merge_sort_union=on,index_merge_intersection=on,engine_condition_pushdown=on,index_condition_pushdown=on,mrr=on,mrr_cost_based=on,block_nested_loop=on,batched_key_access=off,materialization=on,semijoin=on,loosescan=on,firstmatch=on,duplicateweedout=on,subquery_materialization_cost_based=on,use_index_extensions=on,condition_fanout_filter=on,derived_merge=on,use_invisible_indexes=off,skip_scan=on,hash_join=on";
mysql> explain select i0,i3 from t where i0 between 1 and 10;
+----+-------------+-------+------------+------+---------------+------+---------+------+---------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+------+---------+------+---------+----------+-------------+
| 1 | SIMPLE | t | NULL | ALL | idx_i0 | NULL | NULL | NULL | 1121902 | 19.48 | Using where |
+----+-------------+-------+------------+------+---------------+------+---------+------+---------+----------+-------------+
1 row in set, 1 warning (0.00 sec)
mysql> select i0,i3 from t where i0 between 1 and 10;
112000 rows in set (1.52 sec)
可以看到当 mrr_cost_based = OFF 的情况下用时 4.86s，mrr_cost_based = ON 的情况下用时 1.52s，总的来说 mrr_cost_based 是非常关键的建议始终打开。
MRR 的参数优化
MRR 要把主键排个序，这样之后对磁盘的操作就是由顺序读代替之前的随机读。从资源的使用情况上来看就是让 CPU 和内存多做点事，来换磁盘的顺序读。然而排序是需要内存的，这块内存的大小就由参数 read_rnd_buffer_size 来控制。
read_rnd_buffer_size 太小无法启用 MRR 功能。
mysql> select @@read_rnd_buffer_size;
+------------------------+
| @@read_rnd_buffer_size |
+------------------------+
| 262144 |
+------------------------+
1 row in set (0.00 sec)
mysql> explain select i0,i3 from t where i0 between 1 and 12;
+----+-------------+-------+------------+------+---------------+------+---------+------+---------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+------+---------+------+---------+----------+-------------+
| 1 | SIMPLE | t | NULL | ALL | idx_i0 | NULL | NULL | NULL | 1121902 | 23.57 | Using where |
+----+-------------+-------+------------+------+---------------+------+---------+------+---------+----------+-------------+
1 row in set, 1 warning (0.00 sec)
放大 read_rnd_buffer_size 让 MySQL 有足够的资源用于 MRR 。
mysql> set read_rnd_buffer_size = 32 * 1024 * 1024;
Query OK, 0 rows affected (0.00 sec)
mysql> explain select i0,i3 from t where i0 between 1 and 12;
+----+-------------+-------+------------+-------+---------------+--------+---------+------+--------+----------+----------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+-------+---------------+--------+---------+------+--------+----------+----------------------------------+
| 1 | SIMPLE | t | NULL | range | idx_i0 | idx_i0 | 4 | NULL | 264436 | 100.00 | Using index condition; Using MRR |
+----+-------------+-------+------------+-------+---------------+--------+---------+------+--------+----------+----------------------------------+
1 row in set, 1 warning (0.00 sec)
