一、ICP( Index_Condition_Pushdown)

对 where 中过滤条件的处理，根据索引使用情况分成了三种：

如果WHERE条件可以使用索引，MySQL 会把这部分过滤操作放到存储引擎层，存储引擎通过索引过滤，把满足的行从表中读取出。ICP能减少Server层访问存储引擎的次数和引擎层访问基表的次数。

session级别设置：set optimizer_switch="index_condition_pushdown=on
对于InnoDB表，ICP只适用于辅助索引
当使用ICP优化时，执行计划的Extra列显示Using index condition提示
不支持主建索引的ICP(对于Innodb的聚集索引，完整的记录已经被读取到Innodb Buffer，此时使用ICP并不能降低IO操作)
当 SQL 使用覆盖索引时但只检索部分数据时，ICP 无法使用
ICP的加速效果取决于在存储引擎内通过ICP筛选掉的数据的比例

index_condition_pushdown会大大减少行锁的个数，如select for update，因为行锁是在引擎层的

例如：

现在的索引

show index from sm_performance_all;
+--------------------+------------+-------------------------------+--------------+----------------------+-----------+-------------+----------+--------+------+------------+---------+---------------+
| Table              | Non_unique | Key_name                      | Seq_in_index | Column_name          | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment | Index_comment |
+--------------------+------------+-------------------------------+--------------+----------------------+-----------+-------------+----------+--------+------+------------+---------+---------------+
| sm_performance_all |          0 | PRIMARY                       |            1 | id                   | A         |       40527 |     NULL | NULL   |      | BTREE      |         |               |
| sm_performance_all |          1 | FK_a9t29a4b2af1vfny1j2minc1x  |            1 | company_id           | A         |         316 |     NULL | NULL   | YES  | BTREE      |         |               |
| sm_performance_all |          1 | FK_n3ng4a5qju19fw8qy4uskp4g1  |            1 | bill_id              | A         |       21532 |     NULL | NULL   | YES  | BTREE      |         |               |
| sm_performance_all |          1 | FK_eb13u3xwslt9t7wwuycg7vha6  |            1 | car_id               | A         |       16794 |     NULL | NULL   | YES  | BTREE      |         |               |
| sm_performance_all |          1 | FK_2bfhskvklf6mdk557tc3yy3y1  |            1 | commission_entity_id | A         |         177 |     NULL | NULL   | YES  | BTREE      |         |               |
| sm_performance_all |          1 | FK_6fr5ib5iyjyu155dncmc48cwr  |            1 | member_card_id       | A         |          34 |     NULL | NULL   | YES  | BTREE      |         |               |
| sm_performance_all |          1 | FK_93p22vcog266wa82i44a6m18b  |            1 | user_id              | A         |         483 |     NULL | NULL   | YES  | BTREE      |         |               |
| sm_performance_all |          1 | FK_p6nc7l6ewnkcpm2y4o3wct81r  |            1 | member_card_bill_id  | A         |           4 |     NULL | NULL   | YES  | BTREE      |         |               |
| sm_performance_all |          1 | billId_userId_memberCarBillId |            1 | bill_id              | A         |       24194 |     NULL | NULL   | YES  | BTREE      |         |               |
| sm_performance_all |          1 | billId_userId_memberCarBillId |            2 | user_id              | A         |       25688 |     NULL | NULL   | YES  | BTREE      |         |               |
| sm_performance_all |          1 | billId_userId_memberCarBillId |            3 | member_card_bill_id  | A         |       25946 |     NULL | NULL   | YES  | BTREE      |         |               |
+--------------------+------------+-------------------------------+--------------+----------------------+-----------+-------------+----------+--------+------+------------+---------+---------------+
11 rows in set (0.00 sec)

现在的语句执行情况

explain select * from sm_performance_all p where p.date_created>'2018-01-01'  and p.date_created< '2018-02-01' and p.type=0;
+----+-------------+-------+------------+------+---------------+------+---------+------+-------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows  | filtered | Extra       |
+----+-------------+-------+------------+------+---------------+------+---------+------+-------+----------+-------------+
|  1 | SIMPLE      | p     | NULL       | ALL  | NULL          | NULL | NULL    | NULL | 40527 |     1.11 | Using where |
+----+-------------+-------+------------+------+---------------+------+---------+------+-------+----------+-------------+
1 row in set, 1 warning (0.00 sec)

添加索引后

ALTER TABLE sm_performance_all add index date_created_type(date_created, type );

explain select * from sm_performance_all p where p.date_created>'2018-01-01'  and p.date_created< '2018-02-01' and p.type=0;
+----+-------------+-------+------------+-------+-------------------+-------------------+---------+------+------+----------+-----------------------+
| id | select_type | table | partitions | type  | possible_keys     | key               | key_len | ref  | rows | filtered | Extra                 |
+----+-------------+-------+------------+-------+-------------------+-------------------+---------+------+------+----------+-----------------------+
|  1 | SIMPLE      | p     | NULL       | range | date_created_type | date_created_type | 6       | NULL |    1 |    10.00 | Using index condition |
+----+-------------+-------+------------+-------+-------------------+-------------------+---------+------+------+----------+-----------------------+
1 row in set, 1 warning (0.00 sec)

二、MRR(Multi-Range Read )

随机 IO 转化为顺序 IO 以降低查询过程中 IO 开销的一种手段，这对IO-bound类型的SQL语句性能带来极大的提升。

MRR can be used for InnoDB and MyISAM tables for index range scans and equi-join operations.

A portion of the index tuples are accumulated in a buffer.
The tuples in the buffer are sorted by their data row ID.
Data rows are accessed according to the sorted index tuple sequence.

上述的SQL语句需要根据辅助索引date_created_type进行查询，但是由于要求得到的是表中所有的列，因此需要回表进行读取。而这里就可能伴随着大量的随机I/O。这个过程如下图所示：

而MRR的优化在于，并不是每次通过辅助索引就回表去取记录，而是将其rowid给缓存起来，然后对rowid进行排序后，再去访问记录，这样就能将随机I/O转化为顺序I/O，从而大幅地提升性能。这个过程如下所示：

然而，在MySQL当前版本中，基于成本的算法过于保守，导致大部分情况下优化器都不会选择MRR特性。为了确保优化器使用mrr特性，请执行下面的SQL语句：

set optimizer_switch='mrr=on,mrr_cost_based=off';

读取全部字段时

explain select * from sm_performance_all p where p.date_created>'2018-01-01'  and p.date_created< '2018-02-01' and p.type=0;
+----+-------------+-------+------------+-------+-------------------+-------------------+---------+------+------+----------+----------------------------------+
| id | select_type | table | partitions | type  | possible_keys     | key               | key_len | ref  | rows | filtered | Extra                            |
+----+-------------+-------+------------+-------+-------------------+-------------------+---------+------+------+----------+----------------------------------+
|  1 | SIMPLE      | p     | NULL       | range | date_created_type | date_created_type | 6       | NULL |    1 |    10.00 | Using index condition; Using MRR |
+----+-------------+-------+------------+-------+-------------------+-------------------+---------+------+------+----------+----------------------------------+
1 row in set, 1 warning (0.00 sec)

只读取部分字段时

读取外键 explain select car_id from sm_performance_all p where p.date_created>'2018-01-01'  and p.date_created< '2018-02-01' and p.type=0;
+----+-------------+-------+------------+-------+-------------------+-------------------+---------+------+------+----------+----------------------------------+
| id | select_type | table | partitions | type  | possible_keys     | key               | key_len | ref  | rows | filtered | Extra                            |
+----+-------------+-------+------------+-------+-------------------+-------------------+---------+------+------+----------+----------------------------------+
|  1 | SIMPLE      | p     | NULL       | range | date_created_type | date_created_type | 6       | NULL |    1 |    10.00 | Using index condition; Using MRR |
+----+-------------+-------+------------+-------+-------------------+-------------------+---------+------+------+----------+----------------------------------+
1 row in set, 1 warning (0.00 sec)
读取主键
 explain select id from sm_performance_all p where p.date_created>'2018-01-01'  and p.date_created< '2018-02-01' and p.type=0;
+----+-------------+-------+------------+-------+-------------------+-------------------+---------+------+------+----------+--------------------------+
| id | select_type | table | partitions | type  | possible_keys     | key               | key_len | ref  | rows | filtered | Extra                    |
+----+-------------+-------+------------+-------+-------------------+-------------------+---------+------+------+----------+--------------------------+
|  1 | SIMPLE      | p     | NULL       | range | date_created_type | date_created_type | 6       | NULL |    1 |    10.00 | Using where; Using index |
+----+-------------+-------+------------+-------+-------------------+-------------------+---------+------+------+----------+--------------------------+
1 row in set, 1 warning (0.00 sec)

For MRR, a storage engine uses the value of the read_rnd_buffer_size system variable as a guideline for how much memory it can allocate for its buffer.

默认256KB

show GLOBAL VARIABLES like '%buffer_size';
+-------------------------+----------+
| Variable_name           | Value    |
+-------------------------+----------+
| bulk_insert_buffer_size | 8388608  |
| innodb_log_buffer_size  | 16777216 |
| innodb_sort_buffer_size | 1048576  |
| join_buffer_size        | 262144   |
| key_buffer_size         | 8388608  |
| myisam_sort_buffer_size | 8388608  |
| preload_buffer_size     | 32768    |
| read_buffer_size        | 131072   |
| read_rnd_buffer_size    | 262144   |
| sort_buffer_size        | 262144   |
+-------------------------+----------+
10 rows in set (0.00 sec)

三、表连接实现方式

3.1 Nested Loop Join

将驱动表/外部表的结果集作为循环基础数据，然后循环该结果集，每次获取一条数据作为下一个表的过滤条件查询数据，然后合并结果，获取结果集返回给客户端。Nested-Loop一次只将一行传入内层循环, 所以外层循环(的结果集)有多少行, 内存循环便要执行多少次，效率非常差。

EXPLAIN SELECT * from sm_performance_all  p LEFT JOIN sm_bill b ON p.bill_id > b.car_id where p.company_id>1024;
+----+-------------+-------+------------+-------+------------------------------+------------------------------+---------+------+--------+----------+------------------------------------------------+
| id | select_type | table | partitions | type  | possible_keys                | key                          | key_len | ref  | rows   | filtered | Extra                                          |
+----+-------------+-------+------------+-------+------------------------------+------------------------------+---------+------+--------+----------+------------------------------------------------+
|  1 | SIMPLE      | p     | NULL       | range | FK_a9t29a4b2af1vfny1j2minc1x | FK_a9t29a4b2af1vfny1j2minc1x | 9       | NULL |  20263 |   100.00 | Using index condition; Using MRR               |
|  1 | SIMPLE      | b     | NULL       | ALL   | car_id_idx                   | NULL                         | NULL    | NULL | 738383 |   100.00 | Range checked for each record (index map: 0x2) |
+----+-------------+-------+------------+-------+------------------------------+------------------------------+---------+------+--------+----------+------------------------------------------------+
2 rows in set, 1 warning (0.00 sec)

3.2 Block Nested-Loop Join

将外层循环的行/结果集存入join buffer, 内层循环的每一行与整个buffer中的记录做比较，从而减少内层循环的次数。主要用于当被join的表上无索引。

CREATE TABLE t1 (a int PRIMARY KEY, b int);
CREATE TABLE t2 (a int PRIMARY KEY, b int);
INSERT INTO t1 VALUES (1,2), (2,1), (3,2), (4,3), (5,6), (6,5), (7,8), (8,7), (9,10);
INSERT INTO t2 VALUES (3,0), (4,1), (6,4), (7,5);
EXPLAIN 
SELECT * FROM t1 LEFT JOIN t2 ON t1.a = t2.a WHERE t2.b <= t1.a AND t1.a <= t1.b;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra                                              |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+
|  1 | SIMPLE      | t1    | NULL       | ALL  | PRIMARY       | NULL | NULL    | NULL |    9 |    33.33 | Using where                                        |
|  1 | SIMPLE      | t2    | NULL       | ALL  | PRIMARY       | NULL | NULL    | NULL |    4 |    25.00 | Using where; Using join buffer (Block Nested Loop) |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+
2 rows in set, 1 warning (0.00 sec)

3.3 Batched Key Access

当被join的表能够使用索引时，就先好顺序，然后再去检索被join的表。对这些行按照索引字段进行排序，因此减少了随机IO。如果被Join的表上没有索引，则使用老版本的BNL策略。

关于Mysql 的 ICP、MRR、BKA等特性

一、ICP( Index_Condition_Pushdown)

二、MRR(Multi-Range Read )

三、表连接实现方式

相关推荐