数据量增加导致MySQL执行计划改变

数据量增加导致MySQL执行计划改变

收到运维同学电话，mysql服务器连接数满了，登录服务器查看，确实满了，好吧，首先增加连接数到2500，暂时提供对外服务。连接继续升高，又快达到2500。发现有大量的查询时间将近到了1200秒，大量的长连接堆积，导致连接数攀升，看来还是sql的问题。在这些长连接中，发现这样的sql

SELECT product_id,gift_id,gift_original_price,gift_count, FROM promo_xxx WHERE promotion_id IN (589994,589994) AND product_id IN (22569455) AND is_valid=1;

mysql> explain SELECT product_id,gift_id,gift_original_price,gift_count FROM promo_gift WHERE promotion_id IN (589994,589994) AND product_id IN (22569455) AND is_valid=1;
+----+-------------+-----------------+------+---------------+------+---------+------+--------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-----------------+------+---------------+------+---------+------+--------+-------------+
| 1 | SIMPLE | promo_gift_list | ALL | id_promo_gift | NULL | NULL | NULL | 249188 | Using where |
+----+-------------+-----------------+------+---------------+------+---------+------+--------+-------------+
1 row in set (0.04 sec)
mysql> show index from promo_gift;
+-----------------+------------+-----------------+--------------+-----------------+-----------+-------------+----------+--------+------+------------+---------+---------------+
| Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment | Index_comment |
+-----------------+------------+-----------------+--------------+-----------------+-----------+-------------+----------+--------+------+------------+---------+---------------+
| promo_gift_list | 0 | PRIMARY | 1 | id | A | 261184 | NULL | NULL | | BTREE | | |
| promo_gift_list | 0 | id_promo_gift | 1 | promotion_id | A | 1140 | NULL | NULL | YES | BTREE | | |
| promo_gift_list | 0 | id_promo_gift | 4 | product_id | A | 261184 | NULL | NULL | YES | BTREE | | |

狗血的sql，竟然走全表扫描，但是promotion_id有索引啊，为什么没有走索引呢？而且以前建立的索引，走的好好的，今天怎么就出现问题了，这是一个问题
那我们可以通过last_query_cost 查看sql消耗

mysql>SELECT product_id,gift_id,gift_original_price,gift_count FROM promo_gift WHERE promotion_id IN (589994,589994) AND product_id IN (22569455) AND is_valid=1;

mysql>show status like 'last_query_cost';
+-----------------+--------------+
| Variable_name | Value |
+-----------------+--------------+
| Last_query_cost | 52626.599000 |
+-----------------+--------------+
1 row in set (0.00 sec)

不走索引，那我们强制使用索引
mysql> explain SELECT product_id,gift_id,gift_original_price,gift_count FROM promo_gift force index(id_promo_gift) WHERE promotion_id IN (589994,589994) AND product_id IN (22569455) AND is_valid=1;
+----+-------------+-----------------+-------+---------------+---------------+---------+------+--------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-----------------+-------+---------------+---------------+---------+------+--------+-------------+
| 1 | SIMPLE | promo_gift_list | range | id_promo_gift | id_promo_gift | 5 | NULL | 124594 | Using where |
+----+-------------+-----------------+-------+---------------+---------------+---------+------+--------+-------------+
1 row in set (0.02 sec)

嗯，加上索引了，那么sql消耗怎么样呢？
mysql> SELECT product_id,gift_id,gift_original_price,gift_count FROM promo_gift force index(id_promo_gift) WHERE promotion_id IN (589994,589994) AND product_id IN (22569455) AND is_valid=1;
+------------+----------+---------------------+------------+
| product_id | gift_id | gift_original_price | gift_count |
+------------+----------+---------------------+------------+
| 22569455 | 23230046 | 147.00 | 1 |
+------------+----------+---------------------+------------+
1 row in set (0.40 sec)

mysql> show status like 'last_query_cost';
+-----------------+---------------+
| Variable_name | Value |
+-----------------+---------------+
| Last_query_cost | 174432.609000 |
+-----------------+---------------+
1 row in set (0.00 sec)

我们发现如果使用这个索引，sql消耗174432.609000>52626.599000,mysql优化器认为使用这个id_promo_gift索引，sql消耗是非常大的，这就是mysql执行不使用这个索引的原因。
后来开发人员说，昨天晚上这个表增加了11万多的数据，嗯，数据量增加，mysql执行计划改变。那好吧，单独product_id列再加一个索引。

mysql> alter table promo_gift_list add index product_id(product_id);
Query OK, 0 rows affected (6.45 sec)
Records: 0 Duplicates: 0 Warnings: 0

mysql> explain SELECT product_id,gift_id,gift_original_price,gift_count FROM promo_gift_list WHERE promotion_id IN (589994,589994) AND product_id IN (22569455) AND is_valid=1 AND IFNULL(is_delete,0)!=1;
+----+-------------+-----------------+------+--------------------------+------------+---------+-------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-----------------+------+--------------------------+------------+---------+-------+------+-------------+
| 1 | SIMPLE | promo_gift_list | ref | id_promo_gift,product_id | product_id | 5 | const | 2 | Using where |
+----+-------------+-----------------+------+--------------------------+------------+---------+-------+------+-------------+
1 row in set (0.00 sec)
使用了刚才新加的索引 product_id

mysql> SELECT product_id,gift_id,gift_original_price,gift_count FROM promo_gift_list WHERE promotion_id IN (589994,589994) AND product_id IN (22569455) AND is_valid=1 AND IFNULL(is_delete,0)!=1;
+------------+----------+---------------------+------------+
| product_id | gift_id | gift_original_price | gift_count |
+------------+----------+---------------------+------------+
| 22569455 | 23230046 | 147.00 | 1 |
+------------+----------+---------------------+------------+
1 row in set (0.00 sec)

mysql> show status like 'last_query_cost';
+-----------------+----------+
| Variable_name | Value |
+-----------------+----------+
| Last_query_cost | 2.399000 |
+-----------------+----------+
1 row in set (0.01 sec)

sql消耗降到了2.399000，ok，问题解决。连接数很快从1000多降到100以内。
数据量的增加导致了mysql执行计划的改变，那么mysql的cost是怎么计算的呢？
cost=io_cost+cpu_cost
cpu_cost位于mysql上层，处理返回的记录所花开销，io_cost存储引擎层，读取也没的IO开销。最直接的方式last_query_cost记录sql的cost。查看last_query_cost可以初步判断sql的cost，明白mysql优化器执行的依据。

推荐阅读：

分析MySQL慢查询日志的好工具--mysqlsla

MySQL前缀索引导致的慢查询

【体系结构】MySQL 日志文件--慢查询日志

MySQL调优三步曲(慢查询、explain profile)