加个order by使得查询效率提高500倍

时间:2021-12-30 14:26:49

很简单的三个表:

p248_user记录用户信息

CREATE TABLE `p248_user` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`list_ids` varchar(4000) NOT NULL DEFAULT '',
`email` varchar(255) NOT NULL,
`mobile` varchar(20) NOT NULL,
`_created` datetime NOT NULL,
`_updated` datetime NOT NULL,
`hb_status` tinyint(4) DEFAULT '0',
`sb_status` tinyint(4) DEFAULT '0',
`unsubscribe_email_status` tinyint(4) DEFAULT '0',
`unsubscribe_sms_status` tinyint(4) DEFAULT '0',
`hb_time` datetime DEFAULT NULL,
`unsubscribe_email_time` datetime DEFAULT NULL,
`unsubscribe_sms_time` datetime DEFAULT NULL,
`_create_operator_name` varchar(100) DEFAULT NULL,
`_update_operator_name` varchar(100) DEFAULT NULL,
`_create_operator_email` varchar(100) DEFAULT NULL,
`_update_operator_email` varchar(100) DEFAULT NULL,
`name` varchar(255) NOT NULL DEFAULT '',
`time` varchar(255) NOT NULL DEFAULT '',
`year` int(11) NOT NULL DEFAULT '0',
PRIMARY KEY (`id`),
UNIQUE KEY `u1` (`email`,`mobile`) USING BTREE,
KEY `_updated` (`_updated`),
KEY `mobile` (`mobile`)
) ENGINE=InnoDB AUTO_INCREMENT=5596286 DEFAULT CHARSET=utf8

p248_list记录组信息

CREATE TABLE `p248_list` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`name` varchar(255) NOT NULL,
`status` enum('active','delete') DEFAULT 'active',
`_created` datetime NOT NULL,
`_updated` datetime NOT NULL,
`user_count` int(11) DEFAULT '0',
`lock_status` int(11) NOT NULL DEFAULT '0',
`lock_reason` varchar(100) DEFAULT NULL,
`lock_time` datetime DEFAULT NULL,
`import_percent` int(11) DEFAULT NULL,
`hb_count` int(11) DEFAULT '0',
`sb_count` int(11) DEFAULT '0',
`unsubscribe_email_count` int(11) DEFAULT '0',
`unsubscribe_sms_count` int(11) DEFAULT '0',
`_create_operator_name` varchar(100) DEFAULT NULL,
`_update_operator_name` varchar(100) DEFAULT NULL,
`_create_operator_email` varchar(100) DEFAULT NULL,
`_update_operator_email` varchar(100) DEFAULT NULL,
PRIMARY KEY (`id`),
KEY `_updated` (`_updated`)
) ENGINE=InnoDB AUTO_INCREMENT=30 DEFAULT CHARSET=utf8

p248_user_list是个多对多的表,记录用户属于哪些组

CREATE TABLE `p248_user_list` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`user_id` int(11) NOT NULL,
`list_id` int(11) NOT NULL,
PRIMARY KEY (`id`),
UNIQUE KEY `user_list_id` (`user_id`,`list_id`),
KEY `list_id` (`list_id`)
) ENGINE=InnoDB AUTO_INCREMENT=5646298 DEFAULT CHARSET=utf8

 

 

p248_user有200万条记录, p248_user_list有1000万条记录。

 

现在要找出属于29分组,并且手机号码不为空,并且没有退订的用户。这样的用户大约有100万个。现在要把这些用户按照4000个一批放到一群临时的记录集里。

 

这个要用到分页了,一开始的想法:

第一页:

SELECT `id`, `email`, `mobile`, `_created`, `_updated`, `_create_operator_name`, `_update_operator_name`, `name`, `time`, `year` FROM `p248_user` WHERE 1 = 1 AND id IN (SELECT DISTINCT user_id FROM `p248_user_list` WHERE list_id IN (29)) AND unsubscribe_sms_status = 0 AND mobile <> '' LIMIT 0, 4000;

第二页就LIMIT 4000, 4000。第三页就LIMIT 8000, 4000。依次类推。

结果这个SQL查询耗时用了整整5秒。

 

分析一下这个查询:

mysql> explain SELECT `id`, `email`, `mobile`, `_created`, `_updated`, `_create_operator_name`, `_update_operator_name`, `name`, `time`, `year` FROM `p248_user` WHERE 1 = 1 AND id IN (SELECT DISTINCT user_id FROM `p248_user_list` WHERE list_id IN (29)) AND unsubscribe_sms_status = 0 AND mobile <> '' LIMIT 0, 4000;
+----+-------------+----------------+--------+----------------------+--------------+---------+-----------------------------+--------+------------------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+----------------+--------+----------------------+--------------+---------+-----------------------------+--------+------------------------------------+
| 1 | SIMPLE | p248_user | range | PRIMARY,mobile | mobile | 62 | NULL | 934446 | Using index condition; Using where |
| 1 | SIMPLE | p248_user_list | eq_ref | user_list_id,list_id | user_list_id | 8 | contacts.p248_user.id,const | 1 | Using index |
+----+-------------+----------------+--------+----------------------+--------------+---------+-----------------------------+--------+------------------------------------+
2 rows in set (0.00 sec)

可以看到用户表扫描了93万行,几乎是全表扫描了。也就是把所有符合条件的结果都取了出来然后再取前4000条。

 

 

把上面的查询加上了ORDER BY `id`,结果查询耗时仅0.01秒,查询速度足足提高了500倍。

为什么会这样呢?

分析一下新的查询:

mysql> explain SELECT `id`, `email`, `mobile`, `_created`, `_updated`, `_create_operator_name`, `_update_operator_name`, `name`, `time`, `year` FROM `p248_user` WHERE 1 = 1 AND id IN (SELECT DISTINCT user_id FROM `p248_user_list` WHERE list_id IN (29)) AND unsubscribe_sms_status = 0 AND mobile <> '' ORDER BY `id` LIMIT 0, 4000;
+----+-------------+----------------+--------+----------------------+--------------+---------+-----------------------------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+----------------+--------+----------------------+--------------+---------+-----------------------------+------+-------------+
| 1 | SIMPLE | p248_user | index | PRIMARY,mobile | PRIMARY | 4 | NULL | 7999 | Using where |
| 1 | SIMPLE | p248_user_list | eq_ref | user_list_id,list_id | user_list_id | 8 | contacts.p248_user.id,const | 1 | Using index |
+----+-------------+----------------+--------+----------------------+--------------+---------+-----------------------------+------+-------------+
2 rows in set (0.00 sec)

这次用户表仅扫描了8000行。也就是查询先使用了主键索引,扫描完前4000条符合条件的记录就直接结束了。

 

 

那取第二页呢:

mysql> explain SELECT `id`, `email`, `mobile`, `_created`, `_updated`, `_create_operator_name`, `_update_operator_name`, `name`, `time`, `year` FROM `p248_user` WHERE 1 = 1 AND id IN (SELECT DISTINCT user_id FROM `p248_user_list` WHERE list_id IN (29)) AND unsubscribe_sms_status = 0 AND mobile <> '' ORDER BY `id` LIMIT 4000, 4000;
+----+-------------+----------------+--------+----------------------+--------------+---------+-----------------------------+-------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+----------------+--------+----------------------+--------------+---------+-----------------------------+-------+-------------+
| 1 | SIMPLE | p248_user | index | PRIMARY,mobile | PRIMARY | 4 | NULL | 15999 | Using where |
| 1 | SIMPLE | p248_user_list | eq_ref | user_list_id,list_id | user_list_id | 8 | contacts.p248_user.id,const | 1 | Using index |
+----+-------------+----------------+--------+----------------------+--------------+---------+-----------------------------+-------+-------------+
2 rows in set (0.00 sec)

这次就要扫描16000行了,因为前4000条是第一页的没用扔掉了。

 

 

这样的话页数越大查询就会越耗时。

 

 

但实际上可以换个方法:

第一次查询结束时,得到最后一条记录的user id, 比如是6500。

第二次查询的时候用这个user_id作为条件去匹配

SELECT `id`, `email`, `mobile`, `_created`, `_updated`, `_create_operator_name`, `_update_operator_name`, `name`, `time`, `year` FROM `p248_user` WHERE id > 6500 AND id IN (SELECT DISTINCT user_id FROM `p248_user_list` WHERE list_id IN (29)) AND unsubscribe_sms_status = 0 AND mobile <> '' ORDER BY `id` LIMIT 0, 4000;

这样扫描的行数和第一页依然是一样的。

直到最后一页也是如此,耗时不会有任何明显的下降。