前言
昨天遇到一个问题,需要对一张关系表进行重构和优化。然而这张关系表由于已有代码没有注重并发导致了很多的脏数据,即重复数据。
表名thread_recommend,帖子推荐表,为两个实体user_id
和thread_id
的(推荐)关系表,表结构很简单如下:
/*用户推荐帖子记录表*/
CREATE TABLE `thread_recommend` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`thread_id` int(11) DEFAULT NULL COMMENT '被用户推荐的帖子编号',
`user_id` int(11) DEFAULT NULL COMMENT '推荐该帖子的用户编号',
`status` int(11) DEFAULT '1' COMMENT '状态0 取消推荐,1推荐',
`created` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP COMMENT '推荐时间',
PRIMARY KEY (`id`),
KEY `userid` (`user_id`) USING BTREE
) ENGINE=InnoDB;
问题在于,由于代码不规范,在高并发时(或数据库压力大时造成的延时积压时)会出现多个(相同thread_id和user_id的)组合,如下:
之后你们懂的,各种和原想不一致的神奇bug喷涌而出,比如:
我刚刚取消了推荐,怎么还显示我推荐着!!
显示的总推荐数怎么和实际推荐用户加起来不一样!!
解决方案一:使用insert where not exists
语句
声明:此方案并不是最佳方案,不推荐使用。
先上代码:(这里拿另一个关系表的真实query举例,原理一样)
INSERT INTO `user_topic` (`user_id`, `topic_id`)
SELECT :userId, :topicid FROM `user_topic`
WHERE NOT EXISTS (SELECT * FROM `user_topic`
WHERE `user_topic`.`user_id` = :userId
AND `user_topic`.`topic_id` = :topicid)
LIMIT 1;
(相同方法见http://stackoverflow.com/a/31...)
通过这种“插入时判断不存在才插入并返回行数为1,存在的话返回行数为0”的方法,可以做到:
只有在返回行数为1的情况下才执行之后逻辑(如缓存内的统计数+1,缓存内帖子推荐人增加此userId等等)
如果返回行数为0,则接口返回error
解决方案二:清理脏数据并建立联合唯一索引
这个方案是本文的核心了,也是我们目前认为的最佳实践。
第一步:查找user_id, thread_id的联合duplication
SELECT a.* FROM `thread_recommend` a
INNER JOIN (SELECT * FROM `thread_recommend` GROUP BY `thread_id`, `user_id` HAVING COUNT(id) > 1) b ON a.`thread_id` = b.`thread_id` AND a.`user_id` = b.`user_id`
ORDER BY a.`user_id` ASC, a.`thread_id` ASC, a.`id` DESC
或简单的版本
SELECT * FROM `thread_recommend`
WHERE (`user_id`, `thread_id`) IN (SELECT `user_id`, `thread_id` FROM `thread_recommend` GROUP BY `user_id`, `thread_id` HAVING COUNT(1) > 1);
得到
哇!所有的重复项都在这里了,好想马上把它们干掉!
现在需要将重复的条目中ID更大的所有条目都删除,只留ID最小的那一个。
删之前先获得需要删除项,比对一下,
SELECT * FROM `thread_recommend`
WHERE (`user_id`, `thread_id`) IN (SELECT `user_id`, `thread_id` FROM `thread_recommend` GROUP BY `user_id`, `thread_id` HAVING COUNT(1) > 1)
AND `id` NOT IN (SELECT MIN(`id`) FROM `thread_recommend` GROUP BY `user_id`, `thread_id` HAVING COUNT(1) > 1);
下一步,SELECT * FROM
改成DELETE FROM
,删除!
DELETE FROM `thread_recommend`
WHERE (`user_id`, `thread_id`) IN (SELECT `user_id`, `thread_id` FROM `thread_recommend` GROUP BY `user_id`, `thread_id` HAVING COUNT(1) > 1)
AND `id` NOT IN (SELECT MIN(`id`) FROM `thread_recommend` GROUP BY `user_id`, `thread_id` HAVING COUNT(1) > 1);
Oops!报错! You can't specify target table 'thread_recommend' for update in FROM clause
这是Mysql的一个小问题,我们参见解决方案 http://stackoverflow.com/a/14... 后修改一下SQL就好:
DELETE FROM `thread_recommend`
WHERE (`user_id`, `thread_id`) IN (SELECT `user_id`, `thread_id` FROM (SELECT * FROM `thread_recommend`) a GROUP BY `user_id`, `thread_id` HAVING COUNT(1) > 1)
AND `id` NOT IN (SELECT MIN(`id`) FROM (SELECT * FROM `thread_recommend`) b GROUP BY `user_id`, `thread_id` HAVING COUNT(1) > 1);
最后,加联合唯一索引!
ALTER TABLE `thread_recommend`
ADD UNIQUE KEY `thread_id_user_id_unique`(`thread_id`,`user_id`) USING BTREE;
Of course,如果上述清理工作没有完成将会报错!
完!