MySQL子查询返回不正确的结果？

Question

我有以下 MySQL 查询/子查询：

SELECT id, user_id, another_id, myvalue, created, modified,
(
    SELECT id 
    FROM users_values AS ParentUsersValue
    WHERE ParentUsersValue.user_id = UsersValue.user_id
    AND ParentUsersValue.another_id = UsersValue.another_id 
    AND ParentUsersValue.id < UsersValue.id 
    ORDER BY id DESC 
    LIMIT 1
) AS old_id

FROM users_values AS UsersValue
WHERE created >= '2009-12-20' 
AND created <= '2010-01-21' 
AND user_id = 9917
AND another_id = 23

鉴于列出的条件，子查询 (old_id) 的结果应该为 null（在我的表中找不到匹配项）。而不是 MySQL 返回 null，它似乎只是删除了“WHERE ParentUsersValue.user_id = UsersValue.user_id”子句并选择与其他两个字段匹配的第一个值。这是一个 MySQL 错误，还是出于某种原因这是预期的行为？

更新：

CREATE TABLE users_values (
    id int(11) NOT NULL AUTO_INCREMENT,
    user_id int(11) DEFAULT NULL,
    another_id int(11) DEFAULT NULL,
    myvalue double DEFAULT NULL,
    created datetime DEFAULT NULL,
    modified datetime DEFAULT NULL,
    PRIMARY KEY (id)
) ENGINE=InnoDB AUTO_INCREMENT=2801 DEFAULT CHARSET=latin1

EXPLAIN EXTENDED:

id  select_type table   type    possible_keys   key key_len ref rows    filtered    Extra
1   PRIMARY UsersValue  index_merge user_id,another_id  user_id,another_id  5,5 NULL    1   100.00  Using intersect(user_id,another_id); Using where
2   DEPENDENT SUBQUERY  ParentUsersValue    index   PRIMARY,user_id,another_id  PRIMARY 4   NULL    1   100.00  Using where

EXPLAIN EXTENDED Warning 1003:

select `mydb`.`UsersValue`.`id` AS `id`,`mydb`.`UsersValue`.`user_id` AS `user_id`,`mydb`.`UsersValue`.`another_id` AS `another_id`,`mydb`.`UsersValue`.`myvalue` AS `myvalue`,`mydb`.`UsersValue`.`created` AS `created`,`mydb`.`UsersValue`.`modified` AS `modified`,(select `mydb`.`ParentUsersValue`.`id` AS `id` from `mydb`.`users_values` `ParentUsersValue` where ((`mydb`.`ParentUsersValue`.`user_id` = `mydb`.`UsersValue`.`user_id`) and (`mydb`.`ParentUsersValue`.`another_id` = `mydb`.`UsersValue`.`another_id`) and (`mydb`.`ParentUsersValue`.`id` < `mydb`.`UsersValue`.`id`)) order by `mydb`.`ParentUsersValue`.`id` desc limit 1) AS `old_id` from `mydb`.`users_values` `UsersValue` where ((`mydb`.`UsersValue`.`another_id` = 23) and (`mydb`.`UsersValue`.`user_id` = 9917) and (`mydb`.`UsersValue`.`created` >= '2009-12-20') and (`mydb`.`UsersValue`.`created` <= '2010-01-21'))

Answer 1

这将为我返回正确的结果 (NULL)：

CREATE TABLE users_values (id INT NOT NULL PRIMARY KEY, user_id INT NOT NULL, another_id INT NOT NULL, created DATETIME NOT NULL);

INSERT
INTO    users_values VALUES (1, 9917, 23, '2010-01-01');

SELECT  *,
        (
        SELECT  id
        FROM    users_values AS ParentUsersValue
        WHERE   ParentUsersValue.user_id = UsersValue.user_id
                AND ParentUsersValue.another_id = UsersValue.another_id
                AND ParentUsersValue.id < UsersValue.id
        ORDER BY id
                DESC
        LIMIT 1
        ) AS old_id
FROM    users_values AS UsersValue
WHERE   created >= '2009-12-20'
        AND created <= '2010-01-21'
        AND user_id = 9917
        AND another_id = 23

能否请您运行此查询：

SELECT  COUNT(*)
FROM    users_values AS UsersValue
WHERE   user_id = 9917
        AND another_id = 23

并确保它返回1?

请注意，您的子查询不会过滤 created，因此子查询可能会返回超出主查询定义范围的值。

更新：

这绝对是MySQL 中的一个错误。

最可能的原因是为UsersValues选择的访问路径是index_intersect。

这会从两个索引中选择适当的范围并构建它们的交集。

由于该错误，依赖子查询在交集完成之前被评估，这就是为什么您会得到正确的another_id 但错误的user_id 的结果。

您能否在对UsersValues 强制进行PRIMARY 扫描时检查问题是否仍然存在：

SELECT  *,
        (
        SELECT  id
        FROM    users_values AS ParentUsersValue
        WHERE   ParentUsersValue.user_id = UsersValue.user_id
                AND ParentUsersValue.another_id = UsersValue.another_id
                AND ParentUsersValue.id < UsersValue.id
        ORDER BY id
                DESC
        LIMIT 1
        ) AS old_id
FROM    users_values AS UsersValue FORCE INDEX (PRIMARY)
WHERE   created >= '2009-12-20'
        AND created <= '2010-01-21'
        AND user_id = 9917
        AND another_id = 23

此外，对于此查询，您应该在 (user_id, another_id, id) 上创建一个复合索引，而不是在 user_id 和 another_id 上创建两个不同的索引。

创建索引并稍微重写查询：

SELECT  *,
        (
        SELECT  id
        FROM    users_values AS ParentUsersValue
        WHERE   ParentUsersValue.user_id = UsersValue.user_id
                AND ParentUsersValue.another_id = UsersValue.another_id
                AND ParentUsersValue.id < UsersValue.id
        ORDER BY
                user_id DESC, another_id DESC, id DESC
        LIMIT 1
        ) AS old_id
FROM    users_values AS UsersValue
WHERE   created >= '2009-12-20'
        AND created <= '2010-01-21'
        AND user_id = 9917
        AND another_id = 23

user_id DESC, another_id DESC 子句在逻辑上是多余的，但它们会使索引用于排序。

Answer 2

您是否尝试运行子查询只是为了查看您是否获得了正确的结果？您能否向我们展示您的 users_values 表的架构？

另外，尝试将子查询中的 SELECT id 替换为 SELECT ParentUsersValue.id