MariadB：使用相关子查询只给出一行

Question

我有两个表

c0

和

m0

，我使用左连接合并它们。结果是一个包含 13 行的表。我想添加接下来 40 天窗口的第一个日期。数据如下：

DROP TABLE IF EXISTS test0.c0;
CREATE TABLE test0.c0 (
  id      INTEGER
, cons_dt DATE
);

INSERT INTO test0.c0 VALUES
 ('1','2000-01-01')
,('1','2000-02-01')
,('1','2000-03-01')
,('1','2000-04-01')
,('1','2000-05-01')
,('1','2000-06-01')
,('1','2000-07-01')
,('1','2000-08-01')
,('1','2000-09-01')
,('1','2000-10-01')
,('1','2000-11-01')
,('1','2000-12-01')
;

DROP TABLE IF EXISTS test0.m0;
CREATE TABLE test0.m0 ( 
  id        INTEGER
, start_dt  DATE
, atc       CHAR(1)
);

INSERT INTO test0.m0 VALUES
 ('1','2000-03-01','A')
,('1','2000-04-01','A')
,('1','2000-08-01','A')
,('1','2000-08-01','B')
,('1','2000-09-01','A')
,('1','2000-10-01','B')
;

这是代码：

SELECT c.*, m.start_dt, m.atc
      , (SELECT MIN(c.cons_dt)
         FROM test0.c0 c2
         WHERE c2.id = c.id 
                   AND c2.cons_dt BETWEEN c.cons_dt + INTERVAL 1 DAY AND c.cons_dt + INTERVAL 40 DAY
         ) AS stop_dt
FROM test0.c0 c
LEFT OUTER JOIN test0.m0 m ON c.id = m.id AND c.cons_dt = m.start_dt

这是输出：

+----+------------+------------+-----+---------+
| id |  cons_dt   |  start_dt  | atc | stop_dt |
+----+------------+------------+-----+---------+
|  1 | 2000-03-01 | 2000-03-01 | A   | \N      |
+----+------------+------------+-----+---------+

如果条件未满足，我预计会得到 13 行有停止日期或没有停止日期的行。代码有什么问题吗？ 我知道我可以使用 Windows 函数，但这不适用于此类数据。考虑这是一个简化的数据集。原版有很多id和各种atc。

更新1： 这里给出了给出正确结果的代码：

SELECT c.*, m.start_dt, m.atc, MIN(c2.cons_dt) AS stop_dt
FROM test0.c0 c
LEFT OUTER JOIN test0.m0 m ON c.id = m.id AND c.cons_dt = m.start_dt
LEFT JOIN test0.c0 c2 ON c.id = c2.id
                      AND c2.cons_dt BETWEEN c.cons_dt + INTERVAL 1 DAY AND c.cons_dt + INTERVAL 30 DAY
GROUP BY c.id, c.cons_dt, m.start_dt, m.atc
;

这是我得到的和我期望的表格：

+----+------------+------------+-----+------------+
| id |  cons_dt   |  start_dt  | atc |  stop_dt   |
+----+------------+------------+-----+------------+
|  1 | 2000-01-01 | \N         | \N  | \N         |
|  1 | 2000-02-01 | \N         | \N  | 2000-03-01 |
|  1 | 2000-03-01 | 2000-03-01 | A   | \N         |
|  1 | 2000-04-01 | 2000-04-01 | A   | 2000-05-01 |
|  1 | 2000-05-01 | \N         | \N  | \N         |
|  1 | 2000-06-01 | \N         | \N  | 2000-07-01 |
|  1 | 2000-07-01 | \N         | \N  | \N         |
|  1 | 2000-08-01 | 2000-08-01 | A   | \N         |
|  1 | 2000-08-01 | 2000-08-01 | B   | \N         |
|  1 | 2000-09-01 | 2000-09-01 | A   | 2000-10-01 |
|  1 | 2000-10-01 | 2000-10-01 | B   | \N         |
|  1 | 2000-11-01 | \N         | \N  | 2000-12-01 |
|  1 | 2000-12-01 | \N         | \N  | \N         |
+----+------------+------------+-----+------------+

我只是想了解为什么子查询不起作用。一个有趣的问题是哪个表现更好。

更新2：ONLY_FULL_GROUP_BY slaakso 指出了问题所在。当 ONLY_FULL_GROUP_BY 启用时（这是默认值），引擎限制聚合函数的使用。然而，这很令人困惑，因为这有效：

SELECT c.* ,m.atc ,m.start_dt 
     , (SELECT COUNT(*)
        FROM test0.c0 c2
        LEFT OUTER JOIN test0.m0 m2 ON c2.id = m2.id AND c2.cons_dt = m2.start_dt
        WHERE c2.id = c.id
        AND m2.atc <=> m.atc
        AND c.cons_dt > c2.cons_dt
        ) + 1 AS counter
FROM test0.c0 C
LEFT OUTER JOIN test0.m0 m ON c.id = m.id AND c.cons_dt = m.start_dt
ORDER BY m.atc, c.cons_dt, counter
;  

这是输出：

+----+------------+-----+------------+---------+
| id |  cons_dt   | atc |  start_dt  | counter |
+----+------------+-----+------------+---------+
|  1 | 2000-01-01 | \N  | \N         |       1 |
|  1 | 2000-02-01 | \N  | \N         |       2 |
|  1 | 2000-05-01 | \N  | \N         |       3 |
|  1 | 2000-06-01 | \N  | \N         |       4 |
|  1 | 2000-07-01 | \N  | \N         |       5 |
|  1 | 2000-11-01 | \N  | \N         |       6 |
|  1 | 2000-12-01 | \N  | \N         |       7 |
|  1 | 2000-03-01 | A   | 2000-03-01 |       1 |
|  1 | 2000-04-01 | A   | 2000-04-01 |       2 |
|  1 | 2000-08-01 | A   | 2000-08-01 |       3 |
|  1 | 2000-09-01 | A   | 2000-09-01 |       4 |
|  1 | 2000-08-01 | B   | 2000-08-01 |       1 |
|  1 | 2000-10-01 | B   | 2000-10-01 |       2 |
+----+------------+-----+------------+---------+

为什么第一个代码不起作用而第二个代码不起作用？这两个代码有何不同？

Answer 1

将普通列与聚合函数混合使用时，需要包含

GROUP BY

子句。

确保您的服务器设置了

ONLY_FULL_GROUP_BY

模式，因为它将捕获许多不正确的查询。