Postgres 中的子查询根据 sum 函数给出不正确的结果

Posted 2023-03-31

【中文标题】Postgres 中的子查询根据 sum 函数给出不正确的结果

【英文标题】：Subquery in Postgres giving incorrect result based on sum function

【发布时间】：2018-05-29 20:57:06

【问题描述】：

WITH eas AS
(
    SELECT 
    DATE_TRUNC('week',create_ts)::DATE - '1 day'::INTERVAL AS DATE           
    ,COUNT(name) AS ase
    FROM local
    WHERE  type='sr'
    GROUP BY DATE_TRUNC('week',create_ts)::DATE - '1 day'::INTERVAL
) 
, ias AS 
(       
    SELECT 
    DATE_TRUNC('week',create_ts)::DATE - '1 day'::INTERVAL AS DATE           
    ,COUNT(name) AS ase
    FROM local
    WHERE  type='sr'
    AND state NOT IN ('AZ','CA')
    GROUP BY DATE_TRUNC('week',create_ts)::DATE - '1 day'::INTERVAL
)
    SELECT 
    DATE_TRUNC('week',eas.date)::DATE - '1 day'::INTERVAL AS l_create
    ,ase + asi AS count
    FROM eas FULL JOIN ias 
    ON ias.date = eas.date
    GROUP BY ase
    ,asi 
    ,DATE_TRUNC('week',eas.date)::DATE - '1 day'::INTERVAL
    ORDER BY DATE_TRUNC('week',eas.date)::DATE - '1 day'::INTERVAL ASC

如果我运行第一条语句，它会给出以下结果：

3/18/18 0:00    21
3/25/18 0:00    46
4/1/18 0:00     25
4/8/18 0:00     25

第二句话：

3/18/18 0:00    67
3/25/18 0:00    129
4/1/18 0:00     148
4/8/18 0:00     138

如果我运行完整的查询：

2018-03-18 00:00:00 175
2018-03-25 00:00:00 173
2018-04-01 00:00:00 163

而想要的结果是：

3/18/18 0:00    88
3/25/18 0:00    175
4/1/18 0:00     173
4/8/18 0:00     163

我错过了什么？？

【问题讨论】：

好吧，首先你缺少循环。您的代码中没有循环。你正在做的是 CTE 中的 SELECT

“我错过了什么？？”发布表的结构（如CREATE TABLE）并解释循环的含义。这是 PL/pgSQL 循环的一部分吗？请出示。

但是对于这个简单的查询，CTE 还不够吗？对不起这里的新手

@stickybit，循环是指代码的“eas”和“aes”部分

我只是说这些不是循环。它们是公用表表达式中的选择语句。

【参考方案1】：

错误在最后一条语句中。您出于某种原因按每一列分组并在此处执行加法：ase + asi。相反，我会尝试替换以下行：

,ase + asi AS count
...
GROUP BY ase, asi, DATE_TRUNC('week',eas.date)::DATE - '1 day'::INTERVAL

与

,sum(ase)+sum(asi) AS count
...
GROUP BY DATE_TRUNC('week',eas.date)::DATE - '1 day'::INTERVAL

实际上使用聚合函数将每个日期的 ase 和 asi 的计数相加，因此查询的最后部分如下所示

SELECT 
DATE_TRUNC('week',eas.date)::DATE - '1 day'::INTERVAL AS l_create
,sum(ase)+sum(asi) AS count
FROM eas FULL JOIN ias 
ON ias.date = eas.date
GROUP BY DATE_TRUNC('week',eas.date)::DATE - '1 day'::INTERVAL
ORDER BY DATE_TRUNC('week',eas.date)::DATE - '1 day'::INTERVAL ASC

---

要进一步调查为什么会发生这种情况，您需要删除 GROUP BY 子句并查看数据以亲自了解 FULL JOIN 的作用。

---

示例

下面我附上了一个重现您的问题的示例历史记录，但我使用整数而不是日期列（对于您的情况没有区别）。

postgres=# create table a(dt int, cnt int);
CREATE TABLE
postgres=# create table b(dt int, cnt int);
CREATE TABLE
postgres=# insert into a values (1,21),(2,46),(3,25),(4,25);
INSERT 0 4
postgres=# insert into b values (1,67),(2,129),(3,148),(4,138);
INSERT 0 4
postgres=# select a.dt,sum(a.cnt)+sum(b.cnt) from a full join b on a.dt=b.dt group by a.dt order by a.dt;
 dt | ?column?
----+----------
  1 |       88
  2 |      175
  3 |      173
  4 |      163

【讨论】：

它对我有用，我刚刚在我的样本上测试过。附上代码。

【参考方案2】：

WITH eas AS ( SELECT DATE_TRUNC('week',create_ts) AS DATE ,COUNT(name) AS ase FROM local WHERE type='sr' GROUP BY DATE_TRUNC('week',create_ts) ) , ias AS ( SELECT DATE_TRUNC('week',create_ts) AS DATE ,COUNT(name) AS ase FROM local WHERE type='sr' AND state NOT IN ('AZ','CA') GROUP BY DATE_TRUNC('week',create_ts) ) SELECT DATE_TRUNC('week',eas.date)::DATE - '1 day'::INTERVAL AS l_create ,ase + asi AS count FROM eas FULL JOIN ias ON ias.date = eas.date GROUP BY ase ,asi ,DATE_TRUNC('week',eas.date)::DATE - '1 day'::INTERVAL ORDER BY DATE_TRUNC('week',eas.date)::DATE - '1 day'::INTERVAL ASC

这解决了它！谢谢大家！