正确分组此查询以获得最小值？

Question

桌车：

id  serial  size    batch
--------------------------
1   x99     Large    NULL
2   x99     Small    Q
3   x99     Med     NULL
4   x99     Large    K

使用表代码分配连接两次：

Id  serial  size    batch   code    precedence
-----------------------------------------------
1   x99     Large   NULL    5000    1
2   x99     NULL      K     3000    2
3   x99     NULL      Q     2000    3
4   x99     NULL    NULL    500     4

使用以下内容：

select 
    c.*, 
    coalesce(ca1.code, ca2.code) as code, 
    min(coalesce(ca1.precedence, ca2.precedence)) as precedence
from
    cars c 
left join
    codeassignment ca1 on ca1.serial = c.serial 
                       and (ca1.size = c.size or ca1.batch = c.batch) 
left join
    codeassignment ca2 on ca2.serial = c.serial 
                       and ca1.size is null and ca2.size is null 
                       and ca1.batch is null and ca2.batch is null
group by 
    c.id, c.serial, c.size, c.batch, ca1.code, ca2.code

但它会导致 'Large' 和批次 'K' 的匹配出现两次：

id  serial  size    batch   code    precedence
----------------------------------------------
1   x99     Large   A       5000    1
2   x99     Small   Q       2000    3
3   x99     Med     P       500     4
4   x99     Large   K       3000    2
4   x99     Large   K       5000    1

我想选择所有上述行，但对于最后两行，只有优先级最低的行（这就是我尝试min(precedence) 的原因），但我认为 group by 正确执行此操作是错误的。所以基本上serial、size和batch相同的地方应该只有一行。

这是一个小提琴：

https://dbfiddle.uk/?rdbms=sqlserver_2019&fiddle=47700d67170a7beb39bfc523ffdfbcfc

Answer 1

为迟来的答案道歉，但我仍然有这个开放。作为替代方案，无论如何都要完成它。

如果我正确理解目标，您想匹配序列号、填充的大小和填充的批次，优先级领先，无论填充的字段是否“更强”匹配？

那么一个连接就足够了：

select c.Serial, c.Size,c.Batch, min(isnull(ca.Precedence,0)) Precedence
from cars c
left join codeassignment ca on ca.serial = c.serial and (ca.size is null or ca.size=c.size) and (ca.batch is null or ca.batch=c.batch)
group by c.Serial,c.Size,c.Batch

这只是为了显示单连接，不需要 ca1 和 ca2（除非我缺少某些要求）

以此为基础，另一种匹配最小值或其他聚合的方法是使用窗口函数而不是分组依据。我对那些有一点偏好（结合 cte，因为你不能直接使用窗口函数作为标准）

with q as(
select c.*, ca.Code, ca.Precedence, row_number() over (partition by c.Serial, c.Size, c.Batch order by Precedence) rnr
from cars c
left join codeassignment ca on ca.serial = c.serial and (ca.size is null or ca.size=c.size) and (ca.batch is null or ca.batch=c.batch)
)
select Serial, Size, Batch, Precedence, Code
from q
where rnr = 1

Answer 2

问题是您在查询中按“太多东西”进行分组，因此您最终会得到不想要的离散行。解决此问题的一种方法是将查询分为两个阶段，一个是获取每个 id、序列、大小和批次的最小优先级，然后是获取丢失的数据。像这样的：

WITH x AS (
    SELECT 
        c.id, 
        c.serial, 
        c.size, 
        c.batch, 
        MIN(ISNULL(ca1.precedence, ca2.precedence)) AS precedence
    FROM 
        cars c 
        LEFT JOIN codeassignment ca1
            ON ca1.serial = c.serial AND (ca1.size = c.size OR ca1.batch = c.batch) 
        LEFT JOIN codeassignment ca2 ON ca2.serial = c.serial AND ca1.size is null and ca2.size is null and ca1.batch is null and ca2.batch is null
     GROUP BY 
        c.id, 
        c.serial, 
        c.size, 
        c.batch)
SELECT
    x.*,
    ISNULL(ca1.code, ca2.code) AS code
FROM
    x
    INNER JOIN cars c ON c.id = x.id AND c.serial = x.serial AND c.size = x.size AND c.batch = x.batch
    LEFT JOIN codeassignment ca1
     ON ca1.serial = c.serial and (ca1.size = c.size or ca1.batch = c.batch) AND ca1.precedence = x.precedence
    LEFT JOIN codeassignment ca2
     ON ca2.serial = c.serial and
     ca1.size is null and ca2.size is null and ca1.batch is null and ca2.batch is null AND ca2.precedence = x.precedence;

由于某种原因，我无法使用它，但我可以在 SQL Server 2019 上本地运行它以获得：

id  serial  size    batch   precedence  code
1   x99     Large   A       1           5000
2   x99     Small   Q       3           2000
3   x99     Med     P       4           500
4   x99     Large   K       1           5000