计算SQL中分组的行数答案

【问题标题】：Calculate number of grouped rows in SQL计算SQL中分组的行数
【发布时间】：2013-02-21 21:08:58
【问题描述】：

我有 2 个表，请求和响应

请求：

RequestId  UserId  InsertDate
1           1       5/4/2013
2           2       6/4/2012
.           .          .
.           .          .

回复：

Responseid  Requestid(FK)  ResponseCode  
    1           1              A   
    2           1              V
    3           1              M   
    4           2              A
    5           2              S   
    6           2              D
    .           .              .
    .           .              .

如果收到响应代码 A 和 D，则请求被视为“通过”（如我的示例中 id 为 2 的请求）。我想写一个 sql 查询，它会返回 3 件事：

“通过”请求的数量
通过的请求的请求 ID
未通过的请求的 RequestID

我写了一些东西，但我不喜欢它，我想还有更好的方法。我的查询是：

SELECT COUNT(*) 
FROM
(
  SELECT count(*) as c, req.RequestID
  FROM Responses res inner join Requests req 
  on req.RequestID = res.RequestID
  where 
      res.ResponseCode = 'A' or
      res.ResponseCode = 'D' 

  group by req.RequestID
)cc

where c = 2

提前致谢。

【问题讨论】：

标签： sql sql-server-2008 tsql count group-by

【解决方案1】：

要返回同时具有响应代码 A 和 D 的请求 ID，一种方法是使用 HAVING 和 COUNT：

SELECT res.requestId
FROM Responses res inner join Requests req 
  on req.RequestID = res.requestid
WHERE res.ResponseCode IN ('A','D')
GROUP BY res.requestId
HAVING COUNT(DISTINCT res.ResponseCode) = 2

SQL Fiddle Demo

【讨论】：

【解决方案2】：

通过

SELECT r1.RequestID
FROM Responses R1
JOIN Responses R2
  on R2.RequestID = R1.RequestID
 and R1.ResponseCode = 'A' 
 and R2.ResponseCode = 'D'
compute count(r1.RequestID)

（对计算语法不肯定）

不通过

SELECT distinct (r1.RequestID)
FROM Responses R1
FULL OUTTER JOIN Responses R2
  on R2.RequestID = R1.RequestID
 and R1.ResponseCode = 'A' 
 and R2.ResponseCode = 'D'
WHERE R2.RequestID is null 
   OR R1.RequestID is null

通过

SELECT r1.RequestID
FROM  Responses R1
WHERE R1.ResponseCode = 'A' 
INTERSECT 
SELECT r1.RequestID
FROM  Responses R1
WHERE R1.ResponseCode = 'D'

不通过

SELECT r1.RequestID
FROM  Responses R1
EXCEPT
SELECT r1.RequestID
FROM  Responses R1
WHERE R1.ResponseCode = 'D'
   OR R1.ResponseCode = 'A'

【讨论】：

【解决方案3】：

我想不出另一种完全不同的方法（即没有等效的子查询或 CTE）。但是，我会分别明确测试“A”和“D”值：

SELECT COUNT(*) as c, req.RequestID
FROM Responses res inner join
     Requests req 
     on req.RequestID = res.EquifaxIDCompareRequestID
where res.ResponseCode in ('A', 'D')
group by EquifaxIDCompareRequestID
having SUM(case when res.ResponseCode = 'D' then 1 else 0 end) > 0 and
       SUM(case when res.ResponseCode = 'A' then 1 else 0 end)

having 子句类似于您的外部 where 子句，除了它检查“A”和“D”是否都在结果集中。

还有另一种方法使用连接或子查询。像这样的：

select EquifaxIDCompareRequestID
from Requests req
where EquifaxIDCompareRequestID in (select RequestID from responses where ResponseCode = 'A') and
      EquifaxIDCompareRequestID in (select RequestID from responses where ResponseCode = 'D')

【讨论】：

【解决方案4】：

对于每个requestId，下面给出了它是否通过以及它所在的组中的总数（通过或失败）（即，如果它通过，则通过的请求总数；如果它失败，失败的请求总数）。

with myCTE as
(
    select rq.requestId, case when rA.responseCode='A' and rD.responseCode='D' then 1 else 0 end as passed
    from dbo.Requests as rq
    outer apply
    (
        select top (1) re.responseCode
        from dbo.Responses as re
        where re.requestId=rq.requestId
            and re.responseCode='A'
    ) as rA
    outer apply
    (
        select top (1) re.responseCode
        from dbo.Responses as re
        where re.requestId=rq.requestId
            and re.responseCode='D'
    ) as rD
)
select *, count(*) over (partition by m.passed) as totalNumInGroup
from myCTE as m

【讨论】：