查找子组的最小值，过滤掉同一父组中前面的最小值

Question

如何获取组（Acc、TranType）的最小值，但过滤掉 Acc 组前面行中使用的任何最小值。前面的行将定义为 Acc asc，TranType asc。

PosCancelID 应该只在每个 Acc 组出现一次。但是相同的 PosCancelID 可能出现在数据集中的另一个 Acc Group 中。

因此，对于给定的数据集：

Acc  | TranType | PosCancelID
100     1               2
808     1               5
808     1               4
808     2               5
808     2               4<--To be filtered from min calc as it min for (808,1)
813     2               3
813     4               3<--To be filtered from min calc as it min for (813,2)
809     1               3
809     1               4
809     2               3<--To be filtered from min calc as (809,1) uses it
809     2               4 
809     3               4<--To be filtered from min calc as (809,2) uses it

我应该得到：

Acc  | TranType | PosCancelID
100     1               2
808     1               4
808     2               5
813     2               3
809     1               3
809     2               4

SELECT ACC, TranType, min(maxPreceeding) as ActualCancelID 
FROM 
(
   SELECT ACC, TranType,  
          MAX(m.posCancelID) OVER (PARTITION BY m.ACC 
                                  ORDER BY m.TranType, m.posCancelID 
                                  ROWS UNBOUNDED PRECEDING) as maxPreceeding
    FROM MCancel as m
) AS x
GROUP BY ACC, TranType

上面的查询几乎给了我我想要的但没有过滤 acc = 813。所以我知道必须有更好的（实际上应用过滤器来删除以前的最小值）方法。

Answer 1

相当棘手的问题。这是一个递归解决方案：

WITH 
  rcte AS
  ( SELECT TOP (1)
        Acc, TranType, posCancelID,
        CAST('=' + CAST(posCancelID AS VARCHAR(20)) + '=' AS VARCHAR(MAX)) AS IDs 
    FROM
        MCancel
    ORDER BY
        Acc, TranType, posCancelID

    UNION ALL

    SELECT
        Acc, TranType, posCancelID,
        CAST(IDs + CAST(posCancelID AS VARCHAR(20)) + '=' AS VARCHAR(MAX))
    FROM 
      ( SELECT
            m.*, 
            r.IDs, 
            ROW_NUMBER() OVER (ORDER BY m.Acc, m. TranType, m.PosCancelID) AS rn
        FROM
            rcte AS r 
            JOIN MCancel AS m
                ON  (m.Acc = r.Acc AND m.TranType > r.TranType)
                OR  (m.Acc > r.Acc)
        WHERE
            r.IDs NOT LIKE ('%=' + CAST(m.posCancelID AS VARCHAR(20)) + '=%')
      ) AS mc
    WHERE
        rn = 1
  )
SELECT Acc, TranType, posCancelID
FROM rcte
ORDER BY Acc, TranType ;

该解决方案假定 aposCancelID不应在结果集中出现两次。如果要求是它们不应在同一Acc组中出现两次 ，则解决方案需要稍作调整：

WITH rcte AS
  ( SELECT 
      Acc, TranType, posCancelID,
      CAST('=' + CAST(posCancelID AS VARCHAR(20)) + '=' AS VARCHAR(MAX)) AS IDs 
    FROM
      ( SELECT 
          Acc, TranType, posCancelID,
          ROW_NUMBER() OVER (PARTITION BY Acc ORDER BY TranType, PosCancelID) AS rnk
        FROM MCancel
      ) AS f
    WHERE rnk = 1

    UNION ALL

    SELECT
      Acc, TranType, posCancelID,
      CAST(IDs + CAST(posCancelID AS VARCHAR(20)) + '=' AS VARCHAR(MAX))
    FROM
      ( SELECT
          m.*, r.IDs, 
          ROW_NUMBER() OVER (PARTITION BY m.Acc
                             ORDER BY m.TranType, m.PosCancelID) AS rn
        FROM
          rcte AS r 
          JOIN MCancel AS m
              ON  (m.Acc = r.Acc AND m.TranType > r.TranType)
        WHERE
          r.IDs NOT LIKE ('%=' + CAST(m.posCancelID AS VARCHAR(20)) + '=%')
      ) AS mc
    WHERE rn = 1
  )
SELECT Acc, TranType, posCancelID
FROM rcte
ORDER BY Acc, TranType ;

两者都在dbfiddle.uk进行了测试