Question

我需要在执行GROUP BY时找出值来自的行ID。

首先创建并填充“播放对象”

-- vertical coalesce with row-source of value :) instead of column-source of value as in horizontal coalesce :)
CREATE TABLE #TBL(ID INT, GR INT, A INT, B INT)
INSERT INTO #TBL
SELECT 1,1,1,NULL
UNION ALL SELECT 2,1,NULL,2
UNION ALL SELECT 3,1,NULL,3
UNION ALL SELECT 4,2,2,NULL
UNION ALL SELECT 5,2,NULL,NULL
UNION ALL SELECT 6,2,6,NULL

选择最小值很简单：

SELECT GR, MIN(A) A, MIN(B) B,
    'Which first available row did A come from?' A_ID, 
    'Which first available row did B come from?' B_ID 
FROM #TBL
GROUP BY GR

但回答问题不是！

我尝试过子查询，但由于值来自不同的行，因此不起作用：

SELECT a.*
from #TBL a
    join (
    SELECT GR, MIN(A) A, MIN(B) B,
        'Which row did A come from?' A_ID, 
        'Which row did B come from?' B_ID
    FROM #TBL
    GROUP BY GR) b on a.GR = b.GR and a.A = b.A and a.B = b.B
--DROP TABLE #TBL

请帮忙。

Answer 1

我不确定它是否会执行连接/外部应用（我认为它会），但这将有效：

WITH CTE AS
(   SELECT  GR, 
            A, 
            B, 
            ID, 
            MinA = MIN(A) OVER(PARTITION BY GR), 
            MinB = MIN(B) OVER(PARTITION BY GR)
    FROM    #TBL
)
SELECT  GR,
        A = MIN(A),
        B = MIN(B),
        A_ID = MIN(CASE WHEN MinA = A THEN ID END),
        B_ID = MIN(CASE WHEN MinB = B THEN ID END)
FROM    CTE
GROUP BY GR;

在这个小样本上，统计数据表明这一点表现更好：

使用聚合

（估计）查询成本13％

表'工作台'。扫描计数3，逻辑读取21，物理读取0，预读读取0，lob逻辑读取0，lob物理读取0，lob预读读取0。

使用外部申请

（估计）查询成本87％

表'工作台'。扫描计数16，逻辑读取42，物理读取0，预读取读取0，lob逻辑读取0，lob物理读取0，lob预读读取0。

但是你可能会发现索引的真实例子会有所不同。

Answer 2

这应该做：

SELECT  DISTINCT 
        T1.GR,
        T2.A,
        T3.B,
        T2.ID A_ID,
        T3.ID B_ID
FROM #TBL T1
OUTER APPLY (SELECT TOP 1 *
             FROM #TBL
             WHERE GR = T1.GR
             AND A IS NOT NULL
             ORDER BY A, ID) T2
OUTER APPLY (SELECT TOP 1 *
             FROM #TBL
             WHERE GR = T1.GR
             AND B IS NOT NULL
             ORDER BY B, ID) T3

结果：

╔════╦═══╦══════╦══════╦══════╗
║ GR ║ A ║  B   ║ A_ID ║ B_ID ║
╠════╬═══╬══════╬══════╬══════╣
║  1 ║ 1 ║ 2    ║    1 ║ 2    ║
║  2 ║ 2 ║ NULL ║    4 ║ NULL ║
╚════╩═══╩══════╩══════╩══════╝

Answer 3

;WITH CTEA AS
(   SELECT  ID, GR, A, ROW_NUMBER() OVER(PARTITION BY GR ORDER BY A) AS 'RowNum'
    FROM    #TBL where A is not null
),    CTEB AS
(   SELECT  ID, GR, B, ROW_NUMBER() OVER(PARTITION BY GR ORDER BY B) AS 'RowNum'
    FROM    #TBL where B  is not null
)
select  CTEA.GR, CTEA.A, CTEA.ID, CTEB.B, CTEB.ID
  from  CTEA 
  jull  outer join  CTEB
    on  CTEA.GR = CTEB.GR
   and  CTEA.RowNum = 1 
   and  CTEB.RowNum = 1

哪行行最小值来自（垂直合并）

3 个答案: