复杂的 SQL MERGE/PIVOT 未按预期排序

Question

我有以下两张表； ICD 使用

创建

CREATE TABLE ICD ([E] VARCHAR(8), [I] VARCHAR(8));
GO
INSERT INTO ICD ([E], [I])
VALUES
    ('1', 'B1'),
    ('1', 'A1'),
    ('1', 'C23q'),
    ('1', '32Dq'),
    ('2', 'FFFq'),
    ('2', 'ERRE'),
    ('2', 'E'),
    ('3', 'WQWW'),
    ('3', 'WQ'),
    ('4', 'AAAA');
GO

其中O 给出了将列I 插入另一个表的顺序。第二个表是Episode，这是使用

创建的

CREATE TABLE Episode ([EpiNum] VARCHAR(4), [SomeField] VARCHAR(4), 
                      [DX1] VARCHAR(10), [DX2] VARCHAR(10), 
                      [DX3] VARCHAR(10), [DX4] VARCHAR(10));
GO
INSERT INTO Episode ([EpiNum], [SomeField], [DX1], [DX2], [DX3], [DX4])
VALUES
    ('1', 'Test', '', '', '', ''),
    ('2', 'Test', '', '', '', ''), 
    ('3', 'Test', '', '', '', '');
GO

所以我的目标是将ICD 中I 列的值插入DXn 字段，具体取决于是否有要插入的值。忍受我；我有这个工作使用

MERGE INTO [Episode] AS T 
USING (
   SELECT [EpiNum], [1], [2], [3], [4] 
   FROM 
      ( 
          SELECT *, ROW_NUMBER() 
            OVER (PARTITION BY [EpiNum] ORDER BY [I]) AS rn
          FROM [ICD] 
      ) AS p
    PIVOT 
   (
      MAX([I]) 
      FOR rn IN ([1], [2], [3], [4])) AS p2
   ) AS S
ON T.[EpiNum] = S.[EpiNum] 
WHEN MATCHED THEN
   UPDATE SET [DX1] = S.[1], [DX2] = S.[2], [DX3] = S.[3], [DX4] = S.[4];

这会产生一个看起来像

的Episode 表

EpiNum  Test    DX1     DX2     DX3     DX4
'1'     'Test'  '32Dq'  'A1'    'B2'    'C23q'  
'2'     'Test'  'E'     'ERRE'  'FFFq'  null
'3'     'Test'  'WQ'    'WQWW'  null    null
'4'     'Test'  'AAAA'  null    null    null

太棒了！等等...不太好，PIVOT/MERGE 中的ORDER BY 正在破坏数据的原始顺序。我试图通过添加另一列 O 来解决这个问题，该列包含我想要的顺序，所以 ICD 由

定义

CREATE TABLE ICD ([E] VARCHAR(8), [I] VARCHAR(8), [O] INT);
GO
INSERT INTO ICD ([E], [I], [O])
VALUES
    ('1', 'B1.2', 1),
    ('1', 'A1.2', 2),
    ('1', 'C23q', 3),
    ('1', '32Dq', 4),
    ('2', 'FFFq', 1),
    ('2', 'ERRE', 2),
    ('3', 'WQWW', 1),
    ('4', 'AAAA', 1);
GO

我正在尝试使用ORDER BY 列O

MERGE INTO [Episode] AS T 
USING (
    SELECT [E], [1], [2], [3], [4] 
    FROM 
        ( 
             SELECT *, ROW_NUMBER() 
               OVER (PARTITION BY [E] ORDER BY [O]) AS rn
             FROM [ICD] 
        ) AS p
    PIVOT 
    (
       MAX([I]) 
       FOR rn IN ([1], [2], [3], [4])) AS p2
    ) AS S
ON T.[E] = S.[E] 
WHEN MATCHED THEN
    UPDATE SET [DX1] = S.[1], [DX2] = S.[2], [DX3] = S.[3], [DX4] = S.[4];

但我收到以下消息

MERGE 语句多次尝试更新或删除同一行。当目标行匹配多个源行时会发生这种情况。 MERGE 语句不能多次 UPDATE/DELETE 目标表的同一行。优化 ON 子句以确保目标行最多匹配一个源行，或使用 GROUP BY 子句对源行进行分组。

如何修改上述MERGE/PIVOT 才能正确订购？

感谢您的宝贵时间

Answer 1

好吧，您的示例运行不正确，因为您在创建表和查询之间混合了 [E] 和 Epinum 列。但是假设这些实际上是相同的列，那么您的第二个查询不起作用，因为您实际上并没有在 [E] 列上进行分组。试试这个：

MERGE INTO [Episode] AS T 
USING (
    SELECT [E], [1], [2], [3], [4] 
    FROM 
        ( 
             SELECT [E], [I], ROW_NUMBER() 
               OVER (PARTITION BY [E] ORDER BY [O]) AS rn
             FROM [ICD] 
        ) AS p
    PIVOT 
    (
       MAX([I]) 
       FOR rn IN ([1], [2], [3], [4])) AS p2
    ) AS S
ON T.[Epinum] = S.[E] 
WHEN MATCHED THEN
    UPDATE SET [DX1] = S.[1], [DX2] = S.[2], [DX3] = S.[3], [DX4] = S.[4];

现在，我真的不明白您为什么使用 ROW_NUMBER() 而不是简单地使用 [O] 列（值对我来说似乎相同），但这两种方式都有效。