我刚刚开始使用MERGE语句的OUTPUT运算符，并立即弹出一个问题：
有一张桌子：
CREATE TABLE t1 (
id INT
, somedata VARCHAR(10)
, someotherdata INT
);
...其中包含一些测试数据：
INSERT INTO dbo.t1
VALUES ( 1, 'aaa', 100 ),
( 2, 'bbb', 200 ),
( 3, 'ccc', 300 ),
( 4, 'ccc', 444 ),
( 5, 'rrr', 543 );
还有一个更改数据捕获表，如下所示：
SELECT TOP 0 CONVERT(TINYINT, 0) action -- delete (-1) / update (0) / insert (1)
, CONVERT(BIGINT, 0) execution_id -- from a sequence,
-- to distinguish rows
-- for each MERGE operation
, CONVERT(BIT, 0) row_version -- 0: old, 1: new
, t2.*
INTO t1_cdc
FROM dbo.t1
LEFT OUTER JOIN dbo.t1 t2 ON 0 = 1;
（左外连接保证CDC表中每列的可为空性）
有一个MERGE声明：
;WITH cte_sample_rows
AS ( SELECT 1 id
, 'aaa' somedata
, 100 someotherdata -- same
UNION ALL
SELECT 2
, 'bbb'
, 200 -- same
UNION ALL
SELECT 3
, 'fff'
, 333 -- update
UNION ALL
SELECT 4
, 'ccc'
, 444 -- same
UNION ALL
SELECT 50
, 'xxx'
, 5050 -- insert id=50 / delete id=5
)
MERGE dbo.t1 tgt
USING cte_sample_rows src
ON tgt.id = src.id
WHEN MATCHED THEN
UPDATE SET
tgt.somedata = src.somedata
, tgt.someotherdata = src.someotherdata
WHEN NOT MATCHED BY TARGET THEN
INSERT
VALUES ( src.id
, src.somedata
, src.someotherdata
)
WHEN NOT MATCHED BY SOURCE THEN
DELETE;
现在，我需要捕获每行的旧版本和新版本，作为CDC表中的两个单独行。
我知道可以将OUTPUT子句添加到上面的MERGE语句中，如下所示：
...
OUTPUT
CASE $action
WHEN 'DELETE' THEN -1
WHEN 'UPDATE' THEN 0
ELSE 1 -- 'INSERT'
END AS action
, 1 AS execution_id
, Deleted.id id_old
, Deleted.somedata somedata_old
, Deleted.someotherdata someotherdata_old
, Inserted.id id_new
, Inserted.somedata somedata_new
, Inserted.someotherdata someotherdata_new ;
...但是这会将所有旧值和新值作为单行返回（对于每个输入行）。我需要“取消”它们以获得单独的行：一个用于插入，另一个用于已删除。对于删除，只有已删除，对于插入 - 仅插入，对于更新 - 两者（我将它们与版本号区分开来：0：old，1：new）
我知道我可以填充临时#temp表，然后在单独的步骤中“取消”来自该表的数据，但我正在寻找一步操作。
类似的东西：
...
OUTPUT
(SELECT -1 AS action -- DELETE
, 1 AS execution_id
, 0 AS row_version -- old
, Deleted.*
WHERE $action = 'DELETE'
UNION ALL SELECT 0 AS action -- UPDATE'
, 1 AS execution_id
, 0 AS row_version -- old
, Deleted.*
WHERE $action = 'UPDATE'
UNION ALL SELECT 0 AS action -- UPDATE'
, 1 AS execution_id
, 1 AS row_version -- new
, Inserted.*
WHERE $action = 'UPDATE'
UNION ALL SELECT 1 AS action -- INSERT
, 1 AS execution_id
, 1 AS row_version -- new
, Inserted.*
WHERE $action = 'INSERT')
因此，再次：每个已删除记录的一个CDC行，每个插入记录一个CDC行和每个更新记录的两个CDC行。
有没有办法在一步中实现这一目标？
SQL Server 2012。
答案 0 :(得分：0)
我知道我可以填充一个临时的#temp表，然后＆＃34; unpivot＆＃34;来自该表的数据在单独的步骤中，但我正在寻找一步操作。
这是一个非常好的问题。 MERGE声明是 Swiss-Army Knife ，所以我尝试了这种方法：
INSERT INTO
SELECT ...
FROM (
MERGE
...
OUTPUT
...
)
但是有一个问题，我不能将行分成两行，因为我无法使用GROUP BY, CROSS JOIN, CROSS APPLY, nested subquery, PIVOT, UNPIVOT, UNION ALL, cte。每次尝试都以：
A nested INSERT, UPDATE, DELETE, or MERGE statement
is not allowed as the table source of a PIVOT or UNPIVOT operator.
A nested INSERT, UPDATE, DELETE, or MERGE statement is not allowed
in a SELECT statement that is not the immediate source of rows
for an INSERT statement.
A nested INSERT, UPDATE, DELETE, or MERGE statement is not allowed
on either side of a JOIN or APPLY operator.
The GROUP BY clause is not allowed when the FROM clause contains
a nested INSERT, UPDATE, DELETE, or MERGE statement.
我得到的最接近的结果集是（对于UPDATE，您将只获得新的/旧值）：
;INSERT INTO #cdc(action, execution_id, rowversion,id, somedata, someotherdata)
SELECT
[action]
,[execution_id]
,[rowversion] = CASE action WHEN -1 THEN 0
WHEN 1 THEN 1
WHEN 0 THEN 1
END
,[id] = CASE action WHEN -1 THEN id_old
WHEN 1 THEN id_new
WHEN 0 THEN id_old
END
,[somedata] = CASE action WHEN -1 THEN somedata_old
WHEN 1 THEN somedata_new
WHEN 0 THEN somedata_old
END
,[someotherdata] = CASE action WHEN -1 THEN someotherdata_old
WHEN 1 THEN someotherdata_new
WHEN 0 THEN someotherdata_old
END
FROM (
MERGE #t1 tgt
USING (VALUES (1, 'aaa', 100),(2, 'bbb', 200),
(3, 'fff', 333), (4, 'ccc', 4444),
(50, 'xxx', 5050)) AS src(id, somedata, someotherdata)
ON tgt.id = src.id
WHEN MATCHED AND NOT EXISTS (SELECT src.somedata, src.someotherdata
INTERSECT
SELECT tgt.somedata, tgt.someotherdata)
THEN
UPDATE SET
tgt.somedata = src.somedata
, tgt.someotherdata = src.someotherdata
WHEN NOT MATCHED BY TARGET THEN
INSERT
VALUES ( src.id
, src.somedata
, src.someotherdata
)
WHEN NOT MATCHED BY SOURCE THEN
DELETE
OUTPUT
CASE $action
WHEN 'DELETE' THEN -1
WHEN 'UPDATE' THEN 0
ELSE 1 -- 'INSERT'
END AS action
, 1 AS execution_id
, Inserted.*
, Deleted.*)
AS m(action, execution_id, id_new, somedata_new, someotherdata_new
,id_old, somedata_old, someotherdata_old);
的 LiveDemo
另请注意，在当前表单中，您标记为MERGE的{{1}}行将UPDATE。我添加了额外的条件来检查是否至少更改了一列。
WHEN MATCHED AND NOT EXISTS (SELECT src.somedata, src.someotherdata
INTERSECT
SELECT tgt.somedata, tgt.someotherdata)
THEN
另一种方法是在UPDATE/INSERT/DELETE的目标表上添加触发器，这些触发器将记录插入cdc表并忘记OUTPUT中的MERGE子句。