How to delete WITH CONDITION in MERGE-clause? - sql

Question

How can I have a condition in the delete-clause of a merge-statement ?
I want to have user-group-mappings deleted (for the respective user), if the mapping is not present in the source data.
E.g. in the example below, I can do this with:
WHEN NOT MATCHED BY SOURCE
AND A.MUG_USR = 1 -- <== this is the ID of the user I want mappings deleted, but i don't want ONLY ONE id and put it here manually
-- What I actually want here is: WHEN NOT MATCHED BY SOURCE AND A.MUG_USR = CTE.MUG_USR
AND A.MUG_FromSAML = 1
THEN DELETE
But this only works if I insert ONE user at a time and fetch its ID (MUG_USR) manually.
In the example data below, it is critical that the mapping for user 2 and 3 won't get deleted when the mappings for user 1 are deleted.
Note:
I don't want to duplicate the query or create a temporary table (or table variable).
I also don't want a separate delete-statement after the insert statement (see 1)
Duplicating the query or creating a separate delete-statement via temp-table after the insert is something I can do myselfs - I do not need any help with that.
Is this at all possible ?
I would have expected this to work, since I already have
USING CTE ON A.MUG_USR = CTE.MUG_USR AND A.MUG_Group = CTE.MUG_Group
in the using-clause, therefore I expected the delete to be partitioned by this using clause, just like the update.
Example data:
/*
-- DROP TABLE [dbo].Map_User_Groups
CREATE TABLE [dbo].Map_User_Groups
(
MUG_UID uniqueidentifier NOT NULL
,MUG_USR int NULL
,MUG_Group int NULL
,MUG_FromSAML bit NULL
);
*/
TRUNCATE TABLE dbo.Map_User_Groups ;
INSERT INTO Map_User_Groups
(
MUG_UID
,MUG_USR
,MUG_Group
,MUG_FromSAML
)
SELECT
NEWID() AS MUG_UID
,1 AS MUG_USR
,1000 AS MUG_Group
,1 AS MUG_FromSAML
;
INSERT INTO Map_User_Groups
(
MUG_UID
,MUG_USR
,MUG_Group
,MUG_FromSAML
)
SELECT
NEWID() AS MUG_UID
,2 AS MUG_USR
,2000 AS MUG_Group
,1 AS MUG_FromSAML
;
INSERT INTO Map_User_Groups
(
MUG_UID
,MUG_USR
,MUG_Group
,MUG_FromSAML
)
SELECT
NEWID() AS MUG_UID
,3 AS MUG_USR
,3000 AS MUG_Group
,1 AS MUG_FromSAML
;
INSERT INTO Map_User_Groups
(
MUG_UID
,MUG_USR
,MUG_Group
,MUG_FromSAML
)
SELECT
NEWID() AS MUG_UID
,4 AS MUG_USR
,4000 AS MUG_Group
,NULL AS MUG_FromSAML
;
INSERT INTO Map_User_Groups
(
MUG_UID
,MUG_USR
,MUG_Group
,MUG_FromSAML
)
SELECT
NEWID() AS MUG_UID
,5 AS MUG_USR
,5000 AS MUG_Group
,NULL AS MUG_FromSAML
;
INSERT INTO Map_User_Groups
(
MUG_UID
,MUG_USR
,MUG_Group
,MUG_FromSAML
)
SELECT
NEWID() AS MUG_UID
,6 AS MUG_USR
,6000 AS MUG_Group
,0 AS MUG_FromSAML
;
INSERT INTO Map_User_Groups
(
MUG_UID
,MUG_USR
,MUG_Group
,MUG_FromSAML
)
SELECT
NEWID() AS MUG_UID
,7 AS MUG_USR
,7000 AS MUG_Group
,0 AS MUG_FromSAML
;
SELECT * FROM dbo.Map_User_Groups ORDER BY MUG_USR;
;WITH CTE AS
(
SELECT
MUG_UID
,MUG_USR
,MUG_Group
,MUG_FromSAML
FROM
(
SELECT
NEWID() AS MUG_UID
,1 AS MUG_USR
,7000 AS MUG_Group
,1 AS MUG_FromSAML
UNION
SELECT
NEWID() AS MUG_UID
,33 AS MUG_USR
,33000 AS MUG_Group
,1 AS MUG_FromSAML
) AS tSource
WHERE (1=1)
-- AND EXISTS ( SELECT * FROM dbo.Users AS tAlias1 WHERE (1=1) AND tAlias1.USR_ID = tSource.MUG_USR )
-- AND EXISTS ( SELECT * FROM dbo.Groups AS tAlias2 WHERE (1=1) AND tAlias2.GRP_ID = tSource.MUG_Group )
) -- End CTE
-- SELECT * FROM CTE
MERGE INTO dbo.Map_User_Groups AS A
USING CTE ON A.MUG_USR = CTE.MUG_USR AND A.MUG_Group = CTE.MUG_Group
WHEN MATCHED THEN
UPDATE
SET A.MUG_FromSAML = CTE.MUG_FromSAML
WHEN NOT MATCHED BY TARGET THEN
INSERT
(
MUG_UID
,MUG_USR
,MUG_Group
,MUG_FromSAML
)
VALUES
(
CTE.MUG_UID
,CTE.MUG_USR
,CTE.MUG_Group
,CTE.MUG_FromSAML
)
WHEN NOT MATCHED BY SOURCE
-- AND A.MUG_USR = CTE.MUG_USR -- <== i want something like this
-- AND A.MUG_USR = 1 -- <== this i don't want
AND A.MUG_FromSAML = 1
THEN DELETE
;
SELECT * FROM dbo.Map_User_Groups ORDER BY MUG_USR;

Answer 1

What you do is instead of using MERGE Map_User_Groups you do a CTE2 subquery to only get users you wanna potentially delete:
...
, cte2 AS (
select *
from dbo.Map_User_Groups AS A
where exists(
select 1 from cte c where c.mug_usr = a.mug_usr
)
)
MERGE INTO cte2 AS A
...
This way, you're only working with correct subset.
I use WHERE EXISTS instead of JOIN to avoid "contaminating" target data by duplicate and other merge-unfriendly issues.
With that being said, in my opinion, and i'm probably wrong, one should avoid MERGE if you know what's best for you, because of all the various bugs and quirks of it.