SQL Server before trigger - sql-server

Question

I need to control table values uniqueness. It cannot be done by an index or a constraint (error message must show data from another table). I thought of after trigger but since it fires after the insert the below trigger will fire even if values are unique.
--table
CREATE TABLE Names (Id IDENTITY(1,1) NOT NULL, Name VARCHAR(20) NOT NULL)
--first record
INSERT INTO Names VALUES ('John')
--trigger
CREATE TRIGGER [dbo].[Names_Insert_Trigger]
ON [dbo].[Names]
FOR INSERT, UPDATE
AS
SET NOCOUNT ON
IF EXISTS (SELECT Name
FROM inserted
WHERE EXISTS (SELECT * FROM Names N JOIN inserted ON N.Name=inserted.Name))
BEGIN
RAISERROR('This name is already registered in file XYZ.', 16, 1)
ROLLBACK TRAN
SET NOCOUNT OFF
RETURN
END
SET NOCOUNT OFF
--I add another record with different value and the trigger fires
INSERT INTO Names VALUES ('Steven')
I also thought of an instead of insert trigger but the actual table has identity set and will likely get new columns in the future which would require updating the trigger code at each change so I can't use the below code:
CREATE TRIGGER [dbo].[Names_Insert_Trigger]
ON [dbo].[Names]
INSTEAD OF INSERT
AS
SET NOCOUNT ON
IF EXISTS (SELECT Name
FROM inserted
WHERE EXISTS (SELECT * FROM Names N JOIN inserted ON N.Name=inserted.Name))
BEGIN
RAISERROR('This name is already registed in file XYZ.', 16, 1)
ROLLBACK TRAN
SET NOCOUNT OFF
RETURN
END
ELSE
INSERT INTO Names
SELECT * FROM inserted
SET NOCOUNT OFF
Any ideas how to solve it?
Regards,
Przemek

Answer 1

You can use an after trigger. Just use COUNT instead of EXISTS. You should still have a non-unique index on name to optimize performance and concurrency.
IF (SELECT COUNT(*)
FROM inserted AS i
JOIN dbo.Names AS N ON
N.Name = i.Name
GROUP BY N.Name
) > 1
BEGIN
RAISERROR...
END;

Answer 2

The real solution is to use the UNIQUE constraint to this problem, it's designed to solve it and it's much more performant and safer than a trigger for this usage. The error message is better built client-side and ignore the server genereated one, save for determining the exact reason.
But if you really want to follow the trigger route, use the AFTER version, but fix the query for detect duplicates:
SELECT * FROM (
SELECT Name
FROM Names
WHERE id NOT IN (SELECT id FROM inserted)
) PreviousNames
INNER JOIN inserted ON PreviousNames.Name = inserted.Name
(I'm just showing the query to check duplication that goes into the IF EXIST instruction, not the whole trigger).
It begins by creating a subquery that gets the names NOT being inserted (so that you don't get a false positive), then simply joins again to inserted to check if any value is in both tables.
There is an additional problem that can happen when using SNAPSHOT issolation level. In this mode, the trigger will NOT see the changes made by other connections, nor they'll be blocked until the trigger ends. I'm not quite familiar with the details, but will leave this article as reference and possible solutions: https://sqlserverfast.com/?s=snapshot+integrity