You need to do this in transaction to ensure two simultaneous clients won't insert same fieldValue twice:
SET TRANSACTION ISOLATION LEVEL SERIALIZABLE
BEGIN TRANSACTION
DECLARE @id AS INT
SELECT @id = tableId FROM table WHERE fieldValue=@newValue
IF @id IS NULL
BEGIN
INSERT INTO table (fieldValue) VALUES (@newValue)
SELECT @id = SCOPE_IDENTITY()
END
SELECT @id
COMMIT TRANSACTION
you can also use Double-checked locking to reduce locking overhead
DECLARE @id AS INT
SELECT @id = tableID FROM table (NOLOCK) WHERE fieldValue=@newValue
IF @id IS NULL
BEGIN
SET TRANSACTION ISOLATION LEVEL SERIALIZABLE
BEGIN TRANSACTION
SELECT @id = tableID FROM table WHERE fieldValue=@newValue
IF @id IS NULL
BEGIN
INSERT INTO table (fieldValue) VALUES (@newValue)
SELECT @id = SCOPE_IDENTITY()
END
COMMIT TRANSACTION
END
SELECT @id
As for why ISOLATION LEVEL SERIALIZABLE is necessary, when you are inside a serializable transaction, the first SELECT that hits the table creates a range lock covering the place where the record should be, so nobody else can insert the same record until this transaction ends.
Without ISOLATION LEVEL SERIALIZABLE, the default isolation level (READ COMMITTED) would not lock the table at read time, so between SELECT and UPDATE, somebody would still be able to insert. Transactions with READ COMMITTED isolation level do not cause SELECT to lock. Transactions with REPEATABLE READS lock the record (if found) but not the gap.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…