Question

I’m trying to optimize a long transaction and I’ve seen that the following is done quite a few times:

Declare @myCursor CURSOR FAST_FORWARD FOR
SELECT field1, MIN(COALESCE(field2, -2)) FROM MyTable tempfact
LEFT JOIN MyTable sd
ON tempfact.ID = sd.ID AND sd.TransactionId = @transactionId
WHERE tempfact.SomeField IS NULL
AND tempfact.TransactionId = @transactionId
GROUP BY tempfact.field1

OPEN @myCursor

FETCH NEXT FROM @myCursor INTO @field1Variable, @field2Variable

WHILE @@FETCH_STATUS = 0
BEGIN
   EXEC USP_SOME_PROC @field1Variable, @field2Variable
   FETCH NEXT FROM @myCursor INTO @field1Variable, @field2Variable
END

CLOSE @myCursor
DEALLOCATE @myCursor

The code for the USP_SOME_PROC sproc is as follows:

IF NOT EXISTS (SELECT * FROM SomeTable WHERE Field1 = @field1)
BEGIN
   INSERT INTO SomeTable (Field1, Field2)
   VALUES (@field1, @field2)
END

Like I mentioned this is done in quite a few places, tables and fields involved are different but the idea remains the same, and I’m sure that there might be a way to increase the performance of these sprocs if cursors are not used and probably by making this transaction faster an issue that we’re having with a deadlock (a subject for another post) might be solved.

Answer 1

You can use MERGE for this

;WITH Source AS
(
SELECT field1,
       MIN(COALESCE(field2, -2)) as field2
FROM   MyTable tempfact
       LEFT JOIN MyTable sd
         ON tempfact.ID = sd.ID
            AND sd.TransactionId = @transactionId
WHERE  tempfact.SomeField IS NULL
       AND tempfact.TransactionId = @transactionId
GROUP  BY tempfact.field1  
)
MERGE SomeTable AS T
USING Source S
ON (T.Field1 = S.Field1)


WHEN NOT MATCHED BY TARGET THEN
    INSERT (Field1, Field2)
    VALUES (field1, field2)
    ;