TSQL Try / Catch within Transaction or vice versa?

Question

I\'m writing a script that will delete records from a number of tables, but before it deletes it must return a count for a user to confirm before committing.

This is

Answer 1

Never wait for an end user to commit the transaction, unless it's a single-user mode database.

In short, it's about blocking. Your transaction will take some exclusive locks on resources being updated, and will hold on to those lock untill the transaction is ended (committed or rolled back). Nobody will be able to touch those rows. There are some different problems if snapshot isolation is used with version store cleanup.

Better to first issue a select query to determine a number of qualifying rows, present this to the end user, and after he confirms do the actual delete.

Answer 2

Only open a transaction once you are inside the TRY block and just before the actual statement, and commit it straightaway. Do not wait for your control to go to the end of the batch to commit your transactions.

If something goes wrong while you are in the TRY block and you have opened a transaction, the control will jump to the CATCH block. Simply rollback your transaction there and do other error handling as required.

I have added a little check for any open transaction using @@TRANCOUNT function before actually rolling back the transaction. It doesn't really make much sense in this scenario. It is more useful when you are doing some validations checks in your TRY block before you open a transaction like checking param values and other stuff and raising error in the TRY block if any of the validation checks fail. In that case, the control will jump to the CATCH block without even opening a transaction. There you can check for any open transaction and rollback if there are any open ones. In your case, you really don't need to check for any open transaction as you will not enter the CATCH block unless something goes wrong inside your transaction.

Do not ask after you have executed the DELETE operation whether it needs to be committed or rolled back; do all these validation before opening the transaction. Once a transaction is opened, commit it straightaway and in case of any errors, do error handling (you are doing a good job by getting detailed info by using almost all of the error functions).

BEGIN TRY

  BEGIN TRANSACTION SCHEDULEDELETE
    DELETE   -- delete commands full SQL cut out
    DELETE   -- delete commands full SQL cut out
    DELETE   -- delete commands full SQL cut out
 COMMIT TRANSACTION SCHEDULEDELETE
    PRINT 'X rows deleted. Operation Successful Tara.' --calculation cut out.
END TRY

BEGIN CATCH 
  IF (@@TRANCOUNT > 0)
   BEGIN
      ROLLBACK TRANSACTION SCHEDULEDELETE
      PRINT 'Error detected, all changes reversed'
   END 
    SELECT
        ERROR_NUMBER() AS ErrorNumber,
        ERROR_SEVERITY() AS ErrorSeverity,
        ERROR_STATE() AS ErrorState,
        ERROR_PROCEDURE() AS ErrorProcedure,
        ERROR_LINE() AS ErrorLine,
        ERROR_MESSAGE() AS ErrorMessage
END CATCH

Answer 3

In addition to the good advice by M.Ali and dean above, a little bit help for those looking to use the new(er) TRY CATCH THROW paradigm in SQL SERVER:

(I couldn't easily find the complete syntax, so adding it here)

GIST : HERE

Sample stored procedure code here (from my gist):

SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO
 
CREATE PROC [dbo].[pr_ins_test]
@CompanyID INT
AS
 
SET NOCOUNT ON
 
BEGIN
 
    DECLARE @PreviousConfigID INT
    
    BEGIN TRY
        BEGIN TRANSACTION MYTRAN; -- Give the transaction a name
        SELECT 1/0  -- Generates divide by zero error causing control to jump into catch
 
        PRINT '>> COMMITING'
        COMMIT TRANSACTION MYTRAN;
    END TRY
    BEGIN CATCH
        IF @@TRANCOUNT > 0
        BEGIN 
            PRINT '>> ROLLING BACK'
            ROLLBACK TRANSACTION MYTRAN; -- The semi-colon is required (at least in SQL 2012)
            
            
        END; -- I had to put a semicolon to avoid error near THROW
        THROW
    END CATCH
END