Question

I am doing a financial application in which I am expecting a data concurrency issue.

Suppose there is an account ABC which has $500 in it. User from web can transfer these funds to other accounts. This will involve 2 steps 1st checking availability of funds and 2nd transferring. I am making a transaction and doing both acts in it.

Problem is when in a time (say Time1) there are 2 or 3 seprate requests for transferring (say transaction1,transaction2, transaction3) same amount. Now committed available amount is $500. If all translations starts in same time, all will test is amount ($500) available ? which will true and next statement will transfer funds to other account.

I have read about Transaction isolation levels but I couldn’t decide which isolation level I should use, actually I am confuse in its understanding. Please help me.

Thanks

Answer 1

The aim is to prevent another process reading the balance but minimise blocking for other users. So use the “table as a queue” type locks thus:

SET XACT_ABORT, NOCOUNT ON;
BEGIN TRY

    BEGIN TRANSACTION

    SELECT @balance = Balance
    FROM SomeTable WITH (ROWLOCK, HOLDLOCK, UPDLOCK)
    WHERE Account = 'ABC'

    --some checks

    UPDATE ...

    COMMIT TRANSACTION
END TRY
BEGIN CATCH
  ...
END CATCH

The alternative is to do it in one, which is more feasible if there is one table involved.
The CROSS JOIN is a test to

SET XACT_ABORT, NOCOUNT ON;
BEGIN TRY

    --BEGIN TRANSACTION

    UPDATE SomeTable WITH (ROWLOCK, HOLDLOCK, UPDLOCK)
    SET Balance = Balance - @request
    WHERE
       ST.Account = 'ABC' AND Balance > @request;
    IF @@ROWCOUNT <> 1
       RAISERROR ('Not enough in account', 16, 1);

    --COMMIT TRANSACTION
END TRY
BEGIN CATCH
  ...
END CATCH