Question

In my EF4 program, I have an Applicant and Application table. Multiple instances of the program runs periodically to create application for applicants based on some business logic. In Application table I cannot have more than one Submitted/BeingSubmitted records for an Applicant.

So here’s the piece of code which checks if there’s an Submitted/BeingSubmitted application and inserts it. It is run inside a foreach loop for a list of Applicants.

public Application SaveApplication(Int32 applicantId)
    {
        using (TransactionScope txScope = new TransactionScope(TransactionScopeOption.RequiresNew))
        {
            if (ApplicantHasPendingApplication(applicantId))
                return null;

            Application app = null;
            try
            {
                app = new Application()
                {
                    // Create the object...
                };

                _unitOfWork.DisclosureApplications.Add(app);
                _unitOfWork.Commit();
                _unitOfWork.Refresh(app); // We save and refresh it to get the Id.

                txScope.Complete();
            }
            catch (UpdateException ex)
            {
                // We get an Update exception here when multiple instances tries to insert Application.
            }

            return app;
        }
    }

Above piece of code prevents insertion of duplicate records apart from the fact that it throws UpdateException while running multiple instances of the program. If I swallow that exception and carry on then everything is fine.

However, I tried to test/run above code in parallel but it inserts duplicate records in database.

Parallel.Invoke(
            () => CreateApplications("Parallel Instance 1"),
            () => CreateApplications("Parallel Instance 2"));

private void CreateApplications(String dummyInstanceName)
{
   var unitOfWork = new SqlUnitOfWork();
   var applicants = unitOfWork.Applicants.FindAll().Take(100).ToList();

   var facade = new ProviderFacade(unitOfWork, new Log4NetLogger(dummyInstanceName));

   foreach (Applicant applicant in applicants)
            {
                facade.ApplicationProvider.SaveApplication(applicant.applicantID);
            }
}

In above piece of code, it throws UpdateException and inserts multiple Application row for an Applicant.

Note that, the table only has a surrogate primary key and no other unique constraints.

My question is: Why the TransactionScope inserts duplicate rows by running it in Parallel.Invoke but not when I fire off multiple instances of the program? What would be a sound approach achieving it?

Update: The ctor of SqlUnitOfWork is

public SqlUnitOfWork()
    {
        _context = new MyEntities();
    }

The ctor of MyEntities is generated by EF –

    public const string ConnectionString = "name=Entities";
    public const string ContainerName = "Entities";

    public TPIEntities() : base(ConnectionString, ContainerName)
    {
        this.ContextOptions.LazyLoadingEnabled = true;
    }

Thanks.

Answer 1

This is a classic race condition. Both transactions check to see if a pre-existing Application exists. Both find that there is none. Then, both try to insert.

You need to synchronize your checking, either by locking in the application or by taking a SQL Server UPDLOCK, HOLDLOCK when checking whether the Application exists.