[wildfly-dev] What should the EJB3 StatefulSessionSynchronizationInterceptor.processInvocation() do for transactions that are already committed or rolled back

[Scott Marlow]

On 08/13/2014 03:07 PM, Scott Marlow wrote:
> I created WFLY-3731 for a "No transaction is running" ISE [1], that is
> thrown when
> org.jboss.as.ejb3.component.stateful.StatefulSessionSynchronizationInterceptor.processInvocation()
> checks if the current transaction is not null and not STATUS_COMMITTED.
>
> One obvious fix for WFLY-3731, would be to change
> StatefulSessionSynchronizationInterceptor.processInvocation() to also
> check for STATUS_ROLLEDBACK which would help my current case.  Doing so

Checking for STATUS_ROLLEDBACK, helps my test case somewhat.  I am still 
seeing the below mention (WFLY-3730) JPA container failure as a result 
of the transaction status changing from active, to rolled back, after I 
called transaction.getStatus() (which returned STATUS_ACTIVE).

I'm not quite sure what to do with the "IllegalStateException" that is 
thrown in either case (EJB or JPA container), since the transaction 
status can change after StatefulSessionSynchronizationInterceptor checks 
the status (e.g. in the transaction time out case that my testing is 
purposely hitting).

The other related issue that I'm looking at, is how to better handle 
transaction time out (to avoid unexpected state or error conditions).  I 
think that we might be missing an opportunity to coordinate the actions 
that occur when a background thread rolls back a transaction.

> might defer clearing the transaction from the current thread, until we
> later reached EJB container code (not really sure though, if we will
> clear the ended tx).
>
> But, what does it mean exactly for
> StatefulSessionSynchronizationInterceptor.processInvocation() to see an
> ended transaction?  Seems to me that we missed our chance to register
> the StatefulSessionSynchronization, so we won't run the
> StatefulSessionSynchronization.beforeCompletion() and
> StatefulSessionSynchronization.afterCompletion() methods which I think
> skips other logic (beforeCompletionMethod/afterCompletionMethod).
>
> Currently, it looks like we just continue with the invocation as if
> there is no current transaction (SFSB level lock appears to be released
> after the invocation instead of after the transaction ends).  Which
> makes some sense since the transaction already ended.
>
> When running the test case [2], I also see a JPA (EE) container level
> failure as well which is reported as WFLY-3730.
>
> Scott
>
> [1] http://fpaste.org/125252 show the exception that I see when the
> transaction times out in a background (tm reaper) thread.
>
> [2] test case is pushed to
> https://github.com/scottmarlow/wildfly/tree/transactiontimeout_clientut
> _______________________________________________
> wildfly-dev mailing list
> wildfly-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/wildfly-dev
>