How to notify client of CMT TX Timeout

[markus78]

Hi,

Our previous application server (IBM Websphere.. *shruggs*) it had some pretty handy ejb spec violations, whenever we would get a CMT TX timeout an exception was raised letting the client know that the TX timedout.

JBoss7 does not do this, and after reading the ejb3.1 specification I'm pretty sure JBoss7 follows the spec to the letter so there should not be any exception due to TX timeout.

What techniques are available to notify the client that the TX timed out? Do we have to implement asyncronous calls from the client and add timeout checks on the client?

Any ideas are welcome.

/Markus

[nickarls]

Hmm. Haven't tested but at which point does an AFTER_FAILURE transactional CDI event fire? Or can you just use a normal interceptor and try digging out the underlying cause?

[markus78]

CDI Events would have been a nice solution, but I can't quite get it to work

15:20:30,560 INFO  [stdout] (http-localhost/127.0.0.1:8080-1) Waiting for timeout 20 sec

15:20:40,552 WARN  [com.arjuna.ats.arjuna] (Transaction Reaper) ARJUNA012117: TransactionReaper::check timeout for TX 0:ffff82ef5225:-5dbeb42b:51111419:15 in state  RUN

15:20:40,555 WARN  [com.arjuna.ats.arjuna] (Transaction Reaper Worker 0) ARJUNA012095: Abort of action id 0:ffff82ef5225:-5dbeb42b:51111419:15 invoked while multiple threads active within it.

15:20:40,556 WARN  [com.arjuna.ats.arjuna] (Transaction Reaper Worker 0) ARJUNA012108: CheckedAction::check - atomic action 0:ffff82ef5225:-5dbeb42b:51111419:15 aborting with 1 threads active!

15:20:40,559 ERROR [org.jboss.weld.Event] (Transaction Reaper Worker 0) WELD-000401 Failure while notifying an observer of event com.my.ejb.HelloEvent@e6e445d

Very simple code...

package com.my.ejb;


import javax.ejb.Stateless;
import javax.enterprise.event.Observes;
import javax.enterprise.event.TransactionPhase;


@Stateless
public class HelloListener {
          public void listenToHello(@Observes HelloEvent helloEvent) {
                    System.out.println("Test: " + helloEvent);
          }


          public void listenToTxFail(@Observes(during=TransactionPhase.AFTER_FAILURE) HelloEvent helloEvent) {
                    System.out.println("TX FAILED: " + helloEvent);
          }


}

@Named("messenger")
@Stateless
public class HelloMessenger {


          @Inject 
          private Event<com.my.ejb.HelloEvent> myEvent;


          public void test() {
                    myEvent.fire(new HelloEvent("from bean " + System.currentTimeMillis())); 
                    System.out.println("Waiting for timeout 20 sec");
  
                    try{
                    Thread.sleep((1000*10)*2);
                    }catch(Exception e){
                              e.printStackTrace();
                    }
                    System.out.println("...");
          } 
}

[nickarls]

For every problem there is a solution that is simple, elegant and doesn't work ;-)

[jbertram]

Is your goal simply to have the client timeout the call to the server after a certain time, ideally corresponding with the timeout of the transaction on the server?

[markus78]

Yes that is basically the functionality that I need.

[jbertram]

The first road-block here is that when the transaction times out the associated thread is not interrupted.  This means that the thread will happily keep running long after the transaction is actually dead (which could potentially be quite awhile).  Only when it tries to perform more work with the transaction (e.g. enlist another resource, commit, rollback, etc.) will an exception actually occur (as per spec).  At this point the application itself can catch the exception and inform the client as necessary.

If you really want the client to abort its call after a certain amount of time then you'd have to execute the call in a special way to facilitate that behavior (e.g. using java.util.concurrent).

[markus78]

Yea thanks, It occured to me that perhaps all we really need is to set <set-tx-query-timeout> to true on our datasource, I will test that out first of all. Thanks for the help!

[jbertram]

If you are working with a JDBC datasource then <set-tx-query-timeout> will almost certainly help you (granted your JDBC driver obeys the query timeout). 

[markus78]

The property setting worked like a charm, our DB driver now throws an exception when the TX timeout occurs! thanks.

[markus78]

Well , this worked for a while, but now we are getting problems with OOMEM unable to create new native thread, and a quick look in visualvm shows that the culprit is thousands of "Timer-" threads, created by our db2jcc driver..

/Markus