JBAS-2773 - LRUEnterpriseContextCachePolicy should be update

[dlofthouse]

I am planning to look at the Jira issue JBAS-2773

At the moment the class 'org.jboss.ejb.plugins.LRUEnterpriseContextCachePolicy' makes use of 'java.util.Timer' for the scheduling of TimerTasks. The classes 'EntitySynchronizationInterceptor' and 'LRUStatefulContextCachePolicy' both also schedule timer tasks using this timer. As a result if any of the tasks fail with a RuntimeException the timer is killed and no further timers will fire.

I was planning to make use of 'org.jboss.util.TimerQueue' to replace the 'java.util.Timer' as this was implemented to be similar to 'java.util.Timer' but with an Exception safe run implementation so the failure of one task will not kill the timer.

Alternatively it has been suggested that I should use 'org.jboss.util.timeout.TimeoutFactory' instead.

I have had a look at where the two different implementations are used and it appears that the TimerQueue is not used anywhere in Branch_4_0 and the TimeoutFactory is used in a couple of places.

I think I should change LRUEnterpriseContextCachePolicy to use the TimeoutFactory.

As the TimerQueue is not used anywhere is there any reason to keep it? Or can it be removed to make it easier to decide which one to use?

[adrian.brock]

I would think the TimerQueue is what you want?
The TimeoutFactory only fires the event once.

But it doesn't look like the TimerQueue has been used since 2.4.x?

[dimitris]

If we add a "period" to the TimeoutFactory we'd have a good replacement for java.util.Timer. No?

[starksm64]

It also needs an ability to inject a thread pool for dispatching and possibly a notion of insufficient dispatching horsepower. See this remoting/jms discussion for where these issues came up:

http://www.jboss.com/index.html?module=bb&op=viewtopic&t=79144

[dlofthouse]

I have had a look at the remoting discussion, it appears that they did make use of the org.jboss.util.TimerQueue for remoting 1.4.1 but this has now subsequently been dropped in favour of the java.util.Timer for remoting 2.

So I think we are back where we started and no-one is currently using the org.jboss.util.TimerQueue.

Adding the period parameter to the TimeoutFactory doesn't appear to be too difficult although some additional consideration will be required to handle some locking when tasks are rescheduled to also cope if the task is simultaneously cancelled.

[starksm64]

Make the changes and add the approriate testcases so we can move forward with this.

[dimitris]

If the problem is really the uncought runtime exceptions that can kill the timer, I think we should just fix those for now, and introduce the change to TimeoutFactory (support for periodic execution) in the next (4.0.5) release.

[ldsanes]

If the problem is really the uncought runtime exceptions that can kill the timer, I think we should just fix those for now, and introduce the change to TimeoutFactory (support for periodic execution) in the next (4.0.5) release.