The NPE bug is addressed by https://github.com/jbossas/jboss-as/commit/6fb2053524d04655413d7fbe82adf9058e3952c3
The NPE occurred because the AS7 PU service was closed pre-maturely before the EM was. I added logging for when we create the container managed EMF and when we close it. That makes it clearer what happened for case like this. Scott On 06/22/2011 02:09 PM, Steve Ebersole wrote: > As for the nullness, you can see this if the ServiceRegistry is reused. > I have not come up with a good solution for that aside from us > explicitly checking (after adding needed method to ask) that the passed > registry is still valid. > > Not getting the concern here about concurrency. The map and list in > question are immutable after the registry instantiation is complete, > which has to happen on a single thread. But its way complicated to code > that. We would need a "InitializationContext" or something to hold them > during the constructor processing. I used a ConcurrentHashMp because > sure it will be *accessed* concurrently. This is stuff held at the > SessionFactory level. > > > On 06/22/2011 11:00 AM, Scott Marlow wrote: >> Also appears that AbstractServiceRegistryImpl.serviceList is a mutable >> object and accessed through a mutable object reference. >> AbstractServiceRegistryImpl.serviceList should also be changed to an >> immutable object reference (final and let >> AbstractServiceRegistryImpl.destroy just clear it). >> >> Something also needs to be done to make updates to the >> AbstractServiceRegistryImpl.serviceList object thread safe (used to >> ensure services are destroyed in reverse order). Maybe Steve should >> make this change, so he can choose the concurrent data structure (maybe >> CopyOnWriteArrayList?) Or maybe rolling our own concurrent ArrayList >> for needs like this. >> >> I'd jump on IRC but the place I'm at this week, seems to be blocking >> freenode irc... >> >> On 06/22/2011 11:30 AM, Scott Marlow wrote: >>> Good point, it should be "final". >>> >>> I'll make that change (if not done already). >>> >>> On 06/22/2011 11:20 AM, Emmanuel Bernard wrote: >>>> Could it be some thread concurrency visibility? >>>> Since it's a concurrent hashmap, I'd venture it's accessed >>>> concurrently :) >>>> >>>> Try and make the variable final (or volatile) to ensure proper >>>> visibility across threads. >>>> >>>> On 22 juin 2011, at 02:44, Scott Marlow wrote: >>>> >>>>> I'm not sure if we will see this NPE again but found the location >>>>> interesting. I wanted to report the failure in case it happens >>>>> again in >>>>> the future or anyone else is interested in playing detective for >>>>> mysteries like this. ;) >>>>> >>>>> The exception call stack is here http://pastie.org/2104060 >>>>> >>>>> It looks to me like AbstractServiceRegistryImpl.serviceBindingMap is >>>>> null for some reason. It makes me wonder if >>>>> AbstractServiceRegistryImpl.destroy() was called prematurely for some >>>>> reason (for some unexpected reason). >>>>> >>>>> I wanted to mention the NPE here in case it happens again, so we >>>>> can see >>>>> what is in common with this one. Next time, we will have to capture >>>>> the >>>>> AS server.log output which might have additional clues but didn't >>>>> this time. >>>>> >>>>> Scott >>>>> _______________________________________________ >>>>> hibernate-dev mailing list >>>>> hibernate-dev@lists.jboss.org >>>>> https://lists.jboss.org/mailman/listinfo/hibernate-dev >>>> >>> >>> _______________________________________________ >>> hibernate-dev mailing list >>> hibernate-dev@lists.jboss.org >>> https://lists.jboss.org/mailman/listinfo/hibernate-dev >> >> _______________________________________________ >> hibernate-dev mailing list >> hibernate-dev@lists.jboss.org >> https://lists.jboss.org/mailman/listinfo/hibernate-dev > _______________________________________________ hibernate-dev mailing list hibernate-dev@lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev
