Hi,
"Bordet, Simone" wrote:
> Hey Ole,
> > - The client hangs.
> > The call to MarshalledObject.get() in
> > org/jnp/interfaces/NamingContext.java line 299 goes
> > in, but control never comes back.
> >
> > For the last week or two, I have only seen client hangs
> > and no exceptions.
>
> Yeah, this is also what I get.
> In particular, randomly, it happens that sometimes it takes 1-2 sec to
> create the InitialContext then hangs for 10-20 seconds then I finally get
> the home interface. Some other times (most of the times) though, I ctrl-C
> the client, tired of waiting more than 1 minute and getting no home
> interface.
> Last time I tried it happened that I ctrl-C'ed some 15 times the client,
> then I gave up and went out running for 45 minutes: the latter was a better
> decision than going on with the dumb client.
I have seen (almost) no random behaviour. When I've
compiled a binary, I get the same result on each run.
But when I change some code (having nothing to do with
RMI or JNDI) and recompile, the new code sometimes has
a new behaviour with no explanation whatsoever.
> > When debugging, I am always stopped by some native
> > method that I cannot step into to find out about the
> > problem.
>
> Did you try a profiler to see where is most of the time spent ? I think I
> will download OptimizeIt evaluation and give it a try...
Not yet. It should be simple to write some C code
to get raw profiling data out from a running VM,
so I am still looking for an opensource project
to pop up.
> > Please help!!
> > I have been spending a lot of time on this, and I do
> > not mind spending a lot more time if needed to get
> > this fixed.
> > But I simply do not know what to do about it!
>
> Did you already try to take jbosstest sources (client and bean+interfaces)
> jarring them from scrap and add the needed jars one by one, trying to add
> them in the manifest or leaving the manifest empty and adding them to normal
> classpath ?
> I'll try writing a plain bean and see if there are problems also, but it
> seems to me something very tight coupled with jbosstest module...
This made me think (again, my brains seemed to have
stopped). I did some dissecting of the tests, and found
that jboss-client.jar is still in the CLASSPATH of the
test jars. Removing it does not help. But if I don't
remove jboss.client from the jar CLASSPATH and instead
copy this jar from the jBoss distribution, everything
runs fine!
This makes sense as all problems I have seen were when
home proxies (and their superclasses) were loaded from
JNDI. If the jboss-client.jar is local, the superclasses
of home proxies are loaded locally instead.
Thanks. At least I have a set of tests that work now.
Should IMHO still look into the problem so we do not
have to copy jboss-client.jar after every jBoss build.
Best Regards,
Ole Husgaard.
