[jacorb-developer] JacORB 3.6 race condition establishing connections to multiple clients simultaneously?

[Nick Cross]

Hi,

Test cases are always welcome ;-) You could attach the code/logs to a 
bugzilla ticket. You might be able to make it consistently reproducible 
on any system by using byteman (https://www.jboss.org/byteman).

Thanks

Nick


On 11/03/15 23:41, Stout, Eric A. wrote:
> We've been running JacORB 3.4, but we're in the process of upgrading
> from Java 7 to Java 8, and we ran into a bug (BZ1002) that I found
> was fixed in JacORB 3.6.  So, I upgraded to JacORB 3.6, and ran into
> a different problem.
>
> We have some places where multiple client processes send an object
> reference to a server, which holds onto them, and sometime later,
> calls all of them at once using multiple threads.  The problem is,
> sometimes the ORB decides to use the same connection for more than
> one of the client objects - in other words, some of the clients get
> called more than once, and some don't get called at all.
>
> I have distilled this down into a pretty minimal test case, if
> someone from the development team (Nick Cross?) would like to have
> it.  However, the test case only reproduces the problem for me when
> the server is run on a pretty serious many-core Sparc-Solaris host,
> so I don't know if you have an environment in which to reproduce it
> or not.  I can include the test server output from a run that
> demonstrates the failure, as well.
>
> By the way, the test fails with JacORB 3.6 under both Java 1.8.0u25
> and 1.7.0u13.  The code on which the test is modelled works fine with
> JacORB 3.4 and either Java version.
>
> Please let  me know to whom I should send the test code.  Any help
> would be appreciated!
>
> Thanks, Eric Stout _______________________________________________
> jacorb-developer maillist  -
> jacorb-developer at lists.spline.inf.fu-berlin.de
> https://lists.spline.inf.fu-berlin.de/mailman/listinfo/jacorb-developer
>