well the point is we shouldn't end the method without ensuring all thread are done. join is brutal but the most sure way to do so.
Main issue on this test is buildbot since locally it fails very very rarely (got it once watching tv on the computer during the build ;)) Romain Manni-Bucau Twitter: @rmannibucau Blog: http://rmannibucau.wordpress.com/ LinkedIn: http://fr.linkedin.com/in/rmannibucau Github: https://github.com/rmannibucau 2014-07-21 12:47 GMT+02:00 Andy Gumbrecht <agumbre...@tomitribe.com>: > +1 > > Side note: I have been looking at the 'sometimes' failing > StatelessInstanceManagerPoolingTest. > > The issue is not the Thread.join, I tried the same fix but does not work - > If we join on each thread then we get a better chance, but only due to > forcing threads to run sequentially and timing. > The issue is the the teardown is called which calls OpenEJB.destroy. The > shutdown does not take into account running call contexts and they get > interrupted. > This kills the threads in the 'explode' call and more often than not they > never complete - discardedInstances.incrementAndGet() never gets called > in this case, and the test fails. > It seems wrong to me for the destroy to be so aggressive, it should surely > give call context threads some time to complete. > > Andy. > > > On 21/07/2014 10:17, Romain Manni-Bucau wrote: > >> 1.7 is affected as well, we just have luck sometimes since this code is >> the >> same. (depend test ordering for some of them but code was clearly wrong) >> >> I mainly focused on trunk since it showed some bugs of the 1.7 we can't >> see >> while we don't run tests against java 7. All are fixed excepted this one >> which should be trivial once we have the paths. >> >> >> >> Romain Manni-Bucau >> Twitter: @rmannibucau >> Blog: http://rmannibucau.wordpress.com/ >> LinkedIn: http://fr.linkedin.com/in/rmannibucau >> Github: https://github.com/rmannibucau >> >> >> 2014-07-21 10:09 GMT+02:00 Jean-Louis Monteiro <jlmonte...@tomitribe.com >> >: >> >> Looks like we are having issues with examples and I was wondering why >>> suddenly. >>> Regarding Apache DS fullstack example, I can give that a try if Andy does >>> not have time. >>> >>> Well, while trunk is important, I think we must focus on the branch until >>> it gets released. >>> >>> Jean-Louis >>> >>> -- >>> Jean-Louis Monteiro >>> http://twitter.com/jlouismonteiro >>> http://www.tomitribe.com >>> >>> >>> On Mon, Jul 21, 2014 at 8:41 AM, Romain Manni-Bucau < >>> rmannibu...@gmail.com >>> wrote: >>> >>> Hi >>>> >>>> this sample is failing on buildbot cause we don't find all jars from the >>>> app in the classloader (which is wrong since it is embedded ;)). >>>> >>>> Here is the info: >>>> >>>> INFO - Creating dedicated application classloader for classpath.ear >>>> >>>> It is in Assembler if someone wants to have a look , look for this: >>>> >>>> >>>> if (skipLoaderIfPossible) { ... >>>> >>>> >>>> Surely a File.equals issue or something like that. >>>> >>>> >>>> We can log all files in debug mode and activate debug for this sample >>>> to go further >>>> >>>> >>>> wdyt? >>>> >>>> >>>> I'll try to find some time to debug further tonight if nobody already >>>> did >>>> it >>>> >>>> >>>> Then we should be green everywhere even on trunk ;) >>>> >>>> >>>> >>>> Romain Manni-Bucau >>>> Twitter: @rmannibucau >>>> Blog: http://rmannibucau.wordpress.com/ >>>> LinkedIn: http://fr.linkedin.com/in/rmannibucau >>>> Github: https://github.com/rmannibucau >>>> >>>> > -- > Andy Gumbrecht > > http://www.tomitribe.com > agumbre...@tomitribe.com > https://twitter.com/AndyGeeDe > > TomEE treibt Tomitribe! | http://tomee.apache.org > >
