On Dec 29, 2009, at 12:59 AM, Daniel S. Haischt wrote:
btw the buildbot status page URL is:
http://ci.apache.org/builders/openejb-trunk/
On Mon, Dec 28, 2009 at 11:16 PM, Jacek Laskowski <[email protected]
>wrote:
On Mon, Dec 28, 2009 at 9:32 PM, David Blevins <[email protected]
>
wrote:
Going to see if I can't add more checks in the test case in some
way and
maybe drive out some more information.
I wish I could help with it, but have no idea how. Looking for
low-hanging fruits and thought it could be one, but I couldn't have
been mistaken more :(
I figured it out after jamming all the system.out statements in
there. Basically we never got finished creating all the "client"
threads within the allowed 20 seconds. The test is setup so that all
the threads will 'await' on a CyclicBarrier till everyone has reported
in and is ready to go -- i.e. all the runners are at the starting
line. There were simply too many "runners" for the machine to create
in the 20 seconds allowed.
That can happen really with any setting, so rather than just lower the
number of threads (runners) or increase the timeout, I fixed the test
so that it *should* brake more obviously if this condition occurs
again, which it could easily since it's largely dependent on machine
speed. Hopefully continuum will show the failure with a better stack
trace now. I basically just made it so the main test thread also
waits on the CyclicBarrier (the starting pistol), which seems
completely obvious to me now, but of course that's the way it goes :)
Anyway, I'll lower the thread count tomorrow on that test if all
breaks as expected.
-David
