[vtk-developers] Test failing since Nov 13, request stack trace

[David E DeMarle]

Marcus and I also spent some quality time on londinium, but not enough time
to resolve the issue. Perhaps before the next release we'll find the time
to get it resolved.

Run VTK/Testing/Core/vtk_site_history.py to get a sense for, and actual
data about, what machines and what particular tests on those machines are
erratic.

londinium, amber10 and recently karego-at fall into that category now.




David E DeMarle
Kitware, Inc.
R&D Engineer
21 Corporate Drive
Clifton Park, NY 12065-8662
Phone: 518-881-4909


On Thu, Nov 14, 2013 at 10:05 AM, David Gobbi <david.gobbi at gmail.com> wrote:

> Thanks for the info.  If it's caused by the nVidia libraries/drivers
> doing stuff whenever a thread is created, then I was overly optimistic
> about finding a quick fix (this doesn't effect any of my machines,
> I was just hoping to green up the dashboard).
>
>   David
>
> On Thu, Nov 14, 2013 at 7:43 AM, Berk Geveci <berk.geveci at kitware.com>
> wrote:
> > Hi David,
> >
> > I (attempted to) tracked this quite a while ago. It has been failing on
> and
> > off for much longer. There is something very strange happening on that
> > machine. Some weird interaction between nVidia drivers and pthreads. The
> > stack trace shows that the program sometimes crashes when creating a new
> > threads. This can be reduced to a program that does very little but
> creating
> > threads (even if those threads don't do any work). And the stack trace
> > include calls into the nVidia libraries, which seem to somehow get
> involved
> > with thread creation. The crash goes away when not linking against
> nVidia GL
> > or when not creating threads. I traced the program through Valgrind
> > (memcheck and helgrind) and debugged it quite thoroughly and this is
> still
> > all I know. I kind of gave up at the time. I should probably go back and
> > work on it some more since I am introducing more threaded code into VTK.
> Ah
> > another thing. This happens on some other machines but we could find
> > correlation between the nVidia card, the drivers used, the libc
> > implementation and the crash. So it is very hard to report anything to
> > nVidia too...
> >
> > -berk
> >
> >
> > On Thu, Nov 14, 2013 at 9:08 AM, David Gobbi <david.gobbi at gmail.com>
> wrote:
> >>
> >> Hi All,
> >>
> >> I'm wondering if someone can give me a stack trace for a test segfault
> >> on the following dashboard:
> >>
> >> Dashboard: londinium.kitware Arch-GCC-4.8-x86_64-release
> >> Test: Filters/Core/Testing/Python/multipleComponentContour.py
> >>
> >> This test only fails on the londinium release dashboard, it does not
> >> fail on the debug dashboard.  I'm hoping that a stack trace will
> >> provide hints about why the segfault occurs.
> >>
> >>   David
> _______________________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Follow this link to subscribe/unsubscribe:
> http://www.vtk.org/mailman/listinfo/vtk-developers
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/vtk-developers/attachments/20131114/61eaef30/attachment.html>