[Paraview] Crash using openmpi with more than 4 processes

[Jeff Baumes]

Just a thought: Have you tried an entirely clean rebuild? Someone got
an error similar to this before and it went away after a rebuild.

Jeff

On Fri, Feb 5, 2010 at 10:32 AM, Bart Janssens <bart.janssens at rma.ac.be> wrote:
> On Thursday 04 February 2010 09:33:24 pm you wrote:
>> What about Valgrind? Any chance you can start one of the processes such
>> that you can run it in Valgrind?
>>
>
> I ran things in valgrind, but this showed exactly the same crash. I have
> recompiled mesa with TLS, so the PTHREADS block in the ASM code mentioned
> previously is ignored, and now I get a little further (see attached GDB
> trace). The problem appears to be glGetString(GL_VERSION) returning null.
>
> I added extra debug comments in vtkOpenGLExtensionManager.cxx, starting at
> line 496:
>
> --------------
>  if(version)
>  {
>    vtkWarningMacro(<< "GL Version: " << version);
>  }
>  else
>  {
>    vtkWarningMacro("GL version was null, IsCurrent is: " << this-
>>RenderWindow->IsCurrent());
>  }
> ----------------
>
> The output of this confirms that version is null when running on 8 processes,
> while RenderWindow->IsCurrent() is shown to be true (1).
> When running on 4 processes, on the other hand, everything still works and no
> text is generated at all! I would expect to at least see the contents of
> 'version' being printed, in that case? Is all this code somehow only becoming
> active when running many processes?
>
> I also tried aborting the extension manager in the same way as when there are
> no extensions. This avoids the crash but shows only a black screen.
>
> I think the error itself is related to the GL context being messed up somehow,
> even though RenderWindow->IsCurrent() appears to think it is OK. Is anyone
> using OSMesa on many processes? If so, could you please share the exact
> version info and configuration options?
>
> Finally, I also tested using on-screen rendering. This works fine, I tested up
> to 16 processes on the head node alone (not wanting to configure the whole
> cluster with X yet). Unfortunately, rendering became slower and slower as I
> increased the number of processes. Is launching 8 on-screen rendering
> processes on the same node futile, even if there are 8 CPU cores? Would OSMesa
> perform any better in this use case?
>
> Cheers,
>
> Bart
>
> _______________________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html
>
> Please keep messages on-topic and check the ParaView Wiki at: http://paraview.org/Wiki/ParaView
>
> Follow this link to subscribe/unsubscribe:
> http://www.paraview.org/mailman/listinfo/paraview
>