[Paraview-developers] xdisplay-test + display=1 => segfault

Mon May 15 12:00:50 EDT 2017

Hi Utkarsh,

thanks you for the hints!

I will compile ParaView then with EGL and check if that fixes the problem. Probably EGL would be the better solution anyway.

Currently I get errors when linking pvpython:
/homeb/zam/goebbert/workspace/paraview/pv_compile/build/lib/libvtkIOVisItBridge-pv5.3.so.1: undefined reference to `gluTessEndPolygon'
It seems as if libGL is still required - but this is a different story. I have to check it.

Best,
Jens Henrik

P.S:
We are currently running on Nvidia driver version 375.26 (with GLVND style)

-------------------------------------------------------------------------------------
Dipl.-Ing. Jens Henrik Goebbert
Institute for Advanced Simulation (IAS)
Jülich Supercomputing Centre (JSC)
Wilhelm-Johnen-Straße, 52425 Jülich, Germany
phone: +49 2461 61-96498
email: j.goebbert at fz-juelich.de<mailto:jh.goebbert at fz-juelich.de>
http://www.fz-juelich.de/ias/jsc

________________________________
From: Utkarsh Ayachit [utkarsh.ayachit at kitware.com]
Sent: Monday, May 15, 2017 4:05 PM
To: Göbbert, Jens Henrik
Cc: paraview-developers at paraview.org
Subject: Re: [Paraview-developers] xdisplay-test + display=1 => segfault

It seems that the workaround for this for now is to have users start their X servers with the “-noreset” option.  Alternatively, you could use EGL enabled server executables and this will be totally overcome.

Utkarsh

On Sat, May 13, 2017 at 12:11 PM, Utkarsh Ayachit <utkarsh.ayachit at kitware.com<mailto:utkarsh.ayachit at kitware.com>> wrote:
Let me try to contact NVidia developers and see if they can provide some info. What driver versions did you try this with?

Utkarsh

On Thu, May 11, 2017 at 9:34 AM, "Göbbert, Jens Henrik" <j.goebbert at fz-juelich.de<mailto:j.goebbert at fz-juelich.de>> wrote:
Hi Utkarsh,

you are right - I forgot to set the threshold to 0 before testing.
Now it crashes when opening a new render view.

We tested different versions of the NVIDIA driver already - they all behave the same way.
Any hints how to track this error down?

Best,
Jens Henrik

________________________________________
From: Utkarsh Ayachit [utkarsh.ayachit at kitware.com<mailto:utkarsh.ayachit at kitware.com>]
Sent: Thursday, May 11, 2017 3:21 PM
To: Göbbert, Jens Henrik
Cc: paraview-developers at paraview.org<mailto:paraview-developers at paraview.org>
Subject: Re: [Paraview-developers] xdisplay-test + display=1 => segfault

Jens,

One final thing, can you ensure that you're doing remote rendering
when you're closing/creating render views i.e. make sure your Remote
Render Threshold is set to 0 before you try the test  I suggested.

On Thu, May 11, 2017 at 4:37 AM, "Göbbert, Jens Henrik"
<j.goebbert at fz-juelich.de<mailto:j.goebbert at fz-juelich.de>> wrote:
> Hello Utkarsh,
>
> thank you for the hint.
> I checked if ParaView/5.3.0 (connected to 12 pvservers on the same node) crashes, if I close the render view and open a new one. This is not the case - I can close and open render views multiple times without problems (as long as I have called pvserver with "--disable-xdisplay-test").
>
> Best,
> Jens Henrik
>
> P.S:
> In this context I would like to mention this bugreport:
> If using VirtualGL and NVIDIA driver in non-GLVND style ParaView 5.3.0 segfaults on startup (even without pvservers).This is not the case with the driver in GLVND style.
> "Segfault with ParaView 5.3.0RC2" - https://github.com/VirtualGL/virtualgl/issues/44
>
> ________________________________________
> From: Utkarsh Ayachit [utkarsh.ayachit at kitware.com<mailto:utkarsh.ayachit at kitware.com>]
> Sent: Wednesday, May 10, 2017 7:24 PM
> To: Göbbert, Jens Henrik
> Cc: paraview-developers at paraview.org<mailto:paraview-developers at paraview.org>
> Subject: Re: [Paraview-developers] xdisplay-test + display=1 => segfault
>
> Jens,
>
> Can you try closing current render view and creating new view? Does
> that fail even when you pass `--disable-xdisplay-test`? In my
> experience, whenever I have seen a crash in pvserver startup which can
> be circumvented by `--disable-xdisplay-test`, it's been a driver bug
> with creating multiple opengl contexts.
>
> Utkarsh
>
> On Wed, May 10, 2017 at 11:51 AM, "Göbbert, Jens Henrik"
> <j.goebbert at fz-juelich.de<mailto:j.goebbert at fz-juelich.de>> wrote:
>> Hello ParaView-Team,
>>
>> we recently have experienced reproducible a segfaults of Xorg on display=1
>> (on CentOS 7.2 server with two NVIDIA Tesla K40), when used with:
>> --server--
>>   parallel ParaView/5.3.0 session (compiled with VTK_USE_X)
>>   over turboVNC(2.1.1) +VirtualGL(2.2.1)
>> --client--
>>   from a Windows7 client
>>   with turboVNC(2.1.1)
>> (other versions of turboVNC and VirtualGL led to the same
>>
>> This crash does _NOT_ happen, if the vncviewer(yes: vncVIEWER!) is running
>> on a Linux machine.
>> As soon as ParaView GUI (running on the server with turboVNC+VirtualGL, too)
>> connects to the pvservers running on the same machine, at least one of the
>> pvservers on display=1 crashes (MPI leads then to a stop of the other
>> pvservers).
>> The crash of the pvserver on display 1 leads to a crash of Xorg of display 1
>> - or it is the other way round.
>>
>> We found that this crash can be avoided by starting the pvservers with the
>> flag "--disable-xdisplay-test"
>> I report this here, as it took some time to stumble over the flag
>> "--disable-xdisplay-test" and this mail might help someone else.
>> It seems as if the xdisplay-test is doing some nasty stuff ...
>>
>> As this bug is related to the OS(=Windows) of the client running turboVNC
>> the xdisplay-test might trigger a bug in turboVNC or Xorg itself...
>>
>> Best,
>> Jens Henrik
>>
>> P.S:
>> We have tested this with older ParaView versions, too (5.2.0 and 5.1.2) and
>> got the same segfaults of Xorg of display 1.
>> The only difference was, that ParaView 5.3.0 results in a segfault when
>> connecting and ParaView 5.2.0+5.1.2 results in segfaults when loading data.
>>
>>
>>
>>
>> ------------------------------------------------------------------------------------------------
>> ------------------------------------------------------------------------------------------------
>> Forschungszentrum Juelich GmbH
>> 52425 Juelich
>> Sitz der Gesellschaft: Juelich
>> Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498
>> Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher
>> Geschaeftsfuehrung: Prof. Dr.-Ing. Wolfgang Marquardt (Vorsitzender),
>> Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt,
>> Prof. Dr. Sebastian M. Schmidt
>> ------------------------------------------------------------------------------------------------
>> ------------------------------------------------------------------------------------------------
>>
>>
>> _______________________________________________
>> Powered by www.kitware.com<http://www.kitware.com>
>>
>> Visit other Kitware open-source projects at
>> http://www.kitware.com/opensource/opensource.html
>>
>> Search the list archives at:
>> http://markmail.org/search/?q=Paraview-developers
>>
>> Follow this link to subscribe/unsubscribe:
>> http://public.kitware.com/mailman/listinfo/paraview-developers
>>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/paraview-developers/attachments/20170515/58c70243/attachment.html>