[Paraview] [EXTERNAL] BadAlloc Error

[Scott, W Alan]

Lloyd, 
I see the same thing,  ... um ... sort of.  Mine occurs when ssh -X'ing into a remote blade.  New ParaView (i.e., 5.0.0 and later) needs OpenGL 3.2.  I suspect that X forwarding isn't supporting OpenGL 3.2.

If I log onto the blade directly, the OGL version is sufficient.  But, when ssh -X'ing into it, if I do a "glxinfo | grep OpenGL", three of the lines will say 

OpenGL vendor string: NVIDIA Corporation
OpenGL renderer string: Quadro 3000M/PCIe/SSE2
OpenGL version string: 2.1.2 NVIDIA 337.25

I wonder if you aren't seeing the same thing?  


-----Original Message-----
From: ParaView [mailto:paraview-bounces at paraview.org] On Behalf Of Lloyd Brown
Sent: Wednesday, March 30, 2016 1:37 PM
To: paraview at paraview.org
Subject: [EXTERNAL] [Paraview] BadAlloc Error

Hi, all.

I'm trying to get Paraview 5.0 to interact well with the Tesla k80 GPUs in my HPC lab, and I'm encountering an interesting error.  I'm hoping that someone can point me in the right direction to diagnose it.

So, on an HPC node, with some k80s installed, I'm launching Xorg (example config attached) as root, then launching pvserver
("DISPLAY=:0.0 pvserver") as my user.  Then when I try to connect from the Paraview frontend on my desktop, pvserver exits with this error:

> Waiting for client...
> Connection URL: cs://m8g-1-5:11111
> Accepting connection(s): m8g-1-5:11111 Client connected.
> X Error of failed request:  BadAlloc (insufficient resources for
> operation)
>   Major opcode of failed request:  135 (GLX)
>   Minor opcode of failed request:  34 ()
>   Serial number of failed request:  26
>   Current serial number in output stream:  27

Now, according to the guys on the xorg users list (https://lists.x.org/archives/xorg/2016-March/057984.html), this is occurring as a result of the glXCreateContextAttribsARB call getting denied resources somehow, which lines up with the backtrace from gdb (also attached).

Now, under the assumption that pvserver was somehow running out of memory, I've verified that the problem still occurs when my HPC job requests 64GB (which means the cgroup will limit me to that).

Since both the client and server are precompiled 5.0 binaries from paraview.org, I'm not entirely sure where to go here.  Can anyone shed any insight on what might be going on?  A misconfiguration in my Xorg config?  Some software package I'm missing?

It's worth noting that several GLX-based benchmarks (eg. glxgears, glxspheres64, glmark2) and utilities (eg. glxinfo, glewinfo) seem to be able to interact with the Xorg just fine.

Thanks,

--
Lloyd Brown
Systems Administrator
Fulton Supercomputing Lab
Brigham Young University
http://marylou.byu.edu