[Paraview] Problems with Paraview 3.6.1 on cluster (client - server mode)

Fri Oct 30 14:18:01 EDT 2009

(Responding back to ParaView list.)

No clue.  This still looks like a problem with library linking mismatch to me (although I could be wrong).  Even though the error does not specify the OpenGL library, I would still check that.  That is the library that I find is most likely to cause trouble.

-Ken

On 10/28/09 3:39 AM, "myaccountmail9 myaccountmail9" <myaccountmail9 at gmail.com> wrote:

Hi Kein
I apologize again.

We recompiled ParaView this time using BUILD_SHARED_LIBS = OFF (not ON), and the error on the library /usr/lib/libGL.so is gone.
Now it remains the error on the library /lib/libpthread.so.0:

Client connected.
[hpc:15607] *** Process received signal ***
[hpc:15607] Signal: Segmentation fault (11)
[hpc:15607] Signal code: Address not mapped (1)
[hpc:15607] Failing at address: (nil)
[hpc:15607] [ 0] /lib/libpthread.so.0 [0x7f51e4573a80]
[hpc:15607] *** End of error message ***
Segmentation fault

Have you any idea?

Thank you very much.
Luis

2009/10/28 myaccountmail9 myaccountmail9 <myaccountmail9 at gmail.com>
It sounds like you are still linking to the system OpenGL libraries somehow.  Those symbols should be defined in the libGL.so that is built with Mesa (whereas the /usr/lib/libGL.so on your system probably does not have them).

Hi Ken,
thanks again.

Looking the errors such as:

"home/myuser/TempOSMesa/Mesa-7.6/lib/libOSMesa.so: undefined reference to `_mesa_free_context_data'"

I believe those symbols must be defined in libOSMesa.so library and not in /usr/lib/libGL.so library.

In fact, if we install OSMessa using Debian packages we can compile ParaView, but we can't connect client and server; the error always is similar to:

ParaView3/Servers/Common/vtkServerConnection.cxx, line 67
vtkServerConnection (0x3182220): Server Connection Closed!

Executing with: 0
*** Process received signal ***
Signal: Segmentation fault (11)
Signal code:  (128)
Failing at address: (nil)
[ 0] /lib/libpthread.so.0 [0x7f4fc6536a80]
[ 1] /usr/lib/libGL.so.1(glBlendFuncSeparateEXT+0x16) [0x7f4fc7bc12e6]
*** End of error message ***
Segmentation fault

If instead we compile OSMesa ourselves we can't compile ParaView: I believe that something went wrong compiling OSMesa (some simbols aren't present in libOSMesa.so created), but I don't know if compiling OSMesa ourselves can help us to resolve our problem.

We also tried to install on our cluster ParaView 3.6.1already compiled (downloaded from http://www.paraview.org/paraview/resources/software.html) and it works: we can establish client-server connection.

But if we try to run more 2 or more pvserver instances on the same node (mpirun -np 2 pvserver), we have this error:

"vtkProcessModuleConnectionManager (0x23d6da0): Failed to set up server socket"

 Did you do a clean build of ParaView (that is, delete the binary directory and start over from scratch)?  If you just tried to change some CMake variables in an existing build, you may have some stale libraries built.

We did "make clean".

You should also check to make sure that the executables being built are linking to your compiled Mesa libraries instead of the system libraries.  You can check this with the ldd command.  You may have to set your LD_LIBRARY_PATH environment variable to point to /home/myuser/TempOSMesa/Mesa-7.6.

We also tried to set LD_LIBRARY_PATH.

Can you help us?
Luis

-Ken

On 10/23/09 3:27 AM, "myaccountmail9 myaccountmail9" <myaccountmail9 at gmail.com <http://myaccountmail9@gmail.com> > wrote:

Hi Ken,
thank you.
We compiled Mesa with:

./configure --enable-gl-osmesa --disable-gallium
make
make install

and we configured Paraview with:

OPENGL_INCLUDE_DIR=/home/myuser/TempOSMesa/Mesa-.6/include
OPENGL_gl_LIBRARY=/home/myuser/TempOSMesa/Mesa-7.6/lib/libGL.so
OPENGL_glu_LIBRARY=/home/myuser/TempOSMesa/Mesa-7.6/lib/libGLU.so
OPENGL_xmesa_INCLUDE_DIR=/home/myuser/TempOSMesa/Mesa-7.6/include
OSMESA_INCLUDE_DIR=/home/myuser/TempOSMesa/Mesa-7.6/include
OSMESA_LIBRARY=/home/myuser/TempOSMesa/Mesa-7.6/lib/libOSMesa.so

but we have many error compiling Paraview; for example:

/home/myuser/TempOSMesa/Mesa-7.6/lib/libOSMesa.so: undefined reference to `_mesa_free_context_data'
/home/myuser/TempOSMesa/Mesa-7.6/lib/libOSMesa.so: undefined reference to `_mesa_get_current_context'
/home/myuser/TempOSMesa/Mesa-7.6/lib/libOSMesa.so: undefined reference to `_mesa_enable_2_1_extensions'

Do you think there is something went wrong installing Mesa?
Can you help us?

Thank you again.
Luis

2009/10/21 Moreland, Kenneth <kmorel at sandia.gov <http://kmorel@sandia.gov> >
Not really, but I would consider trying downloading the Mesa source and compiling it myself.  Since the error is happening somewhere in the OpenGL libraries (and is not really being replicated on other user's platforms), the only cause I can come up with is a problem with the libraries (and mismatched libraries is the most common problem).  Compiling Mesa is actually pretty easy, so it is worth taking the effort to check.

-Ken

On 10/21/09 12:54 AM, "myaccountmail9 myaccountmail9" <myaccountmail9 at gmail.com <http://myaccountmail9@gmail.com>  <http://myaccountmail9@gmail.com> > wrote:

Hi Ken,
thank you for your reply.
No, we didn't compile OSMesa ourselves, but we installed OSMesa library on each node downloading the .deb packages with:

apt-get install libosmesa6
apt-get install libosmesa6-dev

Have you got any other idea?
Thank you again.
Luis

2009/10/20 Moreland, Kenneth <kmorel at sandia.gov <http://kmorel@sandia.gov>  <http://kmorel@sandia.gov> >
Could it be a problem with mismatched libraries?  I notice in the error below that it occurs in /usr/lib/libGL.so.1.  This is a system library.  Did you compile OSMesa yourself?  If so, then you should not be using the system GL libraries; you should be using the libraries that you compiled when you also generated the OSMesa library.

-Ken

On 10/20/09 1:56 AM, "myaccountmail9 myaccountmail9" <myaccountmail9 at gmail.com <http://myaccountmail9@gmail.com>  <http://myaccountmail9@gmail.com>  <http://myaccountmail9@gmail.com> > wrote:

Hi,
we have a Linux (Debian) cluster without gpu and we have a big problem using Paraview 3.6.1in client - server mode.
The server has been compiled with OSMesa library and following the instructions reported here (in particular the section "OSMesa support"):

http://www.itk.org/Wiki/Setting_up_a_ParaView_Server

If we try to connect the client started with the command:

paraview

and the server, started with the command:

pvserver --use-offscreen-rendering

we must disable the checkbox "Remote rendering threshold" (but this causes bad performances) in the client settings (Edit -> Settings -> Render view -> Server), othervise Paraview crashes after loading a big input or performing some operation such us changing "outline" to "wireframe"...

The error always is (similar to):

ParaView3/Servers/Common/vtkServerConnection.cxx, line 67
vtkServerConnection (0x3182220): Server Connection Closed!

Executing with: 0
*** Process received signal ***
Signal: Segmentation fault (11)
Signal code:  (128)
Failing at address: (nil)
[ 0] /lib/libpthread.so.0 [0x7f4fc6536a80]
[ 1] /usr/lib/libGL.so.1(glBlendFuncSeparateEXT+0x16) [0x7f4fc7bc12e6]
*** End of error message ***
Segmentation fault

This error also occurs if we start client and server logging directly (without any ssh connection) on the master node; on master node X is running and we can execute the test:

mpirun -np 4 /usr/X11R6/bin/glxgears

So, it doen't seem a problem regarding X connection and permission...

Moreover, executing the test:

/ParaView/ParaView-bin$ ctest -I ,,10

we obtain some segmentation fault, but the log file can't give any useful information.

111/475 Testing TestGradientBackground        ***Exception: SegFault
121/475 Testing TestOSConeCxx                 ***Exception: SegFault
131/475 Testing TestTranslucentLUTDepthPeeling***Exception: SegFault
141/475 Testing BoxWidget                     ***Exception: SegFault
151/475 Testing TestParallelopipedWidget      ***Exception: SegFault

..

On our local machine with gpu 100% test passed, and we haven't any problem to connect client and server on that machine.

MPI version is: mpirun (Open MPI) 1.2.7rc2
OSMesa version is: 7.0.3-7

We searched on google, on Paraview wiki... without any succes: can you help us please?

If you need more information let us now

Thank you very much

Luis

   ****      Kenneth Moreland
    ***      Sandia National Laboratories
***********
*** *** ***  email: kmorel at sandia.gov <http://kmorel@sandia.gov>  <http://kmorel@sandia.gov>  <http://kmorel@sandia.gov>
**  ***  **  phone: (505) 844-8919
    ***      web:   http://www.cs.unm.edu/~kmorel <http://www.cs.unm.edu/%7Ekmorel>  <http://www.cs.unm.edu/%7Ekmorel>  <http://www.cs.unm.edu/%7Ekmorel>

   ****      Kenneth Moreland
    ***      Sandia National Laboratories
***********
*** *** ***  email: kmorel at sandia.gov <http://kmorel@sandia.gov>  <http://kmorel@sandia.gov>
**  ***  **  phone: (505) 844-8919
    ***      web:   http://www.cs.unm.edu/~kmorel <http://www.cs.unm.edu/%7Ekmorel>  <http://www.cs.unm.edu/%7Ekmorel>

   ****      Kenneth Moreland
    ***      Sandia National Laboratories
***********
*** *** ***  email: kmorel at sandia.gov <http://kmorel@sandia.gov>
**  ***  **  phone: (505) 844-8919
    ***      web:   http://www.cs.unm.edu/~kmorel <http://www.cs.unm.edu/%7Ekmorel>

   ****      Kenneth Moreland
    ***      Sandia National Laboratories
***********
*** *** ***  email: kmorel at sandia.gov
**  ***  **  phone: (505) 844-8919
    ***      web:   http://www.cs.unm.edu/~kmorel

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.paraview.org/pipermail/paraview/attachments/20091030/e4f81789/attachment-0001.htm>