Salome-Meca 2024.1 custom image with MPI solvers

jcugnoni

Hello to the Code-Aster community,

It has been a while since I posted last time here, but I finally got time to get back and publish some contributions.
Following great inputs gathered from different posts on the forum, I managed to build a working (but experimental) image of Salome-Meca 2024.1 with an MPI version of Code-Aster Testing 17.3.

The image is based on the official Salome-Meca 2024.1 release and the solver is compiled inside the image using the standard process described in GitLab.

There where some challenges to integrate the MPI version of the solver in Salome-Meca / AsterStudy environment (issues with env variables and different as_run paths), but these were "solved" by a quick and dirty patch on "run_aster:main.py" to fix the environment. I guess, there a re better solutions but at least it works.

The only issue that remains is that the solver hangs at the end of calculation if "save databases" is active in AsterStudy (default). So for the moment DISABLE SAVE DATABASES (yellow icon on he right of the green + sign) when submitting a job with MPI version.

All the build instructions, details and the image file can be downloaded on my Github repo:
https://github.com/jcugnoni/SalomeMeca2024_Code-AsterMPI

I hope that it will work as well for you that it is for me ;-)
(at least if the official 2024.1 image works, this one should work too)

Cheers

Joël Cugnoni

AsterO'dactyle

Bonjour,

Ca faisait longtemps !

Merci pour ce travail !

mf

Hello,

thank you for your efforts. Unfortunately, I ran into an error in the first study I tested. Find the details here, I started a new thread, because the error is not unique to your MPI version.

https://forum.code-aster.org/public/d/29842-keyword-nom-cham-is-mandatory-error-although-nom-cham-is-defined

However, this error is unique to your version, the study works in both sequential versions 16.7 and 17.2 (testing) in SM2024 but not in 17.3.1 (testing_mpi):

Unfortunately, the study is quite large, even compressed. But it is very simple at its core: it is just a pressurized tank. Nothing more.

Maybe I'll try compiling 17.2 into SM2024 according to your recipe tomorrow.

Mario.

mf

Hello,

I repeated your recipe, I have to add the "export RDMAV_FORK_SAFE=1" before "./waf configure", then compiling works. Out of the box it chooses 17.3.2 now.

However, creating the symlink "ln -s /usr/local/bin/as_run /opt/salome_meca/V2024.1.0_scibian_univ/tools/Code_aster_frontend-202410/bin/as_run"
does not work, I get a "ln: impossible de créer le lien symbolique '/opt/salome_meca/V2024.1.0_scibian_univ/tools/Code_aster_frontend-202410/bin/as_run': Le fichier existe".
If I force it, I do not have any version to choose from in the version drop-down menu.

I can choose the testing_mpi version if I ignore it, however. But above error persists.

Mario.

jcugnoni

@AsterO'dactyle : merci ça fait plaisir de pouvoir contribuer un peu et de retrouver la communauté. Je n'ai pas arrêté d'utiliser Code-Aster , bien au contraire mais avec le passage de toutes ces applis FEA vers des containers, je ne voyais plus vraiment la valeur ajoutée de faire une nouvelle release de CAELinux...

@Mario:
sorry, my mistake: the "ln" command is wrong; it should be:

ln -s /opt/salome_meca/V2024.1.0_scibian_univ/tools/Code_aster_frontend-202410/bin/as_run /usr/local/bin/as_run

I will fix this and att the export in the Github instructions.
Note that you can also download the image in the Releases and compare if it works
Regarding your crash in the MPI version; I will try to have a look in the other thread,

Cheers

Joël

mf

jcugnoni
Hello,

thanks, I will try in the evening.

The error in this thread was generated with the dowloaded .sif, also the one in the other thread.

Mario.

mathieu.courtois

Il faut éviter la 17.3.0 (la 17.3 ne sera jamais diffusée dans salome-meca) et préférer 17.3.2.

Bug concernant la sauvegarde : https://gitlab.com/codeaster/changelog/-/blob/main/histor/v17/17.3.2?ref_type=heads#L19

jcugnoni

mathieu.courtois
Merci Mathieu,
je vais essayer de faire une update avec la 17.3.2.
A bientôt !
Joël

jcugnoni

UPDATE:
I have rebuilt a new release with Code-Aster v17.3.2 using the same method.

Good news: as mentionned by Mathieu Courtois, now all the MPI jobs finish properly in Salome-Meca too!

The new Singulariy image is available the GitHub as release v0.1.1: https://github.com/jcugnoni/SalomeMeca2024_Code-AsterMPI

Apollo

jcugnoni
Bonjour Joël;

I am trying to install your version of Salome meca on my Windows sytems under WSL2.
I have followed the instruction using your link:
https://github.com/jcugnoni/SalomeMeca2024_Code-AsterMPI/releases/tag/v0.1.1

and run the "sudo..." commands to install and run the .sif file.
Bellow are the commands and outputs:

`To start salome_meca, just use:
/mnt/c/Users/olivier-PC/7zip/SalomeMeca2024_custom
or (in the installation directory):
./SalomeMeca2024_custom

If you want to check your configuration, use:
singularity run --app check SalomeMeca2024_custom.sif
olivier@olivier-PC:/mnt/c/Users/olivier-PC/7zip$ ./SalomeMeca2024_custom
/mnt/c/Users/olivier-PC/7zip/./SalomeMeca2024_custom:433: SyntaxWarning: invalid escape sequence '.'
regexp = re.compile("hwloc_(.*).so")

running salome_meca in GPU mode
++ using container SalomeMeca2024_custom.sif
++ prepend system 'libGL*' libs
olivier@olivier-PC:/mnt/c/Users/olivier-PC/7zip$ SALOME_Session_Server_No_Server: /lib/x86_64-linux-gnu/libc.so.6: version GLIBC_2.38' not found (required by /.singularity.d/libs/libGLdispatch.so.0) SALOME_Session_Server_No_Server: /lib/x86_64-linux-gnu/libc.so.6: versionGLIBC_2.34' not found (required by /.singularity.d/libs/libGLdispatch.so.0)
SALOME_Session_Server_No_Server: /lib/x86_64-linux-gnu/libc.so.6: version GLIBC_2.38' not found (required by /.singularity.d/libs/libGLX.so.0) SALOME_Session_Server_No_Server: /lib/x86_64-linux-gnu/libc.so.6: versionGLIBC_2.34' not found (required by /.singularity.d/libs/libGLX.so.0)
olivier@olivier-PC:/mnt/c/Users/olivier-PC/7zip$`

Any thoughts?
Sincerement.
Olivier

jcugnoni

Apollo
Hi, these errors related to GLIBC and libGLX are usually due to incompatible versions between your wsl distribution libraries and the container for OpenGL and C runtime.
Try to run the app in software rendering to see if it works:
./SalomeMeca2024_custom --soft

And also which distribution do you have in WSL?

raja

Apollo

Check this solution. It seems to work for many people, including me.

Link to forum