Installation Troubles on HPC

[AndrewH]

Hello,

I'm trying to install Code Saturne v3.3.2 on a HPC with SUSE OS. I'm compiling Code Saturne with HDF5-1.8.2, CGNS-3.2.1, and libxml2-2.9.2 libraries. I used the gcc-4.7.3 and mpicc (OpenMPI-1.8.1) to compile the libraries and Code Saturne. For libxml2-2.9.2 and Code Saturne, I also used Python-2.7.8. Everything in the configuration and install process goes fine, but when I go and run a .xml case I generated on a local computer, along with its mesh, I receive errors. All of the errors occur when reading subprocess.py and I recieve "OSError: [Errno 13] Permission denied," and sometimes "OSError: [Errno 9] Bad file descriptor" too. I assume it has something to do with my Python or libxml2 libraries, and I reinstall these libraries several times with different configurations, but nothing seems to help. I have also seen another post suggesting using --without-modules to keep from using unintended modules, but this doesn't help either. I also think I set up the configuration file to run Code Saturne on a batch system correctly, but I'm not sure.

I attached the error file that I commonly get, along with my config.log, batch configuration, batch submission, and .xml file. I tested out my code with a case I generated on a local Ubuntu desktop, which it attached, and the example http://cfd.mace.manchester.ac.uk/twiki/ ... 01TbLamGui. Both run fine on the my Ubuntu desktop.

For my Python complication, I just enabled --with-system-ffi and CC=gcc. And for the libxml2 complication, I enabled --with-ftp=no, --with-http=no, CC=gcc, and used my Python installation.

Any help will be appreciated.

Thank you,
Andrew

[Yvan Fournier]

Hello,

The beginning of your output file starts with:

-bash: BASH_FUNC_module(): line 0: syntax error near unexpected token `)'
-bash: BASH_FUNC_module(): line 0: `BASH_FUNC_module() () { eval $($LMOD_CMD bash "$@");'

so I suspect a problem with the "module" command on your system (some users occasionally encounter issues with that).

To work around this, add "--with-modules=no" to your configure command. This will disable automatic usage of modules by the code, so you will need to load the same modules as those used for building the code in your own environment (or in an "rc file" defined through code_saturne.cfg, or in a top-level code-saturne module), but at least you won't have issues with excess automation (and we can make sure that is the cause of the issue).

Regards,

Yvan

[AndrewH]

Hello,

I reconfigure and compiled my Code Saturne with "--with-modules=no" and I no longer receive the error OSError: [Errno 13] Permission denied. However, I get the new error message stating that the results file already exists and calculations will not be run.

File "/home/external/ccytbf/build/code_saturne-3.3.2/lib/python2.7/site-packages/code_saturne/cs_case.py", line 503, in set_result_dir
raise RunCaseError(err_str)
cs_case_domain.RunCaseError:
Results directory: /gpfs/home/external/ccytbf/test/cylinder_2/CASE1/RESU/20141028-1929 already exists.
Calculation will not be run.

I also spoke with the IT department managing the HPC and they said that the bash error that I'm also receiving should be benign.

Thank you,
Andrew

[Yvan Fournier]

Hello,

This probably means you tried to run the calculation twice withing less than one minute.

You can simply destroy the directory and re-run, or wait one minute and run.

Regards,

Yvan

[AndrewH]

Hello,

That is what I thought at first too because I make that mistake on my desktop often. However, the error continues to happen if I wait or delete the results folder. The error doesn't occur if I delete mpirun in my submission file, which I thought mpirun is required for jobs using mpi, but code saturne says I'm using 12 processors! Although later on in the calculation stage I receive the following error:

cs_gui.c:3824: Fatal error.

Error: can not find the required symbol: velocity[0], velocity[1] ou velocity[2]

Call stack:
1: 0x7f18b6f58812 <uiiniv_+0x20f2> (libsaturne.so.0)
2: 0x7f18b6bd6efa <inivar_+0x38e> (libsaturne.so.0)
3: 0x7f18b6af4771 <caltri_+0x2951> (libsaturne.so.0)
4: 0x7f18b6ad5c98 <cs_run+0x418> (libsaturne.so.0)
5: 0x7f18b6ad572a <main+0x14a> (libsaturne.so.0)
6: 0x7f18b4926c36 <__libc_start_main+0xe6> (libc.so.6)
7: 0x400b79 <> (cs_solver)
End of stack

Is there an error in my configuration of Code Saturne? Do I need to specify my OpenMPI directory in my configuration?

Thank you,
Andrew

[AndrewH]

Hello,

I realize where I went wrong, I had two versions of Code Saturne installed on my Ubuntu desktop, v3.0.5 and v.3.3.2, and I accidentally selected the wrong .xml file. I generated .xml file with v3.0.5 and was trying to run it with my job on the HPC that had version v.3.3.2 compiled. I used the correct .xml and everything seems to be correctly now. Thank you very much your help!

In regards to the mpi command, is mpirun integrated into the code_saturne run command?

Thank you,
Andrew