Vasp GPU (single node) podman

[tonykew]

Instructions to build VASP with Nvidia GPU support in podman
and a sample Slurm script to run the podman tarball image.

Tony

[ANekhai]

These are sort of my initial comments from validating the build pipeline. My meta question is do we need three example run files? I think having a containerfile, build file, and slurm file should be more than plenty for users. I personally don't think we should support any other kinds of runs with scripts, if they want to run VASP GPU locally with a shell I say we leave this up to them. Please correct me if I'm misunderstanding the scope of the run scripts.

[ANekhai]

This comment kinda pertains to this file. I have concerns that there is just too much going on for a user. As a user, if I were looking for a slurm script to try, I would probably keep looking until I found an extremely simple example which I didn't have to read closely to understand.

My concern is that this script has far too many echo and conditional statements to check for directory structure and print out job info. I would rather an example of "hey this is how you can print out a ton of job info in a slurm script" example live somewhere else, uncoupled to trying to run VASP.

I don't think we should check to see if they have all the correct files when running VASP, just to have one less thing that will fail quietly. I'd rather VASP throw it's errors, rather than us check for them.

[tonykew]

I think the changes I've made have resolved this?

[ANekhai]

I would prefer keeping this line, and maybe a one-line comment about directory and file structure so a user can easily identify which lines are running and which lines are failing. Otherwise, we're going to get help desk tickets to try and troubleshoot these scripts, which defeats the purpose of having an example repository.

[tonykew]

The only sections the user needs to change (beyond the SLURM directives)
are clearly marked now.
I think these changes resolve this?

[dsajdak]

Some formatting comments and some questions

[dsajdak]

Can you describe what this section and the next are intended for? If the Slurm script exists then an account is included in it and a data path should be set in it.

[tonykew]

I think you are referring to this section:

if [ -f sample_SLURM_vasp_std.bash ]
then
  default_acct="$(sacctmgr -rnp show User "$(id -un)" | awk -F'|' '{print $2}')"
  sed -E -i -e "/#SBATCH[[:space:]]+--account=/s|\".*\"|\"${default_acct}\"|" \
   -e "/#SBATCH[[:space:]]+--chdir=/s|\".*\"|\"${base_dir}\"|" \
   -e "s|^bin_dir=.*|bin_dir=\"${bin_dir}\"|" \
   "sample_SLURM_vasp_std.bash"
  # try to provide a reasonable data dir path
  grp_group="$(groups | sed 's/ /\n/g' | grep ^grp- | head -1 | sed 's/^grp-//')"
  if [ "${grp_group}" != "" ]
  then
    if [ -d "/projects/academic/${grp_group}" ]
    then
      base_dir="/projects/academic/${grp_group}/$(id -un)/VASP"
      sed -E -i -e "s|^data_dir=.*|data_dir=\"${base_dir}/data\"|" \
       "sample_SLURM_vasp_std.bash"
    fi
  fi
  cp "sample_SLURM_vasp_std.bash" "${bin_dir}/sample_SLURM_vasp_std.bash"
else
  echo "sample script \"sample_SLURM_vasp_std.bash\" missing?" >&2
fi

This substitutes the users installed OCI directory in bin directory path into
the SLURM sample script, set the default SLURM account to the user's
group & sets sample paths based on the user's username & group, and copies
this sample script to the user's script dir

[dsajdak]

Yep, exactly. Thanks for the explanation. I personally think this is too much for the examples repo. I think the simpler the better, less to break and less to explain to users. However, others may have a different opinion.