Job Scripts

For each calculation, its job is handled by JobManager. This class can submit a job, store its jobid, and monitor its progress.

The creation of the actual creation of the vasp.q SLURM submission script is handled by VaspInputCreator using settings from calc_config.json and computing_config.json.

Implementation Details

The vasp.q file is just a template string that needs to be passed 3 separate sections:

vasp.q

#!/bin/bash

{sbatch_params}

{preamble}

starttime=$(date +%s)

{command}

stoptime=$(date +%s)
tottime=$(echo "$stoptime - $starttime" | bc -l)
echo "total time (s): $tottime"
to_hours=$(echo "scale=3; $tottime/3600" | bc -l)
echo "total time (hr): $to_hours"

1. {sbatch_params}: sbatch tags (e.g. #SBATCH -N)
2. {preamble}: environment or module settings (e.g. export OMP_NUM_THREADS=1)
3. {command}: the mpi or srun command to launch VASP

The actual commands for each supercomputer and job type are determined by q_mapper.json which lists the path of a yaml file that contains the sbatch_params, preamble, and command to use.

You can modify the exact commands in the yaml file of your choice if needed. For reference, here's the yaml file for normal jobs on Perlmutter:

vasp.yml

sbatch_params: |-
  #SBATCH -N {n_nodes}
  #SBATCH -q {queuetype}
  #SBATCH -J {jobname}
  #SBATCH -A {allocation}
  #SBATCH -t {walltime}
  #SBATCH -C {constraint}
  #SBATCH --mem=0

preamble: |-
  ulimit -s unlimited
  export OMP_NUM_THREADS=1
  module load {vasp_module}
  mpitasks=$(echo "$SLURM_JOB_NUM_NODES * {ncore_per_node}" | bc)

command: |-
  srun -t {timeout} -u -n $mpitasks --cpu_bind=cores vasp_std > stdout.txt 2> stderr.txt