SLURM batch script

this is my template SLURM batch script for for running python jobs on Greene. Greene requests that python be installed in and run from Singularity containers, so I've included the Singularity incantation here.

in general, for a python script at /path/to/project/script.py, I'll include this in a file called /path/to/project/run_script.sh. then the job is submitted with sbatch run_script.sh.

#!/bin/bash
#SBATCH -J job_name
#SBATCH -t 02-12:00:00
#SBATCH --mem 16G
#SBATCH --nodes 1
#SBATCH --ntasks-per-node 1
#SBATCH --cpus-per-task 1
#SBATCH --gres gpu
#SBATCH mail-type ALL
#SBATCH mail-user address@example.com
#SBATCH -o slurm-%x-%j.out
#SBATCH -e slurm-%x-%j.out

module purge
singularity exec \
    --nv \
    --overlay /path/to/overlay.ext3:ro \
    /scratch/work/public/singularity/*.sif \
    /bin/bash -c \
    "cd /path/to/script/; \
    python py_script.py --args"

some notes on using this:

• omit the lines #SBATCH --gres gpu and --nv if you don't need a GPU
• change :ro to :rw if you'll need to modify files in the /ext3 overlay
• be sure to activate your python env if necessary inside the bash command
• the file slurm-*.out will get written where you run the sbatch command; change it to an absolute path /path/to/slurm-%x-%j.out if you don't want this