Small utility to run top or nvidia-smi on a compute node from the login node

susage

#!/usr/bin/env bash

TEMP=$(getopt -o hsg --long help,snapshot,gpu -n 'susuage' -- "$@")

if [ $? != 0 ] ; then echo "Terminating..." >&2 ; exit 1 ; fi

# Note the quotes around `$TEMP': they are essential!
eval set -- "$TEMP"

SNAPSHOT=false
GPU=false
while true; do
  case "$1" in
    -h | --help ) echo "susage [-h/--help] [-s/--snapshot] [-g/--gpu] JOBID"; exit 0 ;;
    -s | --snapshot ) SNAPSHOT=true; shift ;;
    -g | --gpu ) GPU=true; shift ;;
    -- ) shift; break ;;
    * ) break ;;
  esac
done

JOBID=$1
if [[ -z $JOBID ]]; then
    echo "Please supply SLURM jobid."
    exit 2
fi

# Check that this job is not PENDING/HELD
[[ $(scontrol show job ${JOBID} | egrep -o "JobState=RUNNING") =~ RUNNING$ ]] || { echo "Job ${JOBID} is not currently running.  Exiting."; exit 1; }

# Unfortunately nvidia-smi and top have different defaults,
# with top automatically polling and nvidia requiring an option to poll
top_options="-u ${USER}"
smi_options="-l 3"
if [[ "$SNAPSHOT" == true ]];
then
    top_options="-n 1 -u ${USER}"
    smi_options=""
fi

cmd="top ${top_options}"
srun_options="-p compute -q batch"
if [[ "$GPU" == true ]];
then
    cmd="{ module load cuda10.0/toolkit; nvidia-smi ${smi_options} }"
    srun_options="-q dev"
fi

nodelist="--$(scontrol show job ${JOBID} | egrep -o 'NodeList=([a-z0-9]+)' | tr [:upper:] [:lower:])"

srun ${nodelist} ${srun_options} --pty ${cmd}

Should put parentheses around echo "Job ${JOBID} is not currently running. Exiting."; exit 1

Should put parentheses around echo "Job ${JOBID} is not currently running. Exiting."; exit 1

Changed the offending portion to look like { cmd1; cmd2; } so that the exit happens in the contexts of the current shell (rather than a subshell with (...)).

Note that the nvidia-smi functionality doesn't currently work. srun requires an executable, so the cmd should probably be a function, but can't get it to work 100% correctly.