sreboot

#!/usr/bin/env bash

# Slurm: use scontrol to reboot nodes
# NextState is "Resume" by default, use "-d" to set to Down.
# Homepage: https://github.com/OleHolmNielsen/Slurm_tools/

action=resume
reason="reboot and resume"

# Parse command options
while getopts "dh" options; do
	case $options in
		d )	action=down
			reason="reboot and down"
			;;
		h | * )	echo "Usage:"
			echo "$0 [-d] <nodelist>"
			echo "The NextState after reboot is Resume.  Use -d to set NextState to Down"
			exit 1;;
	esac
done
shift $((OPTIND-1))
NODES="$1"

if [ $# -ne 1 ]
then
	echo "Usage: $0 [-d] nodename-list"
	exit 1
fi

# Check the node list
if [ "`sinfo -hN -n $NODES`" = "" ]
then
	echo "ERROR: Nodelist $NODES does not exist"
	exit 1
fi

echo "Reboot nodes $NODES with nextstate=$action"
scontrol reboot asap nextstate=$action reason="$reason" $NODES 
sinfo -N -n $NODES