nodesuspend

#!/usr/bin/env bash

# Slurm nodes suspend/resume script
# Helper scripts required: power_ipmi power_azure power_noaction

# We must define some node features power_xxx in slurm.conf, for example for IPMI and Azure cloud:
# NodeName=node[001-100] Feature=xeon2650v4,opa,xeon24,power_ipmi
# NodeName=cloud[001-100] Feature=xeon8272cl,power_azure
# We can read the node features using:
#   sinfo -hN -O "Nodelist: ,Features:" --nodes=$* | uniq

# Define the action
action="-s"

# The environment variables passed from slurmctld do not include USER etc.
export USER=`whoami`
export PATH=/usr/local/bin:$PATH

# Check for empty argument list
if [[ $# < 1 ]]
then
	exit 0
fi

# Get the node list including features (in a single call to minimize load on slurmctld)
TMPFILE=`mktemp`
sinfo -hN -O "Nodelist: ,Features:" --nodes=$* > $TMPFILE

# We require the "nodeset" command from the ClusterShell package.  Install it by:
# yum install epel-release
# yum install clustershell

# Node suspend/resume DUMMY action used only for testing the power_save module
# Select the nodelist with the "power_noaction" feature
nodelist=`grep power_noaction $TMPFILE | uniq | awk '{print $1}' | nodeset --fold`
if [[ -n "$nodelist" ]]
then
	power_noaction $action $nodelist
fi

# Node suspend/resume by IPMI
# Select the nodelist with the "power_ipmi" feature
nodelist=`grep power_ipmi $TMPFILE | uniq | awk '{print $1}' | nodeset --fold`
if [[ -n "$nodelist" ]]
then
	power_ipmi $action $nodelist
fi

# Execute Azure VM nodes suspend/deallocate 
# Select the nodelist with the "power_azure" feature
nodelist=`grep power_azure $TMPFILE | uniq | awk '{print $1}' | nodeset --fold`
if [[ -n "$nodelist" ]]
then
	power_azure $action $nodelist
fi

# Cleanup
rm -f $TMPFILE