WIP: A Google Cloud startup-script to automatically revive preemptible compute instances.

google_compute_reviver.sh

#!/bin/bash
#
# GCloud startup script to auto-restart any instances with 'revive' tag.
# The calling machine must have Read/Write access to compute API!!
# I use this to reboot preemptible instances.
# Output is logged to /tmp/revive.log

indent() { sed 's/^/  /'; }

revive_instances() {
	# Go through lines in the provided string
	for line in "$1"; do
		echo "$line"
		# Instance name is the first word in the line.
		instance_name=`echo "$line" | head -n1 | awk '{print $1}'`
		instance_zone=`echo "$line" | head -n1 | awk '{print $2}'`
		# Attempt to reboot the instance
		echo "Rebooting '$instance_name' in zone '$instance_zone'..."
		gcloud compute instances start "--zone=$instance_zone" "$instance_name"
	done
}

auto_reviver () {
	REVIVE_TAG="$1"
	CHECK_INTERVAL="$2"
	LOG_FILE="$3"
	IFS=$'\n'
	date +"%F %T: monitoring instances with revive tag '$REVIVE_TAG', interval $CHECK_INTERVAL" >> "$LOG_FILE"
	while :; do
		# Look for instances with "revive" in their name/tags and TERMINATED status
		offline=`gcloud compute instances list --format='table(name,zone,status,tags.list())' | grep "$REVIVE_TAG" | grep "TERMINATED"`
		if [[ ! -z "$offline" ]] ; then
			# If we found some, reboot them
			date +"%F %T: some instances are down." >> "$LOG_FILE"
			revive_instances "$offline" | indent >> "$LOG_FILE"
		fi
		# Sleep for the check interval
		sleep $CHECK_INTERVAL
	done
}

# Make sure revive.log is readable by general users
printf '' >> "/tmp/revive.log"
chmod 644 "/tmp/revive.log"

# Run auto-reviver with tag "revive", check interval 2 minutes, logging
auto_reviver "revive" 120 "/tmp/revive.log"