Script called from monit to check replication delay

replication-delay.sh

#!/bin/bash

# Specify the threshold over which an error should be flagged
# Defaults to 60, but can be passed like ./script threshold
if [ -z $1 ]
then
	threshold=60
else
	threshold="$1"
	if [[ $threshold != +([0-9]) ]]
	then
		echo "The supplied threshold was not an integer. Ooops..." 1>&2
		exit 1
	fi
fi

# Specify the path to the file where pt-heartbeat writes the data
#filepath="pt-heartbeat"
filepath="/var/spool/pt-heartbeat"

if [ ! -f $filepath ]
then
	echo "Could not find the pt-heartbeat file. Eeeek..." 1>&2
	exit 1
fi

# Grep out the initial digits from the beginning of the line
# Format is ^0.00s [ 0.05s, 0.07s, 0.15s ]$
delay=$(cat $filepath | grep -oE '^[0-9]*')

# Check that $delay is an integer
if [[ $delay != +([0-9]) ]]
then
	# If it fails the first time, wait and try reading it again
	sleep 2.7
	delay=$(cat $filepath | grep -oE '^[0-9]*')

	# Check that $delay is an integer
	if [[ $delay != +([0-9]) ]]
	then
		# If it fails the second time, wait and try reading it again
		sleep 13.7
		delay=$(cat $filepath | grep -oE '^[0-9]*')

		# Check that $delay is an integer
		if [[ $delay != +([0-9]) ]]
		then
			echo "Could not extract a number from the pt-heartbeat file. Baaaad..." 1>&2
			exit 1
		fi
	fi
fi

# Check if the $delay is below the threshold
if [ "$delay" -lt "$threshold" ]
then
	exit 0
else
	echo "This slave is ${delay} seconds behind master, over threshold (${threshold})." 1>&2
fi

# By default, exit with an error code
exit 1