#!/bin/bash
#
# Nagios plugin to check cluster synchronization process
#
# This check plugin is based on 3 files maintain by synchronization
# process :
#  - status file : Empty file updated on each successfully execution
#
#       Default : /var/run/sync_cluster.stat
#
#  - PID file : Present only when synchronization is running and 
#               contain the PID of the running process.
#
#       Default : /var/run/sync_cluster.pid
#
#  - Errors file : Present only when error(s) occured in previous
#                  synchronizations and contains details on this error(s).
#
#       Default : /var/run/sync_cluster.err
#
# Author : Benjamin Renard <brenard@easter-eggs.com>
# Date : Thu, 29 Nov 2018 15:41:48 +0100
# Source : http://gogs.zionetrix.net/check_sync_cluster
#

# Default configuration

# Files path template
TPL_FILE="/var/run/sync_cluster"

# WARNING/CRITICAL time in minutes for status file
WARN_STATUS_FILE=5
CRIT_STATUS_FILE=10

# WARNING/CRITICAL time in minutes for PID file
WARN_PID_FILE=5
CRIT_PID_FILE=10


function usage() {
	echo "Usage : $0 [-f files_path_format] [-wcWC]
	-h			Show this help message
	-f [files path format]	Specify files path template (default : $TPL_FILE)
	-w [minutes]		Specify WARNING time in minutes for status file (default : $WARN_STATUS_FILE)
	-c [minutes]		Specify CRITICAL time in minutes for status file (default : $CRIT_STATUS_FILE)
	-W [minutes]		Specify WARNING time in minutes for PID file (default : $WARN_PID_FILE)
	-C [minutes]		Specify CRITICAL time in minutes for PID file (default : $CRIT_PID_FILE)
"
}

while getopts "hf:w:c:W:C:" OPTION
do
	case "$OPTION" in
		f)
			TPL_FILE="${OPTARG}"
		;;
		w)
			WARN_STATUS_FILE="${OPTARG}"
		;;
		c)
			CRIT_STATUS_FILE="${OPTARG}"
		;;
		W)
			WARN_PID_FILE="${OPTARG}"
		;;
		C)
			CRIT_PID_FILE="${OPTARG}"
		;;
		h)
			usage
			exit 0
		;;
		*)
			echo "Invalid parameter -$OPTION"
			echo
			usage
			exit 1
		;;
	esac
done

STAT_FILE="$TPL_FILE.stat"
ERR_FILE="$TPL_FILE.err"
PID_FILE="$TPL_FILE.pid"

#
# Prepare return
#

EXIT_CODE=0
EXIT_STATUS=OK

ERROR_MSG=()
SUCCESS_MSG=()

function add_error() {
	level="$1"
	if [ $EXIT_CODE -ne 2 ]
	then
		case "$level" in
			CRITICAL)
				EXIT_CODE=2
				EXIT_STATUS=$level
				;;
			WARNING)
				EXIT_CODE=1
				EXIT_STATUS=$level
				;;
			UNKNOWN)
				if [ $EXIT_CODE -ne 1 ]
				then
					EXIT_CODE=3
					EXIT_STATUS=$level
				fi
				;;
		esac
	fi
	ERROR_MSG+=("$2")
}

function add_success() {
	SUCCESS_MSG+=("$1")
}

function join_by {
	local d=$1; shift; echo -n "$1"; shift; printf "%s" "${@/#/$d}";
}

####
# Check status file
####
if [ -r $STAT_FILE ];then

	T=$( ls -al --time-style=+%s $STAT_FILE|awk '{print $6}' )
	CUR=$( date +%s )
	let diff=CUR-T

	let WARN_STATUS_FILE_TIME=WARN_STATUS_FILE*60
	let CRIT_STATUS_FILE_TIME=CRIT_STATUS_FILE*60

	# CRITICAL if diff > $CRIT_STATUS_FILE
	if [ $diff -gt $CRIT_STATUS_FILE_TIME ]
	then
		add_error "CRITICAL" "status file not updated since more than ${CRIT_STATUS_FILE}min"
	# WARNING if diff > $WARN_STATUS_FILE_TIME
	elif [ $diff -gt $WARN_STATUS_FILE_TIME ]
	then
		add_error "WARNING" "status file not updated since more than ${WARN_STATUS_FILE}min"
	else
		add_success "status file updated since less than ${WARN_STATUS_FILE}min"
	fi
else
	add_error "UNKNOWN" "status file not present"
fi




####
# Check presence of error file
####
if [ -e "$ERR_FILE" ]
then
	add_error "WARNING" "some errors occured syncing cluster"
else
	add_success 'no error reported'	
fi

####
# Check pid file
####
if [ -r $PID_FILE ];then

	T=$( ls -al --time-style=+%s $PID_FILE|awk '{print $6}' )
	CUR=$( date +%s )
	let diff=CUR-T

	let WARN_PID_FILE_TIME=WARN_PID_FILE*60
	let CRIT_PID_FILE_TIME=CRIT_PID_FILE*60

	# CRITICAL if diff > $CRIT_PID_FILE
	if [ $diff -gt $CRIT_PID_FILE_TIME ]
	then
		add_error "CRITICAL" "PID file present since more than ${CRIT_PID_FILE}min"
	# WARNING if diff > $WARN_PID_FILE
	elif [ $diff -gt $WARN_PID_FILE_TIME ]
	then
		add_error "WARNING" "PID file present since more than ${WARN_PID_FILE}min"
	else
		add_success "PID file present since less than ${WARN_PID_FILE}min"
	fi
else
	add_success 'PID file not present'
fi


####
# Display return
###
echo "$EXIT_STATUS - $( join_by ', ' "${ERROR_MSG[@]}" "${SUCCESS_MSG[@]}" )"
echo

echo -n "Status file : $STAT_FILE"
if [ -e "$STAT_FILE" ]
then
	echo " ($( find $STAT_FILE -printf "%AY/%Am/%Ad %AX"|cut -d'.' -f1 ))"
else
	echo " (not present)"
fi

echo -n "Errors file : $ERR_FILE"
if [ -e "$ERR_FILE" ]
then
	echo " ($( find $ERR_FILE -printf "%AY/%Am/%Ad %AX"|cut -d'.' -f1 ))"
	echo
	echo "Errors :"
	cat $ERR_FILE|sed 's/^/\t/'
	echo
	echo " => After fixing this problem(s), please remove indicator file : $ERR_FILE"
	echo
else
	echo " (not present)"
fi

echo -n "PID file : $PID_FILE"
if [ -e "$PID_FILE" ]
then
	echo " ($( find $PID_FILE -printf "%AY/%Am/%Ad %AX"|cut -d'.' -f1 ))"
else
	echo " (not present)"
fi

exit $EXIT_CODE