check_resque

#!/bin/bash
# Author: Martin Kammerlander, Phlegx Systems
# Email:  martin.kammerlander@phlegx.com
# Version: 0.1, 2011-05-14
# good example for bash nagios plugin: http://svn.coolcold.org/websvn/filedetails.php?repname=common&path=%2Fmonitoring%2Fnagios%2Fplugins%2Fmy%2Fcheck_tcp_sockets.sh
# Nagios Plugin API: http://nagios.sourceforge.net/docs/3_0/pluginapi.html

# TODO:
# 'ps' is not realiable. We need to check if resque daemons do offer PID mechanism
# Bring performance data into correct form
# Add cpu/memory warning and critcal options
# Add showing version number of resque
# Does ps really show the correct values for cpu/memory usage?

PROGNAME=`basename $0`


print_usage() {
        echo "Usage: $PROGNAME -n <Worker name> -a <number of workers> [-d]"
        echo "Options:"
        echo "-n - Worker/Queue [n]ame [COUNT=5 QUEUE=mailer rake resque:workers]  "
        echo "-a - Number/[a]mount of workers [COUNT=5 QUEUE=mailer rake resque:workers]"
        echo "-d - Show performance [d]ata"
}

print_help() {
        echo $PROGNAME
        echo ""
        print_usage
        echo ""
        echo "Detects if your resque workers are up and running. Shows CPU and Memory usage for every single worker-"
	echo "Returns perf data (cumulated CPU and Memory data of all workers) iff -d option activated."
        echo ""
        exit $UNKNOWN
}

### Parse Arguments

while getopts ":hV:n:a:d" Option; do
        case $Option in
                h)
                        print_help
                        exit 0
                        ;;
                V)
                        print_revision $PROGNAME
                        exit 0
                        ;;
                n)
                        worker_name=${OPTARG}
                        ;;

                a)
                        number_of_workers=${OPTARG}
                        ;;

                d)
                        show_perfdata="yes"
                        ;;
                                                                                        
                *)
                        print_help
                        exit 0
                        ;;
        esac
done
shift $(($OPTIND - 1))



function get_worker_values ()
{
for worker in "${arr[@]}"; do
	if [ "$worker" != "" ]; then
        	workers+="Worker $2 PID: " 
                workers+=`echo $worker | awk '{print $2}'`
                workers+=" - CPU: "
                workers+=`echo $worker | awk '{print $3}'`
		cpu=`echo $worker | awk '{print $3}'`
                workers+="% - Memory: "
                workers+=`echo $worker | awk '{print $4}'`
		memory=`echo $worker | awk '{print $4}'`
                workers+="%\n"
		# suggested by e36freak in channel #awk, not working yet
		# read -r cpu mem < <(awk '{cpu_total += $3; mem_total += $4} END {print cpu, mem}' <<<"$output")
		cpu_total=`echo $cpu $cpu_total| awk '{printf "%f", $1 + $2}'`
		memory_total=`echo $memory $memory_total| awk '{printf "%f", $1 + $2}'`
        fi
done
	# performance data is not yet in the correct form
	if [[ ("x$show_perfdata" == "xyes") ]]
    	then
       		workers+="|'cpu'=${cpu_total};;;0; 'memory'=${memory_total};;;0;"
    	fi
	echo -e $workers
	exit $exitcode
}


if [[ "x$worker_name" == "x" ]]; then
	processes=`ps aux | grep -w resque | grep -w Waiting | grep -v grep` 
else
	processes=`ps aux | grep -w resque | grep -w Waiting | grep "$worker_name" | grep -v grep`
fi

if [[ "x$number_of_workers" != "x" ]]; then
	if [ ! $(echo "$number_of_workers" | grep -E "^[0-9]+$") ]; then
		echo "CRITICAL - $number_of_workers is not a valid integer."
		exit 2
	fi

	if ! [ $number_of_workers -ge 1 ]; then
		echo "CRITICAL - $number_of_workers is an invalid value. Range is [>0]"
		exit 2
	fi
fi

if [ `echo ${#processes}` == 0 ]; then
	echo -e "CRITICAL - Worker(s) $worker_name not running."
	exit 2
else
	i=0
	while read -r li; do 
		arr+=("$li"); 
		i=$(expr $i + 1)
	done <<< "${processes//$1/$'\n'}"; 
	
	# decrease value for 1
	#i=$(expr $i - 1)
	if [[ "x$number_of_workers" != "x" ]]; then
		if [ "$i" != "$number_of_workers" ]; then
			workers="WARNING - Worker(s) $worker_name $i out of $number_of_workers running\n"
			exitcode=1
			get_worker_values workers "$worker_name" arr exitcode
		fi
	fi

	workers="OK - $i $worker_name Worker(s) up and running\n"
	exitcode=0
	get_worker_values workers "$worker_name" arr exitcode
fi