[Nagiosplug-devel] check_sensors rework

[Mikael Lammentausta]

The official check_sensors plugin didn't report any info other than a regexp 'ALARM' from lm_sensors. I rewrote the plugin to send data, and have been making nice plots with nagios-grapher a few days now. 

This version reports cpu, motherboard and hard drive temperatures, fan speeds and voltages (in separate instances). Warning and critical level setting only affects cpu temperature and all fans.

By default this plugin should work like the earlier version, the switches modify the operation mode. See --help for more information.


--

#!/usr/bin/env bash
# gets temperature, fan speed and voltages from lm_sensors
# and smartctl for HDD temp
# Copyright (C) 2006 Mikael Lammentausta


PATH=/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin:/usr/local/sbin:/usr/nagios/sbin:/usr/local/nagios/sbin

PROGNAME=`basename $0`
PROGPATH=`echo $0 | sed -e 's,[\\/][^\\/][^\\/]*$,,'`
REVISION=`echo '$Revision: 2.0 $' | sed -e 's/[^0-9.]//g'`

#. $PROGPATH/utils.sh


print_usage() {
	echo "Usage: $PROGNAME [options]"
	echo "  e.g. $PROGNAME -T temperature -w 50 -c 60 -d /dev/hda,/dev/hdb"
	echo
	echo "Options:"
	echo -e "\t --help | -h          print help"
	echo -e "\t --version | -V       print version"
	echo -e "\t --verbose | -v       be verbose"
	echo -e "\t --type | -T [type]   "
	echo -e "\t    temperature       check and print temperature data"
	echo -e "\t    voltages          check and print voltage data"
	echo -e "\t    fan               check and print fan data"
	echo -e "\t -w [value]           set warning value"
	echo -e "\t -c [value]           set critical value"
	echo -e "\t -d [hdd drives]      set hard drive to scan, comma-separated"
	echo -e "\t -dd [driver type]    set hard drive type for smartctl (default ata)"

	echo
	echo "If no options are given, $PROGRAME will print only status."
	echo "NOTE: you can only ask one data type at once, $PROGNAME will"
	echo "      not output temp, voltage and fan info at the same time."
	echo "      Critical/warning level only watches the CPU temperature."
	echo "      You cannot set levels for voltages."
	echo
}

print_help() {
	print_revision $PROGNAME $REVISION
	echo 
	echo "This plugin checks hardware status using the lm_sensors package."
	echo 
	print_usage
	echo 
	support
	exit 3
}

# set defaults
check_temp=0
check_voltages=0
check_fan=0
check_smart=0
isverbose=0
hddtype="ata"



# get the data
sensordata=$(sensors 2>&1)
status=$?
# test status of lm_sensors
if [ ! "$(type -p sensors)" ]; then
	echo "SENSORS UNKNOWN - command not found (did you install lmsensors?)"
	exit -1
elif [ ! "$(type -p smartctl)" ] && [ ${check_smart} -eq 1 ]; then
	echo "Smartctl binary not found - no data available from hard drives"
	echo
	check_smart=0
elif test ${status} -ne 0 ; then
	echo "WARNING - sensors returned state $status"
	exit -1
elif [ $check_smart -eq 1 ] && [ ! "$hdd_drive" ]; then
	echo "No HDD drive defined. Use the -d switch."
	exit -1
fi


# all is ok
if [ $isverbose -eq 1 ]; then
	echo -e "${sensordata}"
fi


# put the multiple items given at the prompt to an array
form_hdd_array() {
	IFS_bak="$IFS"	
	IFS=","

	local drive
	hdd_drives[0]="reserved"

	for drive in ${1}; do

		hdd_drives[${#hdd_drives[@]}]="$drive"

	done

	IFS="$IFS_bak"
}


# check functions
check_temp() {
	cpu_temp="$(grep -i 'cpu temp' <<< "${sensordata}" | \
				grep -Eo '[0-9\.]+[[:punct:]]?[ ]?[CF]+' | head -n 1)" 

	mobo_temp="$(grep -i 'M/B temp' <<< "${sensordata}" | \
				 grep -Eo '\+[0-9]*[[:punct:]]?[ ]?[CF]+' | head -n 1)"

	local drive
	hdd_temp[0]="reserved"

	for drive in $(seq 1 1 ${#hdd_drives[@]}); do
		[ -e "${hdd_drives[$drive]}" ] && \
		hdd_temp[${#hdd_temp[@]}]="$(smartctl -A ${hdd_drives[$drive]} -d $hddtype | \
		      grep -i temperature | \
	          awk '{for (i=10; i<=NF; i++) printf("%s ",$i);printf ("\n")}') C"
	done
	}

# check fan speed
check_fan() {
	fans="$(grep -i 'fan' <<< "${sensordata}" | cut -d'(' -f 1 )"
	}

# check voltages
check_voltages() {

	# parse all voltages nicely and put into one string
	voltages="$(sensors | grep -E '[0-9]{2}\ V' | \
	            cut -d'(' -f 1 | sed 's/\ *//g ; s/:/\ =\ /' | \
				tr '\n' '; ' )"

	}


main() {
	# temperature
	if [ $check_temp -eq 1 ]; then
		check_temp
		local cpu_status=0

		# if either temp exceeds critical..
		[ "$critical" ] && \
			if [ $(grep -Eo '[0-9]+' <<< ${cpu_temp}) -ge $critical ]; then
#			   [ $(grep -Eo '[0-9]+' <<< ${mobo_temp}) -ge $critical ]; then

				echo -n "[CRITICAL]  "		
				cpu_status=2
			fi

		# if either temp exceeds warning...
		[ "$warning" ] && \
			if [ $(grep -Eo '[0-9]+' <<< ${cpu_temp}) -ge $warning ]; then
#			   [ $(grep -Eo '[0-9]+' <<< ${mobo_temp}) -ge $warning ]; then
				echo "[WARNING]  "
				cpu_status=1
			fi


		# output data
		echo -n "CPU = ${cpu_temp}  MOTHERBOARD = ${mobo_temp}  "

		for drive in $(seq 1 1 ${#hdd_drives[@]}); do
			echo -n "${hdd_drives[$drive]} = ${hdd_temp[$drive]}  "
		done

		exit $cpu_status
		

	# fan
	elif [ $check_fan -eq 1 ]; then
		check_fan
		local fan_status=0
		local worst_state=0
		IFS_bak="$IFS"
		IFS=$'\n'

		for fan in $fans; do
			fan_status=0
			fan_name="$(awk -F':' {'print $1'} <<< $fan )"
			speed="$(awk /[0-9]+/{'print $2'} <<< $fan)"

			# if a fan exceeds critical..
			[ "$critical" ] && \
				if [ $speed -le $critical ]; then
					fan_status=2
				fi

			# if a fan exceeds warning..
			[ "$warning" ] && \
				if [ $speed -le $warning ]; then
					fan_status=1
				fi
				
			# normal status -eq 0
			[ $fan_status -eq 2 ] && echo -n "[CRITICAL]: "
			[ $fan_status -eq 1 ] && echo -n "[WARNING]: "
			echo -n "${fan_name} = ${speed} RPM  "

			# remember worst state
			[ $fan_status -gt $worst_state ] && $worst_state=$fan_status

		done

		echo
		exit $worst_state


	# voltages
	elif [ $check_voltages -eq 1 ]; then
		# no critical/warning for voltages!
		check_voltages
		echo "$voltages"
		exit 0


	# default operation
	else
		if echo ${sensordata} | egrep ALARM > /dev/null; then
			echo SENSOR CRITICAL - Sensor alarm detected!
			exit 2
		else
			echo sensor ok
			exit 0
		fi
	fi
}


# parse cmd arguments
if [ "$#" -gt 0 ]; then
        while [ "$#" -gt 0 ]; do
		case "$1" in
			'--help'|'-h')
				print_help
				exit 3
				;;
			'--version'|'-V')
				print_revision $PROGNAME $REVISION
				exit 3
				;;
			'--verbose'|'-v')
				isverbose=1
				shift 1
				;;
			'-T'|'--type')
				case $2 in
					'temp'|'temperature')
						check_temp=1
						check_smart=1
						;;
					'voltages')
						check_voltages=1
						;;
					'fan')
						check_fan=1
						;;
					*)
						echo "Unknown type!"
						exit 3
						;;
				esac
				shift 2
				;;
			'-c')
				critical="$2"
				shift 2
				;;
			'-w')
				warning="$2"
				shift 2
				;;
			'-d')
				form_hdd_array "$2"
				# drives are now in $hdd_drives[]
				shift 2
				;;
			'-dd')
				hddtype="$2"
				;;
			*)
				echo "Unknown option!"
				print_usage
				exit 3
				;;
		esac
	done
fi

main
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://nagios-plugins.org/archive/devel/attachments/20060908/32583506/attachment.html>