add check_graphite-metric

Signed-off-by: honza <honza@inuits.eu>
inuits · May 14, 2017 · 21e2b67 · 21e2b67
1 parent f34170a
commit 21e2b67
Show file tree

Hide file tree

Showing 3 changed files with 279 additions and 0 deletions.
diff --git a/README.md b/README.md
@@ -304,6 +304,12 @@ An example Vagrant project has been included to get you started right away.
         <td><a href="https://github.com/honzatlusty">Jan Tlusty</a></td>
         <td><a href="https://github.com/honzatlusty/nagios-storm-topology-latency">upstream</a></td>
     </tr>
+    <tr>
+        <td>check_graphite-metric</td>
+        <td><a href="https://github.com/kali-hernandez">kali-hernandez</a></td>
+        <td><a href="https://github.com/kali-hernandez/nagios_plugins/blob/master/check_graphite_metric">upstream</a></td>
+    </tr>
+
 
 
 

diff --git a/build.txt b/build.txt
@@ -65,5 +65,6 @@ check_rabbitmq-sync.rb                1.0
 check_zmstatus.pl                     1.0
 check_ssl-cert                        1.0
 check_topology-latency.rb             1.0
+check_graphite-metric                 1.0
 
 # vim: set ts=2 sw=2 et : #
diff --git a/check_graphite-metric b/check_graphite-metric
@@ -0,0 +1,272 @@
+#!/bin/bash
+
+##
+## check_graphite_metric
+##
+## by kali (kali@kalimotxo.net) created for Sociomantic GmbH, 2014
+##
+##
+## Checks for values provided by graphite interface.
+## Accepts low and high ranges for warning and critical status.
+##
+## Depends on curl to retrieve the values from graphite interface.
+##
+## Specify the metric name with -m parameter, can use wildcards, but only ONE target per call.
+## Retrieves current values collection with a "from" graphite parameter specifying time in the past,
+##  which can be specified in parameters in the check. Useful for metrics with high intervals (e.g. 1 h)
+## Ignores values received as "None" which means graphite has no points collected for the specified metric/timeframe
+##
+## Returns UNKNOWN status if can't retreive values from graphite.
+##
+
+# Declare some defaults
+EXIT_NAGIOS_STATUS=-1
+CHECK_OUTPUT=""
+DEFAULT_NULL_STATUS=3
+# search string in metric to replace with the HOST parameter
+METRIC_HOST_MATCHSTRING="%HOST%"
+# default timeframe to check graphite for values
+GRAPHITE_FROM_VALUE="8minutes"
+# default needed curl options - allow special chars + silent + timeout 30 seconds
+CURL_OPTIONS="-g -s --max-time 30"
+# array to store metric names returned from graphite. used to keep graphite's response order
+declare -a RES_METRICS
+# assoc. array to store values for each metric. declared -A to iterate results and keep only the last not null value
+declare -A RES_VALUES
+# counter of passed threshold parameters
+HAS_LO_PARAMS=0
+HAS_HI_PARAMS=0
+
+# define exit functions for nagios
+exit_ok() {
+	echo "OK - ${CHECK_OUTPUT}"
+	exit 0
+}
+exit_war() {
+	echo "WARNING - ${CHECK_OUTPUT}"
+	exit 1
+}
+exit_cri() {
+	echo "CRITICAL - ${CHECK_OUTPUT}"
+	exit 2
+}
+exit_unk() {
+	echo "UNKNOWN - ${CHECK_OUTPUT}"
+	exit 3
+}
+exit_bad() {
+	echo "OUT OF RANGE - ${CHECK_OUTPUT}"
+	exit 255
+}
+# May this also be the inline documentation :)
+show_usage() {
+cat << EOF
+Usage: $0 -w <warn_lvl_high> -c <crit_lvl_high> [-H host] -m <graphite_metric_substring>
+-g|--graphite_url [url]  : Url (base) where graphite lives
+-G|--graphite_auth [u:p] : User / password to authenticate into graphite (http auth)
+-W|--warn_high [value]   : Value threshold to alert as warning when graphite goes above it (hi level)
+-C|--crit_high [value]   : Value threshold to alert as critical when graphite goes above it (hi/hi level)
+-w|--warn_low [value]    : (Optional) Lower warning value threshold (lo level) Ignored if absent.
+-c|--crit_low [value]    : (Optional) Lower critical value threshold (lo/lo level) Ignored if absent.
+-H|--host [host]         : (Optional) Host string in the format dc-NNN. If present, it will be replaced in the graphite metric.
+-m|--metric [metric]     : Graphite metric to be checked for values. If -H is present it will replace any occurence of the string "${METRIC_HOST_MATCHSTRING}" here.
+-n|--null_returns [code] : (Optional) Return specified status [OK|WARN|CRIT|UNK|0|1|2|3] when graphite gives NULL values only. Defaults to "UNK|3"
+-y|--hyst_warn [value]   : (Optional) Hysteresis for the warning level - how much value should change to turn off Warning
+-Y|--hyst_crit [value]   : (Optional) Hysteresis for the critical level - how much value should change to turn off Error
+-t|--timeframe [minutes] : (Optional) Set MINUTES threshold for metric inspection (e.g.: -t 5 --> checks metric values for 5 minutes in the past). Defaults to "8"
+-q|--quiet               : Quiet output. Returns only the status, and values only if the status is critical or warning.
+-h|--help                : Show usage.
+-d|--debug               : enable debug.
+Example:
+$0 -W 20 -C 40 -m "eu-002.cacti_propulsor.cacti_propulsor_cpu-0*.cpu_usage"
+WARNING - eu-002.cacti_propulsor.cacti_propulsor_cpu-02.cpu_usage=30.3966666667; eu-002.cacti_propulsor.cacti_propulsor_cpu-01.cpu_usage=22.8726666667; eu-002.cacti_propulsor.cacti_propulsor_cpu-03.cpu_usage=29.4873333333; eu-002.cacti_propulsor.cacti_propulsor_cpu-04.cpu_usage=27.258;
+
+EOF
+}
+
+check_status() {
+	local METRIC="${1}"
+	local ADDITION_CRIT="${2:-0}"
+	local ADDITION_WARN="${3:-0}"
+	local STATUS="0"
+	if [ $HAS_LO_PARAMS -eq 2 ] ; then
+		awk "BEGIN {exit !(${RES_VALUES["$METRIC"]} < ($WARN_LVL_LOW + $ADDITION_WARN))}" && STATUS=1
+		awk "BEGIN {exit !(${RES_VALUES["$METRIC"]} < ($CRIT_LVL_LOW + $ADDITION_CRIT))}" && STATUS=2
+	fi
+	if [ $HAS_HI_PARAMS -eq 2 ] ; then
+		awk "BEGIN {exit !(${RES_VALUES["$METRIC"]} > ($WARN_LVL_HIGH - ${ADDITION_WARN}))}" && STATUS=1
+		awk "BEGIN {exit !(${RES_VALUES["$METRIC"]} > ($CRIT_LVL_HIGH - ${ADDITION_CRIT}))}" && STATUS=2
+	fi
+	echo ${STATUS}
+}
+
+while (( "$#" )) ; do
+	case "${1}" in
+	-g|--graphite_url)
+		GRAPHITE_RENDER_URL="${2}/render/"
+		shift
+	;;
+	-G|--graphite_auth)
+		CURL_OPTIONS="${CURL_OPTIONS} -k -u ${2}"
+		shift
+	;;
+	-W|--warn_high)
+		WARN_LVL_HIGH=${2}
+		(( ++HAS_HI_PARAMS ))
+		shift
+	;;
+	-C|--crit_high)
+		CRIT_LVL_HIGH=${2}
+		(( ++HAS_HI_PARAMS ))
+		shift
+	;;
+	-w|--warn_low)
+		WARN_LVL_LOW=${2}
+		(( ++HAS_LO_PARAMS ))
+		shift
+	;;
+	-c|--crit_low)
+		CRIT_LVL_LOW=${2}
+		(( ++HAS_LO_PARAMS ))
+		shift
+	;;
+	-m|--host)
+		METRIC_STRING=${2}
+		shift
+	;;
+	-H|--metric)
+		HOST=${2}
+		shift
+	;;
+	-n|--null_returns)
+		case ${2^^} in
+			OK|0)
+				DEFAULT_NULL_STATUS=0
+			;;
+			WARN|WARNING|1)
+				DEFAULT_NULL_STATUS=1
+			;;
+			CRIT|CRITICAL|2)
+				DEFAULT_NULL_STATUS=2
+			;;
+			*)
+				DEFAULT_NULL_STATUS=3
+			;;
+		esac
+		shift
+	;;
+	-y|--hyst_warn)
+		HYSTERESIS_WARN=${2}
+		shift
+	;;
+	-Y|--hyst_crit)
+		HYSTERESIS_CRIT=${2}
+		shift
+	;;
+	-t|--timeframe)
+		GRAPHITE_FROM_VALUE="${2}minutes"
+		shift
+	;;
+	-q|--quiet)
+		QUIET_OUTPUT=1
+	;;
+	-d|--debug)
+		set -x
+	;;
+	-h|--help)
+		show_usage
+		exit 0
+	;;
+	*)
+		# silently ignore unknown parameters rather than rant about it and exit
+	;;
+	esac
+	shift
+done
+
+# build base curl options for querying graphite. build the target metric passed as parameter over it
+GRAPHITE_QUERY_BASE="${GRAPHITE_RENDER_URL}?format=raw&from=-${GRAPHITE_FROM_VALUE}&target="
+
+# make sure we have all the necessary parameters
+case $HAS_LO_PARAMS in
+	1)
+		if [ -z ${WARN_LVL_LOW} ] ; then
+			# Default warning level to critical if not given
+			WARN_LVL_LOW=${CRIT_LVL_LOW}
+			(( ++HAS_LO_PARAMS ))
+		else
+			CHECK_OUTPUT="Missing CRITICAL (low) level parameter. Passed -w=${WARN_LVL_LOW} -c=${CRIT_LVL_LOW}"
+			exit_bad
+		fi
+	;;
+	2)
+		if `awk "BEGIN {exit !(${WARN_LVL_LOW} < ${CRIT_LVL_LOW})}"` ; then
+			CHECK_OUTPUT="Low Warning threshold cannot be below critical threshold. Passed -w=${WARN_LVL_LOW} -c=${CRIT_LVL_LOW}"
+			exit_bad
+		fi
+	;;
+esac
+case $HAS_HI_PARAMS in
+	1)
+		if [ -z ${WARN_LVL_HIGH} ] ; then
+			WARN_LVL_HIGH=${CRIT_LVL_HIGH}
+			(( ++HAS_HI_PARAMS ))
+		else
+			CHECK_OUTPUT="Missing CRITICAL (high) level parameter. Passed -W=${WARN_LVL_HIGH} -C=${CRIT_LVL_HIGH}"
+			exit_bad
+		fi
+	;;
+	2)
+		if `awk "BEGIN {exit !(${WARN_LVL_HIGH} > ${CRIT_LVL_HIGH})}"` ; then
+			CHECK_OUTPUT="High Warning threshold cannot be over critical threshold. Passed -W=${WARN_LVL_HIGH} -C=${CRIT_LVL_HIGH}"
+			exit_bad
+		fi
+	;;
+esac
+
+## If HOST parameter is present, replace it in the metric variable
+[ "$HOST" != "" ] && METRIC_STRING=$(sed "s/$METRIC_HOST_MATCHSTRING/$HOST/g" <<< $METRIC_STRING)
+
+# Run query against graphite and store last non-none values on the array
+CURL_URI="${GRAPHITE_QUERY_BASE}${METRIC_STRING}"
+for GRAPHITE_OUT in $(curl ${CURL_OPTIONS} ${CURL_URI} | sed 's/ //g') ; do
+	if [[ "${GRAPHITE_OUT:0:1}" =~ ^\<.*|^\{.* ]] ; then
+		GLOBAL_STATE=3
+		CHECK_OUTPUT="Graphite returned an error, check your metric: ${METRIC_STRING}"
+		exit_unk
+	fi
+	GRAPHITE_METRIC=$(cut -f1 -d\| <<< $GRAPHITE_OUT | sed 's/,[0-9]*,[0-9]*,[0-9]*$//')
+	RES_METRICS+=("$GRAPHITE_METRIC")
+	for GRAPHITE_VALUE in $(cut -f2 -d\| <<< $GRAPHITE_OUT | sed 's/,/ /g') ; do
+		[ "${GRAPHITE_VALUE,,}" != "none" ] && RES_VALUES[${GRAPHITE_METRIC}]=${GRAPHITE_VALUE}
+	done
+done
+
+PREV_STATUS=${NAGIOS_LASTSERVICESTATEID:-0}
+if [ ${#RES_VALUES[@]} -eq 0 ] ; then
+	EXIT_NAGIOS_STATUS=$DEFAULT_NULL_STATUS
+	if [ ${DEFAULT_NULL_STATUS} -eq 3 ] ; then CHECK_OUTPUT="Graphite query ${METRIC_STRING} returned nothing." ; fi
+else
+	for METRIC in ${RES_METRICS[*]} ; do
+		METRIC_STATUS=0
+		if [[ "" == "${RES_VALUES["$METRIC"]}" ]] ; then
+			METRIC_STATUS=$DEFAULT_NULL_STATUS
+		else
+			if [[ ! -z ${HYSTERESIS_CRIT} && ${PREV_STATUS} == 2 ]] || [[ ! -z ${HYSTERESIS_WARN} && ${PREV_STATUS} == 1 ]]; then
+				METRIC_STATUS=$(check_status "${METRIC}" "${HYSTERESIS_CRIT}" "${HYSTERESIS_WARN}")
+			else
+				METRIC_STATUS=$(check_status "${METRIC}" 0)
+			fi
+		fi
+		[ $METRIC_STATUS -gt $EXIT_NAGIOS_STATUS ] && EXIT_NAGIOS_STATUS=$METRIC_STATUS
+		[ "${QUIET_OUTPUT}" != 1 ] || [ $METRIC_STATUS -gt 0 ] && CHECK_OUTPUT="${CHECK_OUTPUT}${METRIC}=${RES_VALUES[$METRIC]}; "
+	done
+fi
+
+case $EXIT_NAGIOS_STATUS in
+	0) exit_ok ;;
+	1) exit_war ;;
+	2) exit_cri ;;
+	3) exit_unk ;;
+	*) exit $EXIT_NAGIOS_STATUS ;;
+esac