check_mdstat.sh/check_mdstat.sh

#!/bin/bash
#
# Created by Sebastian Grewe, Jammicron Technology
# Changes by Jasem Elayeb on 02.03.2016
# Chagnes by Jonny007-MKD on 06.02.2020
# JE: add Physical Disks Name RAID_DISKS
# JE: add Physical Disks Status DISKS_STATUS
# JE: add Array Names RAID_ARRAY
# J007: Only check a single array


if [ $# -lt 1 -o -z "$1" ]; then
	echo "ERROR: pass raid name as argument"
	exit 1;
fi

INPUT="/proc/mdstat"
DEBUG=false
while [[ $# -gt 0 ]]; do
	case $1 in
		-r|--raid)	TARGET_RAID_NAME="$2"; shift;;
		-i|--input) INPUT="$2"; shift;;
		-d|--debug) DEBUG=true;;
		-?|--help)
			echo "Check /proc/mdstat. Arguments:"
			echo "--raid NAME: Raid name, e.g. md0"
			echo "--input FILE: Read from this file. Default: /proc/mdstat"
			exit 0
			;;
	esac
	shift
done

if [ -z "$TARGET_RAID_NAME" ]; then
	echo "ERROR: pass --raid as argument"
	exit 255
fi

function log {
	if $DEBUG; then
		echo "  > $@"
	fi
}
function logLine {
	if $DEBUG; then
		echo "  : $@"
	fi
}
NL=$'\n'

CURRENT_RAID=	# string
function setCurrentRaid {
	# This function is called for each line in $1
	# If the line does not start with a whitespace, it denotes a new RAID device and CURRENT_RAID is set
	# Also, the state and the devices are parsed 
	local line="$1"
	if [[ $line =~ ^[^\s].+:.+ ]]; then
		local x1="${line%:*}" # Remove : suffix
		local x2="${x1% *}"   # Trim right
		CURRENT_RAID="$x2"
		log "SET CURRENT RAID"
	fi
	log "CURRENT RAID: $CURRENT_RAID"
}

STATE=		# "active", "started", "inactive"
LEVEL=		# int
declare -A DEVICES	# dict of strings. key = index, value = dev name
declare -A FAILED_DEVICES
function setStateAndDevicesAndLevel {
	# This functions is called for the line that starts a new RAID device by setCurrentRaid, with the complete line in $1
	# It parses the STATE, the LEVEL and the DEVICES
	local line="$1"
	local x1="${line#*: }" 				# Remove : prefix
	STATE="${x1%% *}" 					# Only the first word
	local x2="${x1:$((${#STATE}+1))}"	# Remove the first word
	LEVEL="${x2%% *}" 					# Only the first word
	local x3="${x2:$((${#LEVEL}+1))}"	# Remove the first word

	LEVEL="${LEVEL:4}"					# Remove the raid prefix

	for dev1 in $x3; do
		local devName="${dev1%[*}"		# device name
		local dev2="${dev1#*[}"			# all after device name
		local devIndex="${dev2%]*}"		# device index
		local dev3="${dev2#*]}"			# all after device index
		local devFailed=false
		DEVICES[$devIndex]="$devName"
		FAILED_DEVICES[$devIndex]=false
		if [ -n "$dev3" ]; then
			FAILED_DEVICES[$devIndex]=true
		fi
	done

	log "STATE = $STATE"
	log "LEVEL = $LEVEL"
	log "DEVICES = ${DEVICES[@]}"
}

SIZE_IN_BLOCKS="!"			# int
NUM_DEVICES=0				# int with total number of devices in raid
NUM_ACTIVE_DEVICES=0		# int with number of active devices in raid
function parseConfigStatusLine {
	# This function is called for 1st line after the raid definition line
	# It parses the SIZE_IN_BLOCKS
	local line="$1"
	line=$(echo $line)	# trim left
	SIZE_IN_BLOCKS=${line%% *}

	local x1=${line##* }		# Get the last word [UU_]
	local x2=${line% *}			# Remove the last word
	local x3=${x2##* }			# Get the last word [3/2]
	x3=${x3:1:-1}
	NUM_DEVICES=${x3%/*}
	NUM_ACTIVE_DEVICES=${x3#*/}


	log "SIZE_IN_BLOCKS = $SIZE_IN_BLOCKS"
	log "NUM_DEVICES = $NUM_DEVICES"
	log "NUM_ACTIVE_DEVICES = $NUM_ACTIVE_DEVICES"
	# TODO
}

function parseBitmapOrRecoveryLine {
	# This function is called for the 2nd line after the raid definition line
	# It checks whether it contains a bitmap or a recovery line
	if [[ "$1" == *bitmap* ]]; then
		parseBitmapLine "$1"
	fi
	if [[ "$1" == *recovery* ]]; then
		parseRecoveryLine "$1"
	fi
}

function parseBitmapLine {
	local x
	# TODO
}

RECOVERY_PROGRESS=	# float in percent
RECOVERY_INFO=		# finish and speed
function parseRecoveryLine {
	# This function is called for the 2nd or 3rd lineafter the raid definition line
	# It sets the RECOVERY_PROGRESS
	local line="$1"
	
	local x1="${line##*recovery = }"
	local x2="${x1%%%*}"
	RECOVERY_PROGRESS="$x2"
	
	local x3="${line##*finish=}"
	RECOVERY_INFO="finish=$x3"

	log "RECOVERY_PROGESS = $RECOVERY_PROGRESS"
	log "RECOVERY_INFO = $RECOVERY_INFO"
}

function printOutputAndExit {
	# STATE, LEVEL, DEVICES, SIZE_IN_BLOCKS, NUM_DEVICES, BAD_DEVICES, RECOVERY_PROGRESS, RECOVERY_INFO

	local result="OK"
	local info=""
	local numFailedDevices=0
	local numSpareDevices=0

	if [ $NUM_ACTIVE_DEVICES -lt $NUM_DEVICES ]; then
		result="CRITICAL"
		if [ -n "$info" ]; then info="$info. "; fi
		info="${info}Missing $(($NUM_DEVICES - $NUM_ACTIVE_DEVICES)) of $NUM_DEVICES devices"
	fi

	if [ -n "$RECOVERY_PROGRESS" ]; then
		result="WARNING"
		if [ -n "$info" ]; then info="$info. "; fi
		info="${info}Recovering: progress=$RECOVERY_PROGRESS% $RECOVERY_INFO"
	fi

	local devIndices="$(printf "%s\n" "${!DEVICES[@]}" | sort -n)"
	for devIndex in $devIndices; do
		if [ -z "${DEVICES[$devIndex]}" ]; then
			((numSpareDevies++))
			continue
		fi
		if ${FAILED_DEVICES[$devIndex]}; then
			result="CRITICAL"
			((numFailedDevices++))
			if [ -n "$info" ]; then info="$info. "; fi
			info="${info}Device ${DEVICES[$devIndex]} failed"
		fi
	done

	if [ "$STATE" != "active" -a "$STATE" != "started" ]; then
		result="CRITICAL"
		if [ -n "$info" ]; then info="$info. "; fi
		info="${info}State is $STATE"
	fi

	echo -n "$result"
	if [ -n "$info" ]; then
		echo -n ": $info"
	fi

	echo -n " |"
	echo -n " 'raid level'=$LEVEL"
	echo -n " 'size in blocks'=$SIZE_IN_BLOCKS"
	echo -n " 'num devices'=$NUM_DEVICES"
	echo -n " 'num failed devices'=$numFailedDevices;;1;0;$NUM_DEVICES"
	echo -n " 'num spare devices'=$numSpareDevices;;;0;$NUM_DEVICES"
	if [ -n "$RECOVERY_PROGRESS" ]; then
		echo -n " 'recovery progress'=$RECOVERY_PROGRESS%;0;;0;100"
	fi
	for devIndex in $devIndices; do
		if [ -z "${DEVICES[$devIndex]}" ]; then
			echo -n " 'device $devIndex: spare'=$devIndex"
			continue
		fi
		if ${FAILED_DEVICES[$devIndex]}; then
			echo -n " 'device $devIndex: ${DEVICES[$devIndex]} failed'=$devIndex;;$devIndex'"
			continue
		fi
		echo -n " 'device $devIndex: ${DEVICES[$devIndex]}'=$devIndex"
	done

	echo

	case $result in
		OK)       exit 0;;
		WARNING)  exit 1;;
		CRITICAL) exit 2;;
	esac
}


HAD_TARGET_RAID=0

while IFS= read -r line; do
    logLine "$line"

	setCurrentRaid "$line"

	if [ "$CURRENT_RAID" == "$TARGET_RAID_NAME" ]; then
		case $HAD_TARGET_RAID in
			0) setStateAndDevicesAndLevel "$line";;
			1) parseConfigStatusLine "$line";;
			*) parseBitmapOrRecoveryLine "$line";;
		esac
		((HAD_TARGET_RAID++))
	elif [ $HAD_TARGET_RAID -gt 0 ]; then
		printOutputAndExit
	fi
done < $INPUT

echo "CRITICAL: RAID $TARGET_RAID_NAME not found in $(basename $INPUT)"
exit 2