check_mdstat.sh/check_mdstat.sh

201 lines
5.7 KiB
Bash
Executable File

#!/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="/sys/block"
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
INPUT="$INPUT/$TARGET_RAID_NAME/md"
if [ ! -d $INPUT ]; then
echo "CRITICAL: RAID $TARGET_RAID_NAME not found in $INPUT"
exit 2
fi
function log {
if $DEBUG; then
echo " > $@"
fi
}
NL=$'\n'
log "Reading from $INPUT"
STATE= # string
LEVEL= # int
SIZE_IN_BLOCKS="!" # int
CHUNK_SIZE_IN_BYTES= # size in bytes for chunks
NUM_DEVICES=0 # int with total number of devices in raid
NUM_ACTIVE_DEVICES=0 # int with number of active devices in raid
NUM_SPARE_DEVICES=0 # int with number of spare devices in raid
rYNC_ACTION= # "idle", "check" or "recover"
SYNC_PROGRESS= # float in percent
SYNC_SPEED= # sync speed in K/s
declare -A DEVICE_STATE # dict of strings. key = dev, value = state
declare -A FAULTY_DEVICES # dict with only faulty devices. key = dev, value = true
declare -A TO_REPLACE_DEVICES # dict of devices in raid that shall be replaced. key = dev, value = true
function getDataFromSys {
# https://mjmwired.net/kernel/Documentation/md.txt
LEVEL=$(<$INPUT/level)
NUM_DEVICES=$(<$INPUT/raid_disks)
CHUNK_SIZE_IN_BYTES=$(<$INPUT/chunk_size)
SIZE_IN_BLOCKS=$(<$INPUT/component_size)
STATE=$(<$INPUT/array_state)
SYNC_ACTION=$(<$INPUT/sync_action)
if [ "$SYNC_ACTION" != "idle" ]; then
SYNC_PROGRESS=$(<$INPUT/sync_completed)
SYNC_SPEED=$(<$INPUT/sync_speed)
fi
log "LEVEL = $LEVEL"
log "STATE = $STATE"
log "CHUNK_SIZE_IN_BYTES = $CHUNK_SIZE_IN_BYTES"
log "SIZE_IN_BLOCKS = $SIZE_IN_BLOCKS"
log "SYNC_ACTION = $SYNC_ACTION"
log "SYNC_PROGRESS = $SYNC_PROGRESS"
log "SYNC_SPEED = $SYNC_SPEED"
for dev in `find $INPUT -type d -name 'dev-*' -printf "%f\n"`; do
local deviceName=${dev:4}
local slot=$(<$INPUT/$dev/slot)
local state=$(<$INPUT/$dev/state)
log "$dev [$slot] is '$state'"
DEVICE_STATE[$deviceName]=$state
case "$state" in
*in_sync*) ((NUM_ACTIVE_DEVICES++));;
*writemostly*) ((NUM_ACTIVE_DEVICES++));;
*faulty*) FAULTY_DEVICES[$deviceName]=true;;
*blocked*) FAULTY_DEVICES[$deviceName]=true;;
*write-error*) TO_REPLACE_DEVICES[$deviceName]=true;;
*want_replacement*) TO_REPLACE_DEVICES[$deviceName]=true;;
*replacement*) ((NUM_SPARE_DEVICES++));;
*spare*) ((NUM_SPARE_DEVICES++));;
*) FAULTY_DEVICES[$deviceName]=true;;
esac
done
log "NUM_DEVICES = $NUM_DEVICES"
log "NUM_ACTIVE_DEVICES = $NUM_ACTIVE_DEVICES"
log "NUM_SPARE_DEVICES = $NUM_SPARE_DEVICES"
log "FAULTY_DEVICES = ${FAULTY_DEVICES[*]}"
log "TO_REPLACE_DEVICES = ${TO_REPLACE_DEVICES[*]}"
if [ "$SYNC_ACTION" != "idle" ]; then
SYNC_NUM=${SYNC_PROGRESS% /*}
log "'$SYNC_NUM'"
SYNC_DEN=${SYNC_PROGRESS#*/ }
log "'$SYNC_DEN'"
SYNC_PERCENT=$(($SYNC_NUM*100/$SYNC_DEN))
BLOCK_SIZE=$(<$(dirname $INPUT)/queue/hw_sector_size)
SYNC_REMAINING_S=$(( ($SYNC_DEN-$SYNC_NUM)*$BLOCK_SIZE/1024/$SYNC_SPEED))
SYNC_REMAINING_MIN=$(($SYNC_REMAINING_S/60))
fi
}
function printOutputAndExit {
local result=""
local info=""
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 [ ${#FAULTY_DEVICES[@]} -gt 0 ]; then
result="CRITICAL"
if [ -n "$info" ]; then info="$info. "; fi
info="${info}These devices failed: ${!FAULTY_DEVICES[@]}"
fi
if [ ${#TO_REPLACE_DEVICES[@]} -gt 0 ]; then
if [ -z "$result" ]; then result="WARNING"; fi
if [ -n "$info" ]; then info="$info. "; fi
info="${info}These devices should be replaced: ${!TO_REPLACE_DEVICES[@]}"
fi
if [ "$SYNC_ACTION" == "recover" ]; then
if [ -z "$result" ]; then result="WARNING"; fi
if [ -n "$info" ]; then info="$info. "; fi
info="${info}Recovering: $SYNC_PERCENT%, remaining ${SYNC_REMAINING_MIN}min"
fi
if [ "$SYNC_ACTION" == "check" ]; then
if [ -n "$info" ]; then info="$info. "; fi
info="${info}Checking: $SYNC_PERCENT%, remaining ${SYNC_REMAINING_MIN}min"
fi
if [ -z "$result" ]; then result="OK"; fi
echo -n "$result"
if [ -n "$info" ]; then
echo -n ": $info"
fi
echo -n " |"
echo -n " 'raid level: $LEVEL'=0"
echo -n " 'size in blocks'=$SIZE_IN_BLOCKS"
echo -n " 'num devices'=$NUM_DEVICES"
echo -n " 'num active devices'=$NUM_ACTIVE_DEVICES;;;0;$NUM_DEVICES"
echo -n " 'num failed devices'=${#FAILED_DEVICES[@]};;1;0;$NUM_DEVICES"
echo -n " 'num spare devices'=$NUM_SPARE_DEVICES;;;0"
if [ "$SYNC_ACTION" != "idle" ]; then
echo -n " 'sync progress'=$SYNC_NUM;0;;0;$SYNC_DEN"
echo -n " 'sync speed [1/s]'=${SYNC_SPEED}K"
echo -n " 'sync remaining'=${SYNC_REMAINING_S}s"
fi
for dev in ${!DEVICE_STATE[@]}; do
echo -n " 'dev $dev: ${DEVICE_STATE[$dev]}"
if [ -n "${FAULTY_DEVICE[$dev]}" ]; then
echo -n " faulty"
fi
if [ -n "${TO_REPLACE_DEVICE[$dev]}" ]; then
echo -n " wants replacement"
fi
echo -n "'=0"
done
echo
case $result in
OK) exit 0;;
WARNING) exit 1;;
CRITICAL) exit 2;;
esac
}
getDataFromSys
printOutputAndExit