Implemented check script
This commit is contained in:
parent
ba70f3b737
commit
39214f2907
1 changed files with 193 additions and 36 deletions
229
check_mdstat.sh
229
check_mdstat.sh
|
|
@ -1,47 +1,204 @@
|
||||||
#!/bin/bash
|
#!/bin/bash
|
||||||
#
|
#
|
||||||
# Created by Sebastian Grewe, Jammicron Technology
|
# Created by Sebastian Grewe, Jammicron Technology
|
||||||
# Changes By Jasem Elayeb on 02.03.2016
|
# 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 Name RAID_DISKS
|
||||||
# JE: add Physical Disks Status DISKS_STATUS
|
# JE: add Physical Disks Status DISKS_STATUS
|
||||||
# JE: add Array Names RAID_ARRAY
|
# JE: add Array Names RAID_ARRAY
|
||||||
|
# J007: Only check a single array
|
||||||
|
|
||||||
# Get count of raid arrays
|
|
||||||
RAID_DEVICES=`grep ^md -c /proc/mdstat`
|
|
||||||
|
|
||||||
# Get count of degraded arrays
|
if [ $# -lt 1 -o -z "$1" ]; then
|
||||||
RAID_STATUS=`grep "\[.*_.*\]" /proc/mdstat -c`
|
echo "ERROR: pass raid name as argument"
|
||||||
|
exit 1;
|
||||||
|
fi
|
||||||
|
while [[ $# -gt 0 ]]; do
|
||||||
|
case $1 in
|
||||||
|
-r|--raid) TARGET_RAID_NAME="$2";;
|
||||||
|
-?|--help)
|
||||||
|
echo "Check /proc/mdstat. Arguments:"
|
||||||
|
echo "--raid NAME: Raid name, e.g. md0"
|
||||||
|
;;
|
||||||
|
esac
|
||||||
|
shift
|
||||||
|
done
|
||||||
|
|
||||||
# Is an array currently recovering, get percentage of recovery
|
if [ -z "$TARGET_RAID_NAME" ]; then
|
||||||
RAID_RECOVER=`grep recovery /proc/mdstat | awk '{print $4}'`
|
echo "ERROR: pass --raid as argument"
|
||||||
|
exit 255
|
||||||
# Is an array currently resyncing, get percentage of resync
|
|
||||||
|
|
||||||
RAID_RESYNC=`grep resync /proc/mdstat | awk '{print $4}'`
|
|
||||||
|
|
||||||
RAID_ARRAY=`awk '/md[1-9]/{for (i=1;i<=NF;++i) if ($i~/md[1-2]/)print $i}' /proc/mdstat |xargs`
|
|
||||||
RAID_DISKS=`awk '/sd[a-z]/{for (i=1;i<=NF;++i) if ($i~/sd[a-z]/)print $i}' /proc/mdstat |xargs`
|
|
||||||
DISKS_STATUS=`grep algorithm /proc/mdstat|awk '{print $12}'`
|
|
||||||
|
|
||||||
# Check raid status
|
|
||||||
# RAID recovers --> Warning
|
|
||||||
if [[ $RAID_RECOVER ]]; then
|
|
||||||
STATUS="WARNING - Checked $RAID_DEVICES arrays $RAID_ARRAY, recovering : $RAID_RECOVER"
|
|
||||||
EXIT=1
|
|
||||||
# RAID resync --> Warning
|
|
||||||
elif [[ $RAID_RESYNC ]]; then
|
|
||||||
STATUS="WARNING - Checked $RAID_DEVICES arrays $RAID_ARRAY., resyncing : $RAID_RESYNC"
|
|
||||||
EXIT=1
|
|
||||||
# RAID ok
|
|
||||||
elif [[ $RAID_STATUS == "0" ]]; then
|
|
||||||
STATUS="OK - Checked $RAID_DEVICES arrays $RAID_ARRAY."
|
|
||||||
EXIT=0
|
|
||||||
# All else critical, better save than sorry
|
|
||||||
else
|
|
||||||
STATUS="CRITICAL - Checked $RAID_DEVICES arrays $RAID_ARRAY, $RAID_STATUS have FAILED"
|
|
||||||
EXIT=2
|
|
||||||
fi
|
fi
|
||||||
|
|
||||||
# Status and quit
|
function log {
|
||||||
echo -e "$STATUS \n Physical Disks: $RAID_DISKS Disks Status: $DISKS_STATUS "
|
local x=
|
||||||
exit $EXIT
|
#echo " > $@"
|
||||||
|
}
|
||||||
|
|
||||||
|
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
|
||||||
|
DEVICES= # strings separated by whitespace
|
||||||
|
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 $x3; do
|
||||||
|
local dev2="${dev1%[*}"
|
||||||
|
if [ -n "$DEVICES" ]; then DEVICES="$DEVICES "; fi
|
||||||
|
DEVICES="${DEVICES}$dev2"
|
||||||
|
done
|
||||||
|
|
||||||
|
log "STATE = $STATE"
|
||||||
|
log "LEVEL = $LEVEL"
|
||||||
|
log "DEVICES = $DEVICES"
|
||||||
|
}
|
||||||
|
|
||||||
|
SIZE_IN_BLOCKS="!" # int
|
||||||
|
NUM_DEVICES=0 # int with number of devices in raid
|
||||||
|
BAD_DEVICES=0 # int with number of "_" in "[UUU_UU]"
|
||||||
|
function parseConfigStatusLine {
|
||||||
|
# This function is called for 1st line after the raid definition line
|
||||||
|
# It parses the SIZE_IN_BLOCKS
|
||||||
|
local line="$1"
|
||||||
|
line="${line:6}" # trim left
|
||||||
|
SIZE_IN_BLOCKS=${line%% *}
|
||||||
|
|
||||||
|
local lastWord=${line##* } # Get the last word
|
||||||
|
lastWord=${lastWord:1:-1} # Remove first and last char
|
||||||
|
NUM_DEVICES=${#lastWord} # str length = num devices
|
||||||
|
lastWord=${lastWord//U/} # Remove all U, so only _ remain
|
||||||
|
BAD_DEVICES=${#lastWord} # str length = num bad devices
|
||||||
|
|
||||||
|
|
||||||
|
log "SIZE_IN_BLOCKS = $SIZE_IN_BLOCKS"
|
||||||
|
log "NUM_DEVICES = $NUM_DEVICES"
|
||||||
|
log "BAD_DEVICES = $BAD_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 = $RECOVERY_INFO"
|
||||||
|
}
|
||||||
|
|
||||||
|
function printOutputAndExit {
|
||||||
|
# STATE, LEVEL, DEVICES, SIZE_IN_BLOCKS, NUM_DEVICES, BAD_DEVICES, RECOVERY_PROGRESS, RECOVERY_INFO
|
||||||
|
|
||||||
|
local result="OK"
|
||||||
|
local info=""
|
||||||
|
|
||||||
|
if [ -n "$RECOVERY_PROGRESS" ]; then
|
||||||
|
result="WARNING"
|
||||||
|
if [ -n "$info" ]; then info="$info. "; fi
|
||||||
|
info="${info}Recovering: $RECOVERY_PROGRESS $RECOVERY_INFO"
|
||||||
|
fi
|
||||||
|
|
||||||
|
if [ "$STATE" != "active" -a "$STATE" != "started" ]; then
|
||||||
|
result="CRITICAL"
|
||||||
|
if [ -n "$info" ]; then info="$info. "; fi
|
||||||
|
info="${info}State is $STATE"
|
||||||
|
fi
|
||||||
|
|
||||||
|
if [ $BAD_DEVICES -gt 0 ]; then
|
||||||
|
result="CRITICAL"
|
||||||
|
if [ -n "$info" ]; then info="$info. "; fi
|
||||||
|
info="${info}Missing $NUM_DEVICES devices"
|
||||||
|
fi
|
||||||
|
|
||||||
|
echo -n "$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 bad devices'=$BAD_DEVICES;;1;0;$NUM_DEVICES"
|
||||||
|
if [ -n "$RECOVERY_PROGRESS" ]; then
|
||||||
|
echo -n " 'recovery progress'=$RECOVERY_PROGRESS%;0.1;;0;100"
|
||||||
|
fi
|
||||||
|
echo
|
||||||
|
|
||||||
|
case $result in
|
||||||
|
OK) exit 0;;
|
||||||
|
WARNING) exit 1;;
|
||||||
|
CRITICAL) exit 2;;
|
||||||
|
esac
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
|
HAD_TARGET_RAID=0
|
||||||
|
|
||||||
|
while IFS= read -r line; do
|
||||||
|
#echo " : $line"
|
||||||
|
|
||||||
|
setCurrentRaid "$line"
|
||||||
|
|
||||||
|
if [ "$CURRENT_RAID" == "$TARGET_RAID_NAME" ]; then
|
||||||
|
case $HAD_TARGET_RAID in
|
||||||
|
0) setStateAndDevicesAndLevel "$line";;
|
||||||
|
1) parseConfigStatusLine "$line";;
|
||||||
|
2) parseBitmapOrRecoveryLine "$line";;
|
||||||
|
3) parseRecoveryLine "$line";;
|
||||||
|
esac
|
||||||
|
HAD_TARGET_RAID=$(($HAD_TARGET_RAID+1))
|
||||||
|
elif [ $HAD_TARGET_RAID -gt 0 ]; then
|
||||||
|
printOutputAndExit
|
||||||
|
fi
|
||||||
|
done < /proc/mdstat
|
||||||
|
|
||||||
|
echo "CRITICAL: RAID $TARGET_RAID_NAME not found in mdstat"
|
||||||
|
exit 2
|
||||||
|
|
||||||
|
|
|
||||||
Loading…
Reference in a new issue