contrib-munin/plugins/healthcheck/healthcheck_log

#!/bin/bash
#
#healthcheck on munin
#egrep system log and alert.
#
#programed by rti (hiroyuki fujie) super.rti@gmail.com @super_rti
#LICENSE: NYSL (public domain)
#
#
#config file
#      /etc/munin/plugin-conf.d/munin-node
#
#example minimum config
#---------------------------------------------------
#[healthcheck_log]
#user root
#env.log_1  /var/log/messages
#---------------------------------------------------
#
#check two log
#---------------------------------------------------
#[healthcheck_log]
#user root
#env.log_1  /var/log/messages
#env.log_2  /var/log/syslog
#---------------------------------------------------
#
#check two three
#---------------------------------------------------
#[healthcheck_log]
#user root
#env.log_1  /var/log/messages
#env.log_2  /var/log/syslog
#env.log_3  /var/log/dmesg
#---------------------------------------------------
#
#set name
#---------------------------------------------------
#[healthcheck_log]
#user root
#env.log_1  /var/log/messages
#env.name_1 my_server_messages
#---------------------------------------------------
#
#set egrep string
#---------------------------------------------------
#[healthcheck_log]
#user root
#env.log_1  /var/log/messages
#env.grep_1 alert|warning
#---------------------------------------------------
#
#set egrep string
#---------------------------------------------------
#[healthcheck_log]
#user root
#env.log_1  /var/log/messages
#env.grep_1 alert|warning
#---------------------------------------------------
#
#full option
#/etc/munin/plugin-conf.d/munin-node
#---------------------------------------------------
#[healthcheck_log]
#user root                          #log file is read only root user.
#env.log_1  /var/log/messages       #target log filename
#env.grep_1 critical|error          #egrep string.
                                    #defualt by critical|error|warning|crash|fatal|kernel
#---------------------------------------------------
#


#edakari speed up.
CHECKMAX=`env | grep log_ | wc -l`
let CHECKMAX="$CHECKMAX + 1"
MINUTE_BY_GREP_RANGE=10

if [ "$1" = "autoconf" ]; then
    if [ $CHECKMAX -le 1 ]; then
         echo no
         exit 1
    fi
    echo yes
    exit 0
fi

if [ "$1" = "config" ]; then
    echo 'graph_title log grep (match count)'
    echo "graph_args --base 1000 -l 0 --vertical-label match_count"
    echo 'graph_scale no'
    echo 'graph_vlabel match_count'
    echo 'graph_category healthcheck'
    echo 'graph_info This graph shows the bad event count on log'

    for(( I = 1; I < $CHECKMAX; ++I ))
    do
         eval log=\$log_${I}
         eval name=\$name_${I}
         eval grep=\$grep_${I}
         if [ "x${log}" = "x" ]; then
              continue
         fi
         if [ "x${name}" = "x" ]; then
             name=`echo $log | sed 's#[/|\.]#_#g'`
         fi
         if [ "x${name}" = "x" ]; then
             grep="critical|error|crash|fatal|kernel"
         fi

         echo "$name.label $name"
         echo "$name.info egrep $grep $log | wc -l"
         echo "$name.draw LINE2"
         echo "$name.min 0"
         echo "$name.max 20"
         echo "$name.critical 0:0"
    done

    exit 0
fi

NOWTIME=`date --date "$MINUTE_BY_GREP_RANGE minute ago" +%s`

for(( I = 1; I < $CHECKMAX; ++I ))
do
    eval log=\$log_${I}
    eval name=\$name_${I}
    eval grep=\$grep_${I}
    if [ "x${log}" = "x" ]; then
         continue
    fi
    if [ "x${name}" = "x" ]; then
         name=`echo $log | sed 's#[/|\.]#_#g'`
    fi
    if [ "x${grep}" = "x" ]; then
         grep="critical|error|crash|fatal|kernel"
    fi

    COUNT=0
    MESSAGE=
    IFS=$'\n'
    MATCHLINES=(`egrep -i "$grep" "$log"`)
    for(( N = ${#MATCHLINES[@]} - 1; N >= 0 ; --N ))
    do
         LINE=${MATCHLINES[$N]}
         DATESTRING=`echo $LINE | awk '{ printf("%s %s %s",$1,$2,$3)}'`
         LOGTIME=`date --date "$DATESTRING" +%s`
         if [ $LOGTIME -lt $NOWTIME ]; then
               break
         fi
         let COUNT="$COUNT + 1"
         MESSAGE="$MESSAGE$LINE //@LINE@// "
    done


    if [ $COUNT -eq 0 ]; then
        echo "${name}.value 0"
    else
        echo "${name}.value ${COUNT}"
        echo "${name}.extinfo ${MESSAGE}"
    fi
done