check_proc_ressources

#!/bin/bash
# ======================================================================
#
# NAGIOS CLIENT CHECK :: SHOW PROCESSES USING A LOT OF CPU OR MEMORY
#
# ----------------------------------------------------------------------
#
# SYNTAX:
#  ./check_proc_ressources -m METHOD [-w WARNLIMIT] [-c CRITICALLIMIT]
#    METHOD         string   identify what to check; one of cpu|mem
#    WARNLIMIT      integer  value in percent; default: 70
#    CRITICALLIMIT  integer  value in percent; default: 90
#
# ----------------------------------------------------------------------
# 2020-03-03  v1.0  inititial version
# 2020-03-05  v1.1  <axel.hahn@iml.unibe.ch> switch to ph.* helper functions
# 2020-06-06  v1.2  <axel.hahn@iml.unibe.ch> for cpu: limits are multiplicated with count of cpu 
# 2023-12-22  v1.3  <axel.hahn@unibe.ch>     update help; no temp files
# 2025-02-10  v1.4  <axel.hahn@unibe.ch>      harden sourcing files
# ======================================================================

# shellcheck source=/dev/null
. "$( dirname "$0" )/inc_pluginfunctions" || exit 1

export self_APPVERSION=1.4

# --- limits
iWarnLimitCpu=70
iCriticalLimitCpu=90

iWarnLimitMem=70
iCriticalLimitMem=90

typeset -i iCountWarning=0
typeset -i iCountCritical=0

typeset -i iCriticalLimit=0
typeset -i iWarnLimit=0
typeset -i iMulti=1

# ----------------------------------------------------------------------
# FUNCTIONS
# ----------------------------------------------------------------------

# show help
function showHelp(){
    cat <<EOH
$( ph.showImlHelpHeader )

SYNTAX:

  check_proc_ressources -m METHOD [-w WARNLIMIT] [-c CRITICALLIMIT]
  
OPTIONS:

  -h, --help
      show help

  -c, --critical VALUE
      critical value; defaults: 
        * cpu $iCriticalLimitCpu
        * mem $iCriticalLimitMem

  -m, --mode VALUE
      identify what to check; one of cpu|mem
      Remark: for cpu the given limits are multiplicated with count of cpu.

  -w. --warning VALUE
      warning limit value; defaults: 
        * cpu $iWarnLimitCpu
        * mem $iWarnLimitMem

EXAMPLE:

  check_proc_ressources -m mem -w 50 -c 70
      Show processes consuming more than 50% of memory; mark as critical
      when using 70% and more.

EOH
}

# ----------------------------------------------------------------------
# MAIN
# ----------------------------------------------------------------------


# ----- check limits

while [[ "$#" -gt 0 ]]; do case $1 in
    -h|--help)      showHelp; exit 0;;
    -m|--mode)      sMode=$2; shift ;shift;;
    -c|--critcal)   iCriticalLimit=$2; shift ;shift;;
    -w|--warning)   iWarnLimit=$2; shift ;shift;;
    *) echo "ERROR: Unknown parameter: $1"; showHelp; exit 1;
esac; done

case "${sMode}" in

    "cpu")
        scanfield="pcpu"
        unit="Cpu"

        test $iWarnLimit     -eq 0 && iWarnLimit=$iWarnLimitCpu
        test $iCriticalLimit -eq 0 && iCriticalLimit=$iCriticalLimitCpu
        ;;
    "mem")
        scanfield="pmem"
        unit="Memory"
        test $iWarnLimit     -eq 0 && iWarnLimit=$iWarnLimitMem
        test $iCriticalLimit -eq 0 && iCriticalLimit=$iCriticalLimitMem
        ;;
    *)
        echo ERRROR: [${sMode}] is an INVALID mode
        showHelp
        ph.abort

esac


if [ "$sMode" = "cpu" ]; then
    typeset -i iCpucount=$(cat /proc/cpuinfo | grep vendor_id | wc -l)
    if [ $iMulti -eq 0 ]; then
        ph.abort "ABORT: count of CPUs was not detected."
    fi
    iWarnLimit=$iWarnLimit*$iCpucount
    iCriticalLimit=$iCriticalLimit*$iCpucount
fi

# ----- read processlist and create helper table

out+=$(
    for line in $(ps -eo ${scanfield},pid,comm | awk -v limit=$iWarnLimit '$1 > limit {print $1":"$2":"$3 }' | sort -k 1 -t ":" -n -r )
    do

        Value=$(echo $line | cut -f 1 -d ':')
        iValue=$(echo $line | cut -f 1 -d ':' | sed "s#\..*##")
        iProcessId=$(echo $line | cut -f 2 -d ':')
        processname=$(echo $line | cut -f 3- -d ':')

        test $iValue -lt $iCountWarning  && break

        test $iValue -ge $iCountWarning  && status="Warning"
        test $iValue -ge $iCriticalLimit && status="Critical"

        # echo "${status}: $Value % $unit by process id $iProcessId $processname"
        printf "%-9s %5s %s %-10s %6s  %s\n" ${status} $Value % $unit $iProcessId $processname

    done
)


# ----- Status output

iCountCritical=$( grep -c "^Critical" <<< "$out" )
iCountWarning=$(  grep -c "^Warning" <<< "$out" )

if [ $iCountCritical -gt 0 ]; then
    ph.setStatus "critical"
    ph.status "$iCountCritical processes use $iCriticalLimit % (critical) or more"
    echo "$iCountWarning processes use $iWarnLimit % .. $iCriticalLimit % $unit"
    echo
    echo "Level       Usage        process id  process"
    echo "-----------------------------------------------------"
    echo "$out"
elif [ $iCountWarning -gt 0 ]; then
    ph.setStatus "warning"
    ph.status "$iCountWarning processes use $iWarnLimit % .. $iCriticalLimit % $unit"
    echo
    echo "Level       Usage        process id  process"
    echo "-----------------------------------------------------"
    echo "$out"
else
    ph.setStatus "ok"
    ph.status "all processes below warning limit $iWarnLimit % $unit .. (and critical limit $iCriticalLimit %)"
fi

# ----- additional infos

case "${sMode}" in
    "cpu")
        echo
        echo Cpus detected: $iCpucount
        ;;
    "mem")
        echo
        grep -E '^(MemTotal|MemFree|MemAvailable|Boffers|Cached):' /proc/meminfo
        ;;
esac

ph.exit

# ----------------------------------------------------------------------