check_dns_responsetime

#!/bin/bash
# ======================================================================
#
# Check DNS response time
# requirements:
# - netstat
#
# ----------------------------------------------------------------------
# 2020-06-17  v1.0  <axel.hahn@iml.unibe.ch>
# 2021-11-12  v1.1  <axel.hahn@iml.unibe.ch>  check fqd instead of fixed value
#                                             test tcp 53 first
#                                             check result depends on 1st nameserver only
# ======================================================================


. `dirname $0`/inc_pluginfunctions
tmpfile=/tmp/check_netstat_out_$$
infofile=/tmp/check_netstat_out_2_$$
myHost=$( hostname -f )

echo $myHost | cut -f 3- -d "." | grep "\." >/dev/null
if [ $? -ne 0 ]; then
    ph.setStatus unknown
    ph.status "DNS check for [$myHost] - SKIP: hostname -f returned a FQDN with less than 2 dots"
    ph.exit
fi

# set default / override from command line params
typeset -i iWarnLimit=`     ph.getValueWithParam  300 w "$@"`
typeset -i iCriticalLimit=` ph.getValueWithParam 1000 c "$@"`

rm -f $tmpfile $infofile 2>/dev/null

typeset -i iMax=0
typeset -i iTime=0

typeset -i iCounter=0
typeset -i iNotReachable=0

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


# --- check param -h
if [ "$1" = "-h" ]; then
    echo "
    usage: $0 [ -w value -c value -h ]

        -w  Warning level
        -c  Critical level
        -h  this help
    "
    exit 0
fi

for mydns in `grep ^nameserver /etc/resolv.conf | awk '{ print $2 } ' `
do
    iCounter=$iCounter+1
    typeset -i iSrvMax=0
    echo "" >>$infofile

    echo "---------- $iCounter - $mydns " >>$infofile
    # todo loop
    >/dev/tcp/${mydns}/53 >>$infofile 2>&1 
    if [ $? -ne 0 ]; then
      iNotReachable=$iNotReachable+1
      test $iCounter -eq 1 && ph.setStatus critical
      echo "ERROR: ${mydns} is not reachable on tcp 53" >>$infofile
    else
        for i in `seq 5`
        do
            (time nslookup ${myHost} $mydns) >$tmpfile 2>&1

            iTime=`cat $tmpfile | grep "^real.*m.*\..*s" | cut -f 2 -d "m" | sed "s#[\.s]##g" | sed "s#^0*##g" `
            echo "$mydns #$i >>> $iTime ms" >>$infofile
            test $iTime -ge $iWarnLimit && cat $tmpfile | grep -vE "^(real|user|sys)" >> $infofile
            test $iTime -gt $iSrvMax && iSrvMax=$iTime

        done
        echo "max: $iSrvMax ms" >>$infofile

        # --- set status
        test $iCounter -eq 1 && ph.setStatusByLimit $iSrvMax $iWarnLimit $iCriticalLimit

        label=`echo $mydns | sed "s#\.#-#g" `
        ph.perfadd "response-$label" "${iSrvMax}"
        test $iSrvMax -gt $iMax && iMax=$iSrvMax
    fi
    test $iCounter -eq 1 && (echo " ^"; echo " |"; echo " +--- 1st nameserver is relevant for total status of the check. Limits are warning=$iWarnLimit and critical=$iCriticalLimit") >>$infofile
    echo "" >>$infofile
done


ph.status "DNS check for $myHost - found maximum was $iMax ms - $iNotReachable of $iCounter nameservers not reachable"
cat $infofile

rm -f $tmpfile $infofile


ph.exit

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