#!/bin/bash

# 20210628 - Renamed xymon-chronyd.bash
# based on ntpd xymon monitor

# IWT 2010 stanza below for etc/clientlaunch.cfg
# [ntpd]
#        ENVFILE $HOBBITCLIENTHOME/etc/hobbitclient.cfg
#        CMD $HOBBITCLIENTHOME/ext/xymon-chronyd.cfg
#        LOGFILE $HOBBITCLIENTHOME/logs/xymon-chronyd.log
#        INTERVAL 5m


# General setup 

PROG="xymon-chronyd.bash"
VERSION="0.3"
TEST="ntpd"		# THE NAME OF OUR TEST
script=$(basename $0)
FOOTER="
           Test: ${TEST}${NL}
         Script: ${script}    
        Version: ${VERSION}${NL}
"

PURPLEDELAY="60"		# Time to live = one day
ERRORLOG=$HOBBITCLIENTHOME/logs/$PROG.log
CHRONYC='/usr/bin/chronyc'

# Colors and limits - units are in milliseconds (1/1000s of a second) (other than stratum which goes from 0-16)
((STRATUM_YELLOW = 2))      # if stratum is larger than 2 -- go yellow
((STRATUM_RED = 3))         # if stratum is larger than 3 -- go red
((SLOW_YELLOW = 1))         # if more than 1 ms, go yellow (system time)
((SLOW_RED = 10))           # if more than 10 ms, go red (system time)
((RMS_OFFSET_YELLOW = 1))   # if more than 1 ms, go yellow (RMS Offset)
((RMS_OFFSET_RED = 10))     # if more than 10 ms, go red (RMS Offset)
((DISPERSION_YELLOW = 1))   # if more than 1 ms, go yellow (Root Dispersion)
((DISPERSION_RED = 10))     # if more than 10 ms, go red (Root Dispersion)

# output from: chronyc tracking
# Reference ID    : 35.73.197.144 (time1.google.com)
# Stratum         : 2
# Ref time (UTC)  : Mon Jun 28 16:12:59 2021
# System time     : 0.000489485 seconds slow of NTP time
# Last offset     : -0.000389254 seconds
# RMS offset      : 0.001954304 seconds
# Frequency       : 2.726 ppm slow
# Residual freq   : -0.015 ppm
# Skew            : 0.345 ppm
# Root delay      : 0.016909 seconds
# Root dispersion : 0.001069 seconds
# Update interval : 521.7 seconds
# Leap status     : Normal
#################################################################
green='GREEN'
yellow='YELLOW'
red='RED'
COLOR="${green}"
ERR=''
NL='
'
DIV='<HR>'

if [[ "X$HOBBITCLIENTHOME" == "X" ]]
 then
   echo "ERROR: HOBBITCLIENTHOME variable not set"
   exit 1  
 fi

#################################################################
# actually start getting data and running tests
#################################################################

tracking=$(${CHRONYC} tracking)
sources=$(${CHRONYC} sourcestats)
stratum=$(echo "${tracking}" | grep Stratum | awk '{print $3}')
# sed out minus sign so we can look at absolute distance
slow=$(echo "${tracking}" | grep "System time" | awk '{print $4}' | sed 's/-//' | sed 's/+//')
rmsOffset=$(echo "${tracking}" | grep "RMS offset" | awk '{print $4}' | sed 's/-//' | sed 's/+//')
dispersion=$(echo "${tracking}" | grep "Root dispersion" | awk '{print $4}' | sed 's/-//' | sed 's/+//')
sync=$(echo "${tracking}" | grep 'Leap status' | awk '{print $4}' | tr '[:upper:]' '[:lower:]')

# first test is absolute
if (( stratum > STRATUM_RED ))
 then
   COLOR="${red}"
   ERR="STRATUM: ${startum} &gt; Red Threshold (${STRATUM_RED})${NL}"
 else
   if (( stratum > STRATUM_YELLOW ))
    then
      COLOR="${yellow}"
      ERR="STRATUM: ${startum} &gt; Yellow Threshold (${STRATUM_YELLOW})${NL}"
    fi
 fi

# rest of tests must assume a tes may have already failed
# system time off
tmp=$(echo "${slow} * 1000" | bc)              # multiply by 1000 to get milliseconds
tn=$(echo "0${tmp}" | awk -F'.' '{print $1}')  # add leading zero (in case less than 0) and then truncate at decimal point
if ((tn > SLOW_RED))
 then
   COLOR="${red}"
   ERR="${ERR}System time is off: ${slow}*1000 &gt; Red Threshold (${SLOW_RED})${NL}"
 else
   if ((tn > SLOW_YELLOW))
    then
      if [[ "${COLOR}" == "${green}" ]]
       then
         COLOR="${yellow}"
       fi
      ERR="${ERR}System time is off: ${slow} &gt; Yellow Threshold (${SLOW_YELLOW})${NL}"
    fi
 fi
#RMS Offset
tmp=$(echo "${rmsOffset} * 1000" | bc)         # multiply by 1000 to get milliseconds
tn=$(echo "0${tmp}" | awk -F'.' '{print $1}')  # add leading zero (in case less than 0) and then truncate at decimal point
if ((tn > RMS_OFFSET_RED))
 then
   COLOR="${red}"
   ERR="${ERR}RMS Offset is bad: ${rmsOffset} &gt; Red Threshold (${RMS_OFFSET_RED})${NL}"
 else
   if ((slow > RMS_OFFSET_YELLOW))
    then
      if [[ "${COLOR}" == "${green}" ]]
       then
         COLOR="${yellow}"
       fi
      ERR="${ERR}RMS Offset is bad: ${rmsOffset}*1000 &gt; Yellow Threshold (${RMS_OFFSET_YELLOW})${NL}"
    fi
 fi
# dispersion
tmp=$(echo "${dispersion} * 1000" | bc)         # multiply by 1000 to get milliseconds
tn=$(echo "0${tmp}" | awk -F'.' '{print $1}')  # add leading zero (in case less than 0) and then truncate at decimal point
if ((slow > SLOW_RED))
 then
   COLOR="${red}"
   ERR="${ERR}Dispersion is bad: ${dispersion}*1000 &gt; Red Threshold (${DISPERSION_RED})${NL}"
 else
   if ((slow > SLOW_RED))
    then
      if [[ "${COLOR}" == "${green}" ]]
       then
         COLOR="${yellow}"
       fi
      ERR="${ERR}Dispersion is bad: ${dispersion}*1000 &gt; Yellow Threshold (${DISPERSION_YELLOW})${NL}"
    fi
 fi
# leap second can be: Normal, Insert second, Delete second or Not synchronized
# grab first word after colon, make loercase and compare to not
if [[ 'not' == "${sync}" ]]
 then
   COLOR="${red}"
   ERR="${ERR}Leap status is Not synchronized${NL}"
 fi

if [[ 'X' != "${ERR}X" ]]                  # we had errors
 then
   $ERR="${ERR}${NL}${NL}${DIV}${NL}"
 fi
MSG="<B>Chronyc tracking</b>:${NL}${tracking}${NL}${NL}${DIV}${NL}<b>chronyc sourcestats</b>: ${sources}${nl}${NL}${DIV}${NL}"

# FORMAT IT PROPERLY FOR BB...
time=`$DATE +"%d/%m/%Y %H:%M:%S"`
LINE="status+${PURPLEDELAY} ${MACHINE}.${TEST} ${COLOR} ${time}
${ERR}
${MSG}
${FOOTER}
"
# Send a status update to the Big Brother display
# echo "$LINE"     # SEND IT TO BBDISPLAY
$BB $BBDISP "$LINE"     # SEND IT TO BBDISPLAY
# END