#!/bin/bash
#----------------------------------------------------
# Version:      0.1.7.0
# Author:       Florian "Bluewind" Pritz <flo@xssn.at>
#
# Licensed under WTFPL v2
#   (see COPYING for full license text)
#
#----------------------------------------------------
# Script to check if services are working and take 
# actions if not
#----------------------------------------------------
# NOTE:
# Format for service file 
# <IP or hostname> <service name>

#-------------------------- CONFIGURATION -----------------------------#
# Time to wait between 2 checks; Shouldn't be lower than 10 as that
# could cause overlap with the timeouts
WAIT=30

# Take action after x faild checks
MAX_FAIL=2

# set to /dev/null if you don't want logs
LOGFILE="$HOME/watchdog.log"

# Paths (Shouldn't need to be changed)
CURL="/usr/bin/curl"
WGET="/usr/bin/wget"
PING="/bin/ping"
#----------------------------------------------------------------------#

SCRIPTNAME=$(basename $0)

EXIT_SUCCESS=0
EXIT_FAILURE=1
EXIT_ERROR=2
EXIT_BUG=10

DEBUG=0

# Colors for output
red='\e[0;31m'
RED='\e[1;31m'
green='\e[0;32m'
GREEN='\e[1;32m'
blue='\e[0;34m'
BLUE='\e[1;34m'
cyan='\e[0;36m'
CYAN='\e[1;36m'
NC='\e[0m'

function usage {
  echo -e "${blue}Usage:${NC} ${SCRIPTNAME} [OPTIONS]" >&2
  echo -e "Possible service types: http, ping (default)"
  echo -e "Options:" >&2
  echo -e "-h           this help" >&2
  echo -e "-d           debug output" >&2
  echo -e "-f <file>    loads IP list from a file" >&2
  echo -e "-t <time>    Time between 2 checks"
  echo -e "-m <tries>   Max. fails before taking actions"
  [[ $# -eq 1 ]] && exit $1 || exit $EXIT_FAILURE
}

dbg() {
  if [[ $DEBUG -gt 0 ]]; then
    echo "$@" >&2
  fi
}

if [ ! -f "$CURL" ]; then 
  if [ ! -f "$WGET" ]; then
    echo -e "${red}Wget doesn't exist!$NC"
    exit $EXIT_ERROR
  fi
  HTTP="$WGET -t 2 -T 1 -O /dev/null"
else
  HTTP="$CURL --connect-timeout 1 --retry 2 -O /dev/null"
fi

if [ ! -f "$PING" ]; then
  echo -e "${red}Ping command doesn't exist! Please fix the path.$NC"
  exit $EXIT_ERROR
else
  PING="$PING -c 1 -W 1"
fi

if [ ! "$1" ] && [ ! "$2" ]; then
  usage $EXIT_SUCCESS
fi

while getopts ':f:hdt:m:' OPTION ; do
  case $OPTION in
    f)
      IPFILE="$OPTARG"
    ;;
    t)
      WAIT="$OPTARG"
    ;;
    m)
      MAX_FAIL="$OPTARG"
    ;;
    h)
      usage $EXIT_SUCCESS
	  ;;
    d)
      DEBUG=1
	  ;;
    \?)
      echo "Unknown option \"-$OPTARG\"." >&2
      usage $EXIT_ERROR
    ;;
    :)
      echo "Option \"-$OPTARG\" needs an argument" >&2
      usage $EXIT_ERROR
    ;;
	  *)
      echo "This shouldn't happen, please file a bugreport.">&2
      usage $EXIT_BUG
    ;;
  esac
done


shift $(( OPTIND - 1 ))

watcher () {
  IP="$1"
  TYPE="$2"
  WAIT="$3"

  dbg "watcher: $IP - $TYPE - $WAIT"
  
  case "$TYPE" in
    http) COMMAND="$HTTP";;
    *) COMMAND="$PING";;
  esac

  counter=0

  while :;
  do
    precmd_time=$(date +%s)
    $COMMAND $IP &> /dev/null
    exitcode=$?
    if [ "$exitcode" -ne "0" ]; then
      if [ "$counter" -lt "1" ]; then
        downtime=`date`
      fi
      let counter=$counter+1
      if [ "$counter" -eq $MAX_FAIL ]; then
        mail $USER -s "$IP - $TYPE DOWN! " <<< "$IP - $TYPE is down since $downtime
$(tracepath $IP)"
        dbg "$IP - $TYPE down since $downtime"
      fi
    else
      if [ "$counter" -gt $MAX_FAIL ] || [ "$counter" -eq $MAX_FAIL ]; then
        mail $USER -s "$IP - $TYPE UP! " <<< "$IP - $TYPE is OK again.
	        Downtime: ${downtime} - $(date)"
        dbg "$IP - $TYPE up again $(date)"
        echo "$IP - $TYPE Downtime: ${downtime} - $(date)" >> $LOGFILE
      fi
      counter=0
    fi
    current_time=$(date +%s)
    processing_time=`expr ${current_time} - ${precmd_time}`
    sleeptime=`expr ${WAIT} - ${processing_time}`
    if [ "${sleeptime}" -lt "1" ]; then
      sleeptime=0
    fi
    sleep ${sleeptime} 
  done
}

# stops watchers and exits
cleanup() {
  dbg "cleaning up..."
  kill $jobs &> /dev/null
  exit
}
trap cleanup 2 15

# start watchers
while read line
do
  watcher $line & 
done < $IPFILE

# Prepare killing all watchers
jobs=$(jobs -p)
disown -ar

echo "Press ^C to kill"
while :; do
  sleep 999
done