check_puppet_agent

#!/bin/bash
#
# A nrpe plugin that will poll the state of puppet on a machine. We want
# to have different notification intervals for different states, so this
# script operates in two different modes.
#
# -c: Check for catalog compile failures only
# -t: Check the last time puppet ran and check for run time errors
#
# Based on code by Alexander Swen <a@swen.nu>
#
# CHANGELOG
# 5/15/2013 bhourigan Initial commit
#

# NOTE: Puppet 3.x changed configprint syntax
puppetversion=$(puppet -V)
case $puppetversion in
   2.*)
      statedir=$(puppet --configprint statedir)
      ;;
   3.*)
      statedir=$(puppet config print statedir)
      ;;
   *)
      echo "Sorry, puppet version ${puppetversion:-UNKNOWN} is unsupported by $0"
      exit 1
      ;;
esac

statefile=${statedir}/last_run_summary.yaml
reportfile=${statedir}/last_run_report.yaml
disablefile=${statedir}/disabled.at

usage(){
   if [ $# -gt 0 ]; then
      echo "ERROR: $*"
      echo
   fi

   echo -e "Usage: $0 [-c] [-t <threshold>] [-s <statefile>] [-r <reportfile>] [-d <disablefile>]"
   echo
   echo -e "\t-w last run alert threshold"
   echo -e "\t-c check for catalog compilation failures only"
   echo -e "\t-d disable file location (default: ${statedir}/disabled.at)"
   echo -e "\t-r report file location (default: ${statedir}/last_run_report.yaml)"
   echo -e "\t-s state file location (default: ${statedir}/last_run_summary.yaml)"
   echo
   exit 1
}

result(){
   echo "$1"
   exit $2
}

check_last_run_time(){
   if [ ! -f $disablefile ]; then
      # Parse last_last run from statefile
      last_run_time=$(awk '/\s*last_run:/ {print $2}' $statefile)
      if [ ${last_run_time:-0} -eq 0 ]; then
         result "Can't get last_run from $statefile" 3
      fi

      now=$(date "+%s")
      time_since_last=$((now-last_run_time))
      if [ $time_since_last -gt $threshold ]; then
         result "Last run was ${time_since_last} seconds ago" 3
      fi
   else
      comment=$(grep 'Puppet has been disabled' /etc/motd | sed -r "s/\x1B\[([0-9]{1,3}((;[0-9]{1,3})*)?)?[m|K]//g")
      result "${comment}" 0
   fi
}

check_last_run_errors(){
   failed=$(awk '/\s*failed:/ {print $2}' $statefile)
   failure=$(awk '/\s*failed:/ {print $2}' $statefile)
   fail_to_restart=$(awk '/\s*failed:/ {print $2}' $statefile)
   sum_of_fail=$((failed+failure+failure_to_restart))

   if [ ${sum_of_fail:-0} -gt 0 ]; then
      if [ ${sum_of_fail:-0} -gt 1 ]; then
         s="s"
      fi
      result "Last run had ${sum_of_fail} error${s:-}" 1
   fi
}

check_catalog_compile(){
   failed_catalog=$(awk -F '"' '/\s*Could not retrieve catalog from remote server:|\s*Failed to apply catalog/ {gsub(/on node .*/,"",$2); print $2}' $reportfile)

   if [ ! -z "${failed_catalog:-}" ]; then
      result "${failed_catalog}" 2
   fi
}

result_ok(){
   configuration_version=$(awk -F '"' '/\s*configuration_version:/ {print $2}' $reportfile)
   result "Puppet agent ${puppetversion} running catalog ${configuration_version:-UNKNOWN}" 0
}

while getopts ":t:s:d:c" opt; do
   case $opt in
      t)
         threshold=$OPTARG
         ;;
      t)
         statefile=$OPTARG
         ;;
      d)
         disablefile=$OPTARG
         ;;
      r)
         reportfile=$OPTARG
         ;;
      c)
         check_catalog=1
         ;;
      \?)
         usage "Invalid option: -$OPTARG"
         ;;
      :)
         usage "Option -$OPTARG requires an argument."
         ;;
   esac
done

if [ ! -f $statefile ]; then
   result "State file $statefile doesn't exist" 3
fi

if [ ! -f $reportfile ]; then
   result "Report file $reportfile doesn't exist" 3
fi

# Nagios won't allow us to have different alerting timers for different
# states (we want to see catalog compile failures and recoveries
# immediately, and run time errors daily), so this script will operate
# in two modes which will be defined with two different notification
# intervals in Nagios

if [ ${check_catalog:-0} -eq 1 ]; then
   if [ ${threshold:-0} -gt 1 ]; then
      usage "Threshold argument not allowed with -c"
   fi

   check_catalog_compile
else
   if [ ${threshold:-0} -lt 1 ]; then
      usage "Invalid time threshold $OPTARG"
   fi

   check_last_run_time
   check_last_run_errors
fi

result_ok