#!/bin/bash
# Title: cpu-load-calc.sh
# Description: Calculate a xymon client's cpu load (Run on Xymon Server periodically with cron)
# Dependency: Requires 'bc' package
# Last Change: 2018-05-22
# Recent Changes:-Updated awk search to look for [nproc] at the beginning of the line

#=======================
# Customize Here
#=======================

# Warning and Critical Load Multipliers (num of procs * multiplier)
load_warn_multiplier=1.0
load_crit_multiplier=1.5

# Directory to save auto load thresholds
auto_load_dir="/etc/xymon/analysis.d/auto-cpuload.d"

# Xymon server's hostdata directory
xymon_hostdata_dir="/var/lib/xymon/hostdata"

# Xymon server's main analysis config file
xymon_analysis_cfg="/etc/xymon/analysis.cfg"
#=======================
# End of Customize
#=======================

#=======================
# Pre-Run Error Checking
#=======================
## Dependency Check ##
which bc &> /dev/null
if [[ $? -eq 1 ]]; then
  echo ">> Error! Dependent package 'bc' (byte code) not detected. Exiting..."
  exit 1
fi

## Does the Auto Load Directory exist?
if [[ ! -d ${auto_load_dir} ]]; then
  echo ">> Error! The directory (${auto_load_dir}) does not exist or is not a directory. Exiting..."
  exit
fi

## Write Access Check
touch ${auto_load_dir}/testfile &> /dev/null
if [[ $? -eq 1 ]]; then
  echo ">> Error! User '$(whoami)' does not have write access to ${auto_load_dir}! Exiting..."
  exit 1
else
  rm -f ${auto_load_dir}/testfile &> /dev/null
fi

## Check if the auto_load_dir is included in main analysis config file
grep "^directory ${auto_load_dir}" ${xymon_analysis_cfg} &> /dev/null
if [[ $? -eq 1 ]]; then
  echo -e ">> Warning! Auto load directory (${auto_load_dir}) is not included in ${xymon_analysis_cfg}. Continuing, but auto CPU load settings will not take affect until 'directory ${auto_load_dir}' is added to ${xymon_analysis_cfg}.\n"
fi
#=======================
# End of Pre-Run Error Checking
#=======================

#===============================
# Functions; Main starts after
#===============================

function show_usage
{
  echo -e "\n####==== Xymon Client Auto Load Thresholds ====####"
  echo -e "\nDescripton: Calculate a xymon client's cpu load."
  echo -e "\n--Usage"
  echo -e "$0      => No arguments, configure with no verbosity."
  echo -e "$0 -v   => Verbose output."
  echo -e "$0 -r   => Refresh CPU load data (force hostdata update)."
  echo -e "$0 -h   => Display usage."
}

# Force snapshots of hostdata
function force_hostdata
{
  # Use node name passed as argument
  node_name=${1}

  # Lie to Xymon that the node's cpu is green, then yellow, forcing a hostdata snapshot
  xymon 127.0.0.1 "status ${node_name}.cpu green $(date)"
  xymon 127.0.0.1 "status ${node_name}.cpu yellow $(date)"
}

#=======================
# Get Script Arguments
#=======================
# Reset POSIX variable in case it has been used previously in this shell
OPTIND=1

# By default, no verbose output
verbose_output="no"
refresh_cpus="no"

while getopts "hrv" opt; do
  case "${opt}" in
    h) # -h (help) argument
      show_usage
      exit 0
    ;;
    r) # -r (refersh cpus) argument
      refresh_cpus="yes"
    ;;
    v) # -v (verbose) argument
      verbose_output="yes"
    ;;
    *) # invalid argument
      show_usage
      exit 0
    ;;
  esac
done

#=======================
# Main Program
#=======================
echo -e "== Xymon Client Auto Load Thresholds =="
echo -e "Load Warning Multiplier: ${load_warn_multiplier}"
echo -e "Load Critical Multiplier: ${load_crit_multiplier}"
echo -e "Saving configs to: ${auto_load_dir}"

# For each node reporting host data
for node in $(ls ${xymon_hostdata_dir}); do

  if [[ ${verbose_output} == "yes" ]]; then
    echo -e "\n>> Working on node: ${node}"
  fi

  if [[ ${refresh_cpus} == "yes" ]]; then
    if [[ ${verbose_output} == "yes" ]]; then
      echo -e "\n-> Refreshing hostdata..."
    fi
    # Force an update of hostdata
    force_hostdata ${node}
  fi

  # Get the number of procs reported from node's most recent host data file
  node_num_procs="$(cat ${xymon_hostdata_dir}/${node}/$(ls -tr ${xymon_hostdata_dir}/${node}/ | tail -1) | awk '/^\[nproc]/ { getline; print }')"
  
  # If node_num_procs is empty or not a number, move to the next node
  if [[ -z ${node_num_procs} || ! ${node_num_procs} =~ [0-9][0-9]* ]]; then
    # Did not find 'nproc' in the host data file or no number from nproc returned

    if [[ ${verbose_output} == "yes" ]]; then
      echo "-> Warning! Could not find 'nproc' in ${node}'s host data file or no number returned. Skipping..."
    fi

    continue
  fi

  # Calculate the warning and critical load thresholds (normalize as a floating point with bc)
  load_warning=$(echo "${node_num_procs} * ${load_warn_multiplier}" | bc)
  load_critical=$(echo "${node_num_procs} * ${load_crit_multiplier}" | bc)

  if [[ ${verbose_output} == "yes" ]]; then
    echo -e "-> Number of Procs: ${node_num_procs}"
    echo -e "-> Warning at: ${load_warning}"
    echo -e "-> Critical at: ${load_critical}"
    echo -e "-> Creating node analysis drop in file..."
  fi

  # Create analysis drop in file
  echo "# ${node}'s CPU Load Thresholds (Warning Critical)" > ${auto_load_dir}/${node}.cfg
  echo "HOST=${node}" >> ${auto_load_dir}/${node}.cfg
  echo "  LOAD  ${load_warning}  ${load_critical}" >> ${auto_load_dir}/${node}.cfg
done

echo -e "\n== Auto Load Thresholds Complete =="

exit 0