check_graphite

#!/usr/bin/env ruby

$VERBOSE = nil

require "rubygems"
require "optparse"
require "rest-client"
require "json"

EXIT_OK = 0
EXIT_WARNING = 1
EXIT_CRITICAL = 2
EXIT_UNKNOWN = 3

@@options = {}

optparse = OptionParser.new do |opts|
  opts.banner = "Usage: #{File.basename($0)} [options]"

  @@options[:url] = nil
  opts.on("-u", "--url URL", "Target url") do |url|
    @@options[:url] = url
  end
  @@options[:metric] = nil
  opts.on("-m", "--metric NAME", "Metric path string") do |metric|
    @@options[:metric] = metric
  end
  @@options[:shortname] = nil
  opts.on("-s", "--shortname SHORTNAME", "Metric short name (used for performance data)") do |shortname|
    @@options[:shortname] = shortname
  end
  @@options[:duration] = 5
  opts.on("-d", "--duration LENGTH", "Length, in minute of data to parse (default: 5)") do |duration|
    @@options[:duration] = duration
  end
  @@options[:function] = "average"
  opts.on("-f", "--function \[average \| sum\]", "Function applied to metrics for thresholds (default: average)") do |function|
    @@options[:function] = function
  end
  @@options[:warning] = nil
  opts.on("-w", "--warning VALUE", "Warning threshold") do |warning|
    @@options[:warning] = warning
  end
  @@options[:critical] = nil
  opts.on("-c", "--critical VALUE", "Critical threshold") do |critical|
    @@options[:critical] = critical
  end
  @@options[:units] = nil
  opts.on("-U", "--units VALUE", "Adds a text tag to the metric count in the plugin output. Useful to identify the metric units. Doesn't affect data queries.") do |units|
    @@options[:units] = units
  end
  @@options[:message] = "metric count:"
  opts.on("-M", "--message VALUE", "Text message to output (default: \"metric count:\")") do |message|
    @@options[:message] = message
  end
  @@options[:zero_on_error] = false
  opts.on("-z", "--zero-on-error", "Return 0 on a graphite 500 error") do |zero_on_error|
    @@options[:zero_on_error] = zero_on_error
  end
  @@options[:link_graph] = false
  opts.on("-l", "--link-graph", "Add a link in the plugin output, showing a 24h graph for this metric in graphite. Requires 'escape_html_tags=0' option in cgi.cfg nagios config file") do |link_graph|
    @@options[:link_graph] = link_graph
  end
  opts.on( "-h", "--help", "Display this screen" ) do
    puts opts
    exit
  end
end

optparse.parse!

if (@@options[:url].nil? || @@options[:metric].nil? || @@options[:warning].nil? || @@options[:critical].nil?)
  puts optparse
  exit 2
end

def url
  base_url = @@options[:url]
  metric = @@options[:metric]
  duration = @@options[:duration].to_s
  base_url + "/render/?target=" + metric + "&format=json&from=-" + duration + "mins"
end

def output_url
  base_url = @@options[:url]
  metric = @@options[:metric]
  if @@options[:link_graph]  
    "<a target=\"_blank\" href=\"http://" + base_url + "/render/?target=" + URI.escape(metric) + "&from=-24h&width=700&height=450\"> 24h graph </a>"
  end
end  


data = {}
data["total"] = 0

begin
  response = JSON.parse(RestClient.get(URI.encode(url)))
  if response.empty?
    puts "UNKNOWN no data reported by Graphite"
    exit EXIT_UNKNOWN
  else
    response.each do |cache|
      data["#{cache['target']}"] = 0
      count = 0
      cache["datapoints"].each do |point|
        unless (point[0].nil?)
          data["#{cache['target']}"] += point[0]
          count += 1
        end
      end
      if (count == 0)
        count = 1
      end
      if (@@options[:function] == "average")
        data["#{cache['target']}"] /= count
      end
      data["total"] += data["#{cache['target']}"]
    end
  end
rescue RestClient::InternalServerError
  if @@options[:zero_on_error]
    data["total"] = 0
  else
    puts "CRITICAL 500 error from Graphite"
    exit EXIT_CRITICAL
  end
end

total = data["total"].to_f
perfdata = ""
perfdata = "| #{@@options[:shortname]}=#{total}" if !@@options[:shortname].nil?

if (@@options[:critical].to_f > @@options[:warning].to_f)
  if (total >= @@options[:critical].to_f)
    puts "CRITICAL #{@@options[:message]} #{total}#{@@options[:units]} threshold: #{@@options[:critical]} #{output_url} #{perfdata}"
    exit EXIT_CRITICAL
  elsif (total >= @@options[:warning].to_f)
    puts "WARNING #{@@options[:message]} #{total}#{@@options[:units]} threshold: #{@@options[:warning]} #{output_url} #{perfdata}"
    exit EXIT_WARNING
  else
    puts "OK #{@@options[:message]} #{total}#{@@options[:units]} #{output_url} #{perfdata}"
    exit EXIT_OK
  end
else
  if (total <= @@options[:critical].to_f)
    puts "CRITICAL #{@@options[:message]} #{total}#{@@options[:units]} threshold: #{@@options[:critical]} #{output_url} #{perfdata}"
    exit EXIT_CRITICAL
  elsif (total <= @@options[:warning].to_f)
    puts "WARNING #{@@options[:message]} #{total}#{@@options[:units]} threshold: #{@@options[:warning]} #{output_url} #{perfdata}"
    exit EXIT_WARNING
  else
    puts "OK #{@@options[:message]} #{total}#{@@options[:units]} #{output_url} #{perfdata}"
    exit EXIT_OK
  end
end