check_es.rb

#!/usr/bin/env ruby

require 'rubygems'
require 'json'
require 'open-uri'

require 'yaml'

critical = false
warn= false
critical_msg = []
warn_msg = []
additional_info =[]
ok_msg = []

config_file = '/ftn/apps/elasticsearch/master/current/config/elasticsearch.yml'
# Read node list from es configuration
if File.exists?(config_file)
  conf = YAML.load_file(config_file)
  nodes = conf['discovery']['zen']['ping']['unicast.hosts']
  expected_count = nodes.length
else
  nodes = ['a02.prod.fotonauts.net']
  expected_count = nil
end

clusters = {}

nodes.each do |hostname|
  begin
    nodes = JSON.parse(open("http://#{hostname}:9200/_cluster/state").read)['nodes']
  rescue Exception => e
    critical = true
    critical_msg << "Unable to get cluster state on #{hostname}"
    additional_info << e.inspect
    next
  end
  if expected_count && nodes.length != expected_count
    critical = true
    critical_msg << "Number of node visible on #{hostname} is #{nodes.length} (should be #{expected_count})"
    additional_info << "Visible node for #{hostname}:"+nodes.join(", ")
  else
    nodes = nodes.values.map { |n| n['name'] }.sort.join(", ")
    clusters[nodes] ||= []
    clusters[nodes] << hostname
  end
end


if clusters.keys.length > 1
  critical = true
  critical_msg << "More than one active cluster"
  additional_info << clusters.inspect
end

if critical
  puts "CRITICAL "+critical_msg.join(", ")
  puts additional_info.join("\n")
  Process.exit 2
end
if warn
  puts "WARNING "+warn_msg.join(", ")
  puts additional_info.join("\n")
  Process.exit 1
end

ok_msg << "OK 1 cluster with nodes #{clusters.keys[0]}."

puts ok_msg.join(", ")