NETCONF-speaking script to poll a Juniper SRX NAT pool and stuff the metrics into Graphite

netconf_nat_pool.rb

#!/usr/bin/env ruby

require 'rubygems'
require 'pry'
require 'socket'
$:.unshift '/home/jof/src/net-netconf/lib'
$:.unshift '/home/jof/src/net-ssh/lib'

require 'net/netconf'
require 'net/ssh'
require 'net/ssh/proxy/command'

jump_host = ARGV.shift
target = ARGV.shift
snat_pool = ARGV.shift
unless jump_host and target and snat_pool then
  STDERR.puts "This needs a jump host, target, and snat pool name as arguments"
  exit 1
end

poll_interval = 5

corp_jump_host = Net::SSH::Proxy::Command.new("ssh #{jump_host} nc %h %p")
netconf = Netconf::SSH.new(:target => target, :username => ENV['USER'], :port => 22)
netconf.trans_open(:proxy => corp_jump_host)
netconf.open

graphite_socket = TCPSocket.new('localhost', 2003)

at_exit do
  netconf.close
  graphite_socket.close
end

loop do
  sleep poll_interval
  cluster_status = netconf.rpc.get_chassis_cluster_status(:redundancy_group => '0')
  nodes = cluster_status.xpath('//device-stats/device-name').map { |node| node.text }
  statuses = cluster_status.xpath('//device-stats/redundancy-group-status').map { |status| status.text }
  states = nodes.zip(statuses)
  active_node = states.select { |state| state[1] == 'primary' }.flatten.first

  pools = netconf.rpc.retrieve_source_nat_pool_information(:pool_name => snat_pool)
  # FIXME: Detect <error> tags in RPC response when the remote node times out.
  
  # FIXME: Maybe there is a better way to walk the <multi-routing-engine-item> nodes. xpath'ing a Nokogiri::XML::NodeSet seems to still require the full path from the root.
  total_addresses = pools.xpath("//multi-routing-engine-item[re-name=\"#{active_node}\"]//total-pool-address").text # "16"
  total_addresses = total_addresses.to_i
  translation_range = pools.xpath("//multi-routing-engine-item[re-name=\"#{active_node}\"]//source-pool-port-translation").text # "[1024, 63487]"
  translation_range = translation_range.gsub(/[\[\]]/,'').split(/,\s*/).map { |i| i.to_i }
  translation_range = Range.new(*translation_range)
  single_port_translations = pools.xpath("//multi-routing-engine-item[re-name=\"#{active_node}\"]//single-port").text
  single_port_translations = single_port_translations.to_i
  twin_port_translations = pools.xpath("//multi-routing-engine-item[re-name=\"#{active_node}\"]//twin-port").text
  twin_port_translations = twin_port_translations.to_i

  available_source_port_translations = translation_range.count * total_addresses
  in_use_port_translations = single_port_translations + twin_port_translations

  now = Time.now

  metrics = { 'available'        => available_source_port_translations,
              'single_port_used' => single_port_translations,
              'twin_port_used'   => twin_port_translations,
              'ports_used'       => (single_port_translations + twin_port_translations) }
  metrics.each do |metric, value|
    # Change this metric name for your own Graphite environment
    graphite_socket.puts "jof.nat_pools.#{target}.#{snat_pool}.#{metric} #{value} #{now.to_i}"
  end
  graphite_socket.flush

  metrics.each do |metric, value|
    puts "Reporting that jof.nat_pools.#{target}.#{snat_pool}.#{metric} is #{value} at #{now}"
  end
end

#netconf.close