newrelic_redborder_agent

#! /usr/bin/env ruby

require 'rubygems'
require 'bundler/setup'
require 'snmp'
require 'pry'
require_relative 'src/druid-master'
require_relative 'src/nginx-master'
require_relative 'src/chef-master'
require_relative 'src/check-master'
require_relative 'lib/rb-monitor'
require 'newrelic_plugin'
require('logger')
require 'rationalist'

module RedborderAgent
  class Agent < NewRelic::Plugin::Agent::Base

# Agent configuration
    agent_guid 'com.redborder.redborder'
    agent_version '1.0.2'
    agent_config_options :snmp_host, :snmp_community, :services
    agent_human_labels('Manager') { 'testingCluster' }
    $logger = Logger.new('/var/log/newrelic/plugin.log')
    argv = Rationalist.parse(ARGV)
    # ./newrelic_redborder_agent --log debug
    if (argv[:log] == 'debug')
      $logger.level = Logger::DEBUG
    else
      $logger.level = Logger::INFO
    end
    $metrics = []
    $nginx = []
    $chef = []
    $check = []

    $logger.formatter = proc do |severity, datetime, progname, msg|
      '#{severity}: #{datetime}: #{msg}\n'
    end

    $i=0
    $var = 0
    $number = 0

# Thread managers for detailed services
    druid_master
    nginx_master
    chef_master

    Thread.new do
      $logger.debug('Nginx recolector launched')
      recolector($nginx)
    end

    Thread.new do
      $logger.debug('Druid recolector launched')
      recolector($metrics)
    end

    def poll_cycle
      # poll_cycle is the loop where the agent collect&report metrics
      host = `hostname`.strip

# SNMP manager initialization
      SNMP::Manager.open(community: snmp_community, host: snmp_host) do |manager|
        memory_total = mem_total(manager)

        services.each do |service|
          check_master(service)
        end

        memory = (memory_total -
        mem_free(manager) -
        mem_total_buffer(manager) -
        mem_total_cache(manager)) * 100

        #----------------------------------- RB_MONITOR METRICS ---------------
        report_metric 'CPU_PERC_' + host,
        'Value', cpu(manager)
        report_metric 'LATENCY_' + host,
        'Value', latency(host)
        report_metric 'PKTS_RCV_' + host,
        'Value', pkts_rcv(host)
        report_metric 'MEM_TOTAL_' + host,
        'Value', memory_total
        report_metric 'MEM_FREE_' + host,
        'Value', mem_free(manager)
        report_metric 'MEM_TOTAL_BUFFER_' + host,
        'Value', mem_total_buffer(manager)
        report_metric 'MEM_TOTAL_CACHE_' + host,
        'Value', mem_total_cache(manager)
        report_metric 'MEM_PERC_' + host,
        'Value', memory / memory_total

        # (disk utilization)avio shows the average number of milliseconds p/req
        report_metric 'AVIO_' + host,
                      'Value', get_avio()
        report_metric 'DISK_PERCENT_' + host,
                      'Value', disk_percent(manager)
        report_metric 'DISK_LOAD_' + host,
                      'Value', disk_load()
        report_metric 'MEMORY_DRUID_BROKER_' + host,
                      'Value', 100 * memory_total_druid_broker()/memory_total
        report_metric 'MEMORY_DRUID_COORDINATOR_' + host,
                      'Value', 100 * memory_total_druid_coordinator()/memory_total
        report_metric 'MEMORY_DRUID_HISTORICAL_' + host,
                      'Value', 100 * memory_total_druid_historical()/memory_total
        report_metric 'MEMORY_DRUID_REALTIME_' + host,
                      'Value', 100 * memory_total_druid_realtime()/memory_total
        report_metric 'MEMORY_KAFKA_' + host,
                      'Value', 100 * memory_total_kafka()/memory_total
        report_metric 'MEMORY_NPROBE_' + host,
                      'Value', 100 * memory_total_nprobe()/memory_total
        report_metric 'MEMORY_POSTGRESQL_' + host,
                      'Value', 100 * memory_total_postgresql()/memory_total
        report_metric 'MEMORY_RBWEBUI_' + host,
                      'Value', 100 * memory_total_rbwebui()/memory_total
        report_metric 'MEMORY_ZOOKEEPER_' + host,
                      'Value', 100 * memory_total_zookeeper()/memory_total
        report_metric 'jvm_mem_init',
                      'Value', 1
      end
      $logger.debug('General metrics reported')
      #
      # Druid metrics reporting
      #
      $metrics.each do |m|
        unless m['metric'].nil? || m['service'].nil? || m['value'].nil?
          report_metric  'druid_' +
                          m['metric'] + '_' + m['service'] + '_' + host, 'Value', m['value']
          m['ttl'] -= 1
          # puts  'druid_' + m['metric']  + m['service'] + '_' + host + ' Value: ' + m['value'].to_s
        end
      end
      $logger.debug('Druid metrics reported')
      #
      # Nginx metrics reporting
      #
      $nginx.each do |m|
        unless m['status'].nil? || m['times'].nil?
          report_metric  'nginx_' + m['status'] + '_' + host, 'Value', m['times']
          m['ttl'] -= 1
          # puts  'nginx_' + m['status'] + '_' + host + ' Value: ' + m['times'].to_s
        end
      end
      $logger.debug('Nginx metrics reported')
      #
      # Chef metrics reporting
      #
      $chef.each do |m|
        unless m['error'].nil? || m['times'].nil?
          report_metric  'chef_' + m['error'] + '_' + host, 'Value', m['times']
          # puts  'chef_' + m['error'] + '_' + host + ' Value: ' + m['times'].to_s
        end
      end
      $logger.debug('Chef metrics reported')
      #
      # Check metrics reporting
      #
      $check.each do |m|
        unless m['service'].nil? || m['value'].nil?
          report_metric  'check_' + m['service'] + '_' + host, 'Value', m['value']
          # puts 'check_' + m['service'] + '_' + host + ' Value: ' + m['value'].to_s
        end
      end
      $logger.debug('Checks metrics reported')

      $i += 1
      $number+=1
      if $i%60 == 0
        chef_recolector
        $logger.debug('Chef recolector executed')
      end
    end
    $logger.info('Pollcycle number ' + $number.to_s + ' finished ')
  end
  #
  # Register this agent with the component.
  # The RedborderAgent is the name of the module that defines this
  # driver (the module must contain at least three classes - a
  # PollCycle, a Metric and an Agent class, as defined above).
  #
  NewRelic::Plugin::Setup.install_agent :redborder, RedborderAgent
  #
  # Launch the agent
  #
  NewRelic::Plugin::Run.setup_and_run
end