jam-cloud/websocket-gateway/lib/jam_websockets/server.rb

require 'em-websocket'
require 'bugsnag'
module JamWebsockets

  class Server

    def initialize(options={})
      EM::WebSocket.close_timeout = 10 # the default of 60 is pretty intense
      @log = Logging.logger[self]
      @count=0
      @router = Router.new
      @ar_base_logger = ::Logging::Repository.instance[ActiveRecord::Base]

      @last_conn_check = nil
    end

    def run(options={})
      host = "0.0.0.0"
      port = options[:port]
      trust_port = port + 1
      connect_time_stale_client   = options[:connect_time_stale_client].to_i
      connect_time_expire_client  = options[:connect_time_expire_client].to_i
      connect_time_stale_browser  = options[:connect_time_stale_browser].to_i
      connect_time_expire_browser = options[:connect_time_expire_browser].to_i
      max_connections_per_user = options[:max_connections_per_user].to_i
      chat_enabled = options[:chat_enabled].nil? ? true : options[:chat_enabled]
      chat_blast = options[:chat_blast].nil? ? true : options[:chat_blast]
      gateway_name = options[:gateway_name]
      rabbitmq_host = options[:rabbitmq_host]
      rabbitmq_port = options[:rabbitmq_port].to_i
      allow_dynamic_registration = options[:allow_dynamic_registration].nil? ? true : options[:allow_dynamic_registration]


      Stats::init(options)

      calling_thread = options[:calling_thread]
      trust_check = TrustCheck.new(trust_port, options[:cidr])

      @log.info "starting server #{host}:#{port} staleness_time=#{connect_time_stale_client}; reconnect time = #{connect_time_expire_client}, rabbitmq=#{rabbitmq_host}:#{rabbitmq_port} gateway_name=#{gateway_name}"

      EventMachine.error_handler{|e|
        puts "unhandled error #{e}"
        puts "unhandled error #{e.backtrace}"
        @log.error "unhandled error #{e}"
        @log.error "unhandled error #{e.backtrace}"
        # SHOULD WE JUST DIE HERE?  This was seen in production, and the gateway stopped after this!!!
        begin
          Bugsnag.notify(e)
        rescue => bugsnag_e
          puts "unable to report to bugsnag #{bugsnag_e}"
          @log.error "unable to report to bugsnag  #{bugsnag_e}"
        end
      }

      EventMachine.run do
        @router.start(connect_time_stale_client, connect_time_expire_client, connect_time_stale_browser, connect_time_expire_browser, host: rabbitmq_host, port: rabbitmq_port, max_connections_per_user: max_connections_per_user, gateway: gateway_name, allow_dynamic_registration: allow_dynamic_registration, chat_enabled: chat_enabled, chat_blast: chat_blast) do

          @router.init

          #start_connection_expiration
          #start_client_expiration
          #start_connection_flagger
          start_stats_dump
          start_websocket_listener(host, port, trust_port, trust_check, options[:emwebsocket_debug])
          calling_thread.wakeup if calling_thread
        end

        # if you don't do this, the app won't exit unless you kill -9
        at_exit do
            @log.info "cleaning up server"
            @log.info "Exit error: #{$ERROR_INFO}" if $ERROR_INFO
            @log.info "Exit position: #{$ERROR_POSITION}" if $ERROR_POSITION
            @router.cleanup
        end
      end
    end

    def stop
      EventMachine::stop_event_loop
    end


    def check_for_em_drift(timer)
       # if our timer check is a full second off, say what's up
      drift = Time.now - @last_conn_check
      @router.highest_drift = drift if drift > @router.highest_drift

      @last_conn_check = Time.now
    end



    def start_websocket_listener(listen_ip, port, trust_port, trust_check, emwebsocket_debug)
      EventMachine::WebSocket.run(:host => listen_ip, :port => port, :debug => emwebsocket_debug) do |ws|
        #@log.info "new client #{ws}"
        @router.new_client(ws, false)
      end
      EventMachine::WebSocket.run(:host => listen_ip, :port => trust_port, :debug => emwebsocket_debug) do |ws|
        @log.info "new latency_tester client #{ws}"
        # verify this connection came in from a valid subnet, if specified
        ip = extract_ip(ws)
        if trust_check.trusted?(ip, trust_port)
          @router.new_client(ws, true)
        else
          @log.warn("untrusted client attempted to connect to #{listen_ip}:#{trust_port} from #{ip}")
          ws.close
        end
      end
      @log.debug("started websocket")
    end

    def start_connection_expiration
      # one cleanup on startup
      @router.wipe_all_connections
      #@router.periodical_check_connections

      @last_conn_check = Time.now
      timer = 2
      EventMachine::PeriodicTimer.new(timer) do
        check_for_em_drift(timer)
        time_it('conn_expire') { safety_net { sane_logging { @router.periodical_check_connections } } }
      end
    end

    def start_client_expiration
      # one cleanup on startup
      @router.periodical_check_clients

      EventMachine::PeriodicTimer.new(30) do
        time_it('client_expire') { safety_net { sane_logging { @router.periodical_check_clients } } }
      end
    end

    def start_connection_flagger
      # one cleanup on startup
      @router.periodical_flag_connections

      EventMachine::PeriodicTimer.new(2) do
        time_it('conn_flagger') { safety_net { sane_logging { @router.periodical_flag_connections } } }
      end
    end

    def start_stats_dump
      EventMachine::PeriodicTimer.new(60) do
        time_it('stats_dump') { safety_net { @router.periodical_stats_dump } }
      end
    end

    # this was added for this reason: https://jamkazam.atlassian.net/browse/VRFS-2425
    # if an unhandled exception occurs in PeriodicTimer, it just kills all future timers; doesn't kill the app.
    # not really what you want.

    # so, we signal to Bugsnag, so we know really bad stuff is happening, but we also move
    def safety_net(&blk)
      begin
        blk.call
      rescue => e
        #Bugsnag.notify(e)
        @log.error("unhandled exception in EM Timer #{e}")
        puts "Error during processing: #{$!}"
        puts "Backtrace:\n\t#{e.backtrace.join("\n\t")}"

      end
    end

    def time_it(cat, &blk)
      start = Time.now

      blk.call

      time = Time.now - start


      @router.time_it_sums[cat] = (@router.time_it_sums[cat] || 0) + time

      @log.warn("LONG TIME #{cat}: #{time}") if time > 1
    end


    def sane_logging(&blk)
      # used around repeated transactions that cause too much ActiveRecord::Base logging
      # example is handling heartbeats
      begin
        original_level = @ar_base_logger.level if @ar_base_logger
        @ar_base_logger.level = :info if @ar_base_logger
        blk.call
      ensure
        @ar_base_logger.level = original_level if @ar_base_logger
      end
    end

    private
    def extract_ip(client)
      Socket.unpack_sockaddr_in(client.get_peername)[1]
    end
  end

end