gistfile1.txt

------------------------------------------------------------------------
this 12-line file is "send.rb"
------------------------------------------------------------------------
#!/usr/bin/env ruby
require 'rubygems'
require 'socket'

sock = TCPSocket.new 'localhost', 9814

t0 = Time.now
sock.send 'start working', 0

reply, sender_addrinfo  = sock.recvfrom(16)

printf("%8.4fs later, got the stale reply: %s\n", (Time.now - t0), reply)


------------------------------------------------------------------------
and this file is 'clairvoyant.rb'
------------------------------------------------------------------------
#!/usr/bin/env ruby
require 'rubygems'
require 'eventmachine'

class Feeder < EventMachine::Connection⋅
  def initialize(q)⋅
    @q = q⋅
  end⋅

  def receive_data(cmd)⋅
    send_data  cmd + '.'*(16 - cmd.size)
    puts 'Reply has been sent, entangled with the work which has yet to begin'
    @q.push  cmd⋅
  end⋅
end⋅

# - - - - - - - - - - - - -⋅

class Worker
  # This busy loop takes >50ms on my mac.
  # If you comment out the work, the reply is received in < 1ms.
  def busy(str)
    puts 'The future work period has yet to begin...'
    10.times{system 'ls'}
    puts 'work completed'
  end
end

# - - - - - - - - - - - - -⋅

class Clairvoyant
  def initialize
    puts 'Clair is listening'
  end

  def run
    worker = Worker.new
    EventMachine::run {⋅
      q = EM::Channel.new⋅
      EM::start_server '127.0.0.1', 9814, Feeder, q⋅
      unused = q.subscribe{|msg|  worker.busy(msg)}⋅
    }⋅
  end
end

# - - - - - - - - - - - - -⋅

clair = Clairvoyant.new
clair.run

The problem is you can't use sleep inside an EM program. You'll be putting the entire reactor to sleep and, this will probably happen before the message is sent.

Instead of sleep, can you use something along the lines of EM.add_timer(2) { puts 'wakeup' }?

Thanks - you're right. But the actual program actually does some work instead of sleeping. I just put 'sleep 2' in the isolated example for simplicity, and it was a very misleading simplification. The EM timers aren't a one-line replacement for sleep, so I've changed to: 10.times{system 'ls'} Which takes more than 50ms on my system. With no "work" being done, the reply is received in less than one millisecond, so the problem still remains.

Somehow, the socket data is not being received in the 'send.rb' program until AFTER the work gets done.

That's still going to block the reactor. It will execute all of the 10.times { system 'ls' } commands before the reactor loop runs again. The reactor loop has to turn in order for the messages to get sent out.

You either need to turn the blocking operations into non-blocking or use something like EM.defer to run them on a background thread.

On Wed, May 18, 2011 at 2:23 PM, dj2 < ***@***.***>wrote: That's still going to block the reactor. It will execute all of the 10.times { system 'ls' } commands before the reactor loop runs again. The reactor loop has to turn in order for the messages to get sent out. You either need to turn the blocking operations into non-blocking or use something like EM.defer to run them on a background thread.

Replacing unused = q.subscribe{|msg| worker.busy(msg)} with unused = q.subscribe{|msg| EM.defer lambda {worker.busy(msg)} } Solves the delayed-response problem. But I suspect subsequent socket commands that arrive prior to the completion of the first Worker's job will start additional worker instances, which will cause disastrously interleaved work in my application. I suppose I should change the Channel subscriber to a Queue popper, which does not pop the next entry until the worker completes his job.... if I defer the worker that pops the queue entry so that it does not also block receive_data() and q.push(). Thank you, - Randy