Question

I’m about to put this design into use in an application, but I’m fairly new to threading and Queue stuff in python. Obviously the actual application is not for saying hello, but the design is the same – i.e. there is a process which takes some time to set-up and tear down, but I can do multiple tasks in one hit. Tasks will arrive at random times, and often in bursts.

Is this a sensible and thread safe design?

class HelloThing(object):

  def __init__(self):
    self.queue = self._create_worker()

  def _create_worker(self):
    import threading, Queue

    def worker():
      while True:
        things = [q.get()]
        while True:
          try:
            things.append(q.get_nowait())
          except Queue.Empty:
            break
        self._say_hello(things)
        [q.task_done() for task in xrange(len(things))]

    q = Queue.Queue()
    n_worker_threads = 1
    for i in xrange(n_worker_threads):
      t = threading.Thread(target=worker)
      t.daemon = True
      t.start()

    return q

  def _say_hello(self, greeting_list):
    import time, sys
    # setup stuff
    time.sleep(1)
    # do some things
    sys.stdout.write('hello {0}!\n'.format(', '.join(greeting_list)))
    # tear down stuff
    time.sleep(1)


if __name__ == '__main__':
  print 'enter __main__'

  import time
  hello = HelloThing()

  hello.queue.put('world')
  hello.queue.put('cruel world')
  hello.queue.put('stack overflow')

  time.sleep(2)

  hello.queue.put('a')
  hello.queue.put('b')

  time.sleep(2)

  for i in xrange(20):
    hello.queue.put(str(i))

  #hello.queue.join()

  print 'finish __main__'

Answer 1

1. The thread safety is handled by Queue implementation (also you must handle in your _say_hello implementation if it is required).

2. Burst handler problem: A burst should be handled by a single thread only.(ex: let’s say your process setup/teardown takes 10 seconds; at second 1 all threads will be busy with burst from sec 0, on second 5 a new task(or burst) but no thread available to handle them/it). So a burst should be defined by max number of tasks (or maybe “infinite”) for a specific time-window. An entry in queue should be a list of tasks.

How can you group burst tasks list?
I provide a solution as code, more easy to explain …

producer_q = Queue()
def _burst_thread():
   while True:
      available_tasks = [producer_q.get()]
      time.sleep(BURST_TIME_WINDOW)
      available_tasks.extend(producer_q.get() # I'm the single consumer, so will be at least qsize elements  
                             for i in range(producer_q.qsize()))
      consumer_q.push(available_tasks)

If you want to have a maximum of messages in a burst, you just need to slice the available_tasks in multiple lists.