Priority queueing of tasks to one or more threaded or asynchronous workers.
AsynQueue provides asynchronous task queueing based on the Twisted framework, with task prioritization, a powerful worker interface, and optional deferred iteration. Worker implementations are included for running tasks asynchronously in the main thread, in separate threads, in separate Python interpreters (multiprocessing), and even on separate networked devices via Twisted’s Asynchronous Messaging Protocol (AMP).
Multicore and the Mandelbrot Set
The mcMandelbrot package bundled with AsynQueue provides a usage example and a cool little app as well. It runs under the command line or as a web server (port 8080) via twistd. See mcmandelbrot.main and mcmandelbrot.html.
You can run it and see your multcore CPU cranking away doing asynchronous multiprocessing, with a web server under Twisted running besides! When you install AsynQueue with pip install asynqueue, you’ll get a console entry point for mcmandelbrot. Give it a try and see what asynchronous multiprocessing can do, or just click on the image to run it on my quad-core virtual private server.
Each row of the PNG file is computed on a subordinate Python interpreter, with all the rows dispatched and queued up via a ProcessQueue. As the pixel values are computed, color mapped to RGB, and returned as byte arrays, they appear to the blocking png.Writer object as a regular row iterator, via the magic of asynqueue.threads.Filerator.
Intelligently buffering the asynchronous results and iterating them in a proper sequence, using a thread to both yield and accept blocking iterations, allows the use of a third-party library that wouldn’t ordinarily play nice with Twisted.
There’s a lot going on to make everything happen, and it all works beautifully. Take a look at the mcMandelbrot page for details.
Multi-core Logfile Parsing
Another example is my use of AsynQueue to get some CPU cores parsing thousands of lines of logfiles:
class Reader: # Maximum number of logfiles to process concurrently N = 6 # <Stuff...> @defer.inlineCallbacks def run(self): def dispatch(fileName): filePath = self.pathInDir(fileName) # Get a ProcessConsumer for this file consumer = self.rk.consumerFactory(fileName) self.consumers.append(consumer) # Call the ProcessReader on one of my subordinate # processes to have it feed the consumer with # misbehaving IP addresses and filtered records return self.pq.call( self.pr, filePath, consumer=consumer).addCallback(done) def done(consumer): self.consumers.remove(consumer) dList =  self.lock.acquire() self.pq = asynqueue.ProcessQueue(3) # We have at most two files being parsed concurrently for each # worker servicing my process queue ds = defer.DeferredSemaphore(min([self.N, 2*len(self.pq)])) # "Wait" for everything to start up and get a list of # known-bad IP addresses ipList = yield self.rk.startup() # Warn workers to ignore records from the bad IPs yield self.pq.update(self.pr.ignoreIPs, ipList) # Dispatch files as permitted by the semaphore for fileName in self.fileNames: if not self.isRunning(): break # "Wait" for the number of concurrent parsings to fall # back to the limit yield ds.acquire() # If not running, break out of the loop if not self.isRunning(): break # References to the deferreds from dispatch calls are # stored in the process queue, and we wait for their # results. d = dispatch(fileName) d.addCallback(lambda _: ds.release()) dList.append(d) yield defer.DeferredList(dList) ipList = self.rk.getNewIPs() defer.returnValue(ipList) # Can now shut down, regularly or due to interruption self.lock.release()
The Reader object has a ProcessReader object, referenced by its self.pr attribute. The process reader is passed to subordinate Python processes for doing the logfile parsing.
In fact, each call to the reader object’s task queue, base.TaskQueue.call via the process.ProcessQueue subclass instance, passes along a reference to self.pr as a callable. But that’s not a problem, even over the interprocess pipe. Python’s built-in multiprocessing module pickles the reference very efficiently, and almost no CPU time is spent doing so.
Everything done by each subordinate Python process is contained in the following two methods of its copy of the process.ProcessUniverse object:
def next(self, ID): if ID in self.iterators: try: value = self.iterators[ID].next() except StopIteration: del self.iterators[ID] return None, False return value, True return None, False def loop(self, connection): while True: # Wait here for the next call callSpec = connection.recv() if callSpec is None: # Termination call, no reply expected; just exit the # loop break elif isinstance(callSpec, str): # A next-iteration call connection.send(self.next(callSpec)) else: # A task call status, result = self.runner(callSpec) if status == 'i': # Due to the pipe between worker and process, we # hold onto the iterator here and just # return an ID to it ID = str(hash(result)) self.iterators[ID] = result result = ID connection.send((status, result)) # Broken out of loop, ready for the process to end connection.close()
Bridging the Blocking Gap
Yes, the process blocks when it waits for the next call with connection.recv. So what? It’s not running Twisted; the subordinate Python interpreter’s whole purpose in life is to run tasks sent to it via the task queue. And on the main Twisted-running interpreter, here’s what process.ProcessWorker does. Note the magic that happens in the line with yield self.delay.untilEvent(self.cMain.poll):
@defer.inlineCallbacks def run(self, task): if task is None: # A termination task, do after pending tasks are done yield self.dLock.acquire() self.cMain.send(None) # Wait (a very short amount of time) for the process loop # to exit self.process.join() self.dLock.release() else: # A regular task self.tasks.append(task) yield self.dLock.acquire(task.priority <= -20) # Our turn! #------------------------------------------------------------------consumer = task.callTuple.pop('consumer', None) self.cMain.send(task.callTuple) # "Wait" here (in Twisted-friendly fashion) for a response # from the process yield self.delay.untilEvent(self.cMain.poll) status, result = self.cMain.recv() self.dLock.release() if status == 'i': # What we get from the process is an ID to an iterator # it is holding onto, but we need to hook up to it # with a Prefetcherator and then make a Deferator, # which we will either return to the caller or couple # to a consumer provided by the caller. ID = result pf = iteration.Prefetcherator(ID) ok = yield pf.setup(self.next, ID) if ok: result = iteration.Deferator(pf) if consumer: result = iteration.IterationProducer(result, consumer) else: # The process returned an iterator, but it's not # one I could prefetch from. Probably empty. result =  if task in self.tasks: self.tasks.remove(task) task.callback((status, result))
The iteration.Delay object has this very cool capability of providing a Deferred that fires after an event happens. It checks whatever no-argument callable you provide to see if the event has happened yet, and fires the Deferred if so. If not, it waits a while and checks again, with exponential back off to keep the interval between checks approximately proportionate to the amount of time that’s passed. It’s efficient and works very well.
The main Reader object running on the main Python interpreter also has a RecordKeeper object, reference by self.rk, that can provide implementors of twisted.internet.interfaces.IConsumer. Those consumer objects receive the iterations that are produced by iteration.IterationProducer instances, iterating asynchronously “over the wire” (actually, over the interprocess connection pipe).
Copyright (C) 2006-2007, 2015, 2017 by Edwin A. Suominen, http://edsuom.com/:
See edsuom.com for API documentation as well as information about Ed's background and other projects, software and otherwise. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.