程序媛

python——有一种线程池叫做自己写的线程池

  python的线程一直被称为鸡肋,所以它也没有亲生的线程池,但是竟然被我发现了野生的线程池,简直不能更幸运~~~于是,我开始啃源码,实在是虐心,在啃源码的过程中,我简略的了解了python线程的相关知识,感觉还是很有趣的,于是写博客困难症患者一夜之间化身写作小能手,完成了一系列线程相关的博客,然后恍然发现,python的多线程是一个鸡肋哎。。。这里换来了同事们的白眼若干→_→。嘻嘻,但是鸡肋归鸡肋,看懂了一篇源码给我带来的收获和成就感还是不能小视,所以还是分享下~~~

别人的线程池

  首先介绍别人写的线程池模块,野生threadpool,直接到pypi上去搜,或者pip安装,都可以get到。这里还是先贴上来:

  1 # -*- coding: UTF-8 -*-
  2 """Easy to use object-oriented thread pool framework.
  3 
  4 A thread pool is an object that maintains a pool of worker threads to perform
  5 time consuming operations in parallel. It assigns jobs to the threads
  6 by putting them in a work request queue, where they are picked up by the
  7 next available thread. This then performs the requested operation in the
  8 background and puts the results in another queue.
  9 
 10 The thread pool object can then collect the results from all threads from
 11 this queue as soon as they become available or after all threads have
 12 finished their work. It's also possible, to define callbacks to handle
 13 each result as it comes in.
 14 
 15 The basic concept and some code was taken from the book "Python in a Nutshell,
 16 2nd edition" by Alex Martelli, O'Reilly 2006, ISBN 0-596-10046-9, from section
 17 14.5 "Threaded Program Architecture". I wrapped the main program logic in the
 18 ThreadPool class, added the WorkRequest class and the callback system and
 19 tweaked the code here and there. Kudos also to Florent Aide for the exception
 20 handling mechanism.
 21 
 22 Basic usage::
 23 
 24     >>> pool = ThreadPool(poolsize)
 25     >>> requests = makeRequests(some_callable, list_of_args, callback)
 26     >>> [pool.putRequest(req) for req in requests]
 27     >>> pool.wait()
 28 
 29 See the end of the module code for a brief, annotated usage example.
 30 
 31 Website : http://chrisarndt.de/projects/threadpool/
 32 
 33 """
 34 __docformat__ = "restructuredtext en"
 35 
 36 __all__ = [
 37     'makeRequests',
 38     'NoResultsPending',
 39     'NoWorkersAvailable',
 40     'ThreadPool',
 41     'WorkRequest',
 42     'WorkerThread'
 43 ]
 44 
 45 __author__ = "Christopher Arndt"
 46 __version__ = '1.3.2'
 47 __license__ = "MIT license"
 48 
 49 
 50 # standard library modules
 51 import sys
 52 import threading
 53 import traceback
 54 
 55 try:
 56     import Queue            # Python 2
 57 except ImportError:
 58     import queue as Queue   # Python 3
 59 
 60 
 61 # exceptions
 62 class NoResultsPending(Exception):
 63     """All work requests have been processed."""
 64     pass
 65 
 66 class NoWorkersAvailable(Exception):
 67     """No worker threads available to process remaining requests."""
 68     pass
 69 
 70 
 71 # internal module helper functions
 72 def _handle_thread_exception(request, exc_info):
 73     """Default exception handler callback function.
 74 
 75     This just prints the exception info via ``traceback.print_exception``.
 76 
 77     """
 78     traceback.print_exception(*exc_info)
 79 
 80 
 81 # utility functions
 82 def makeRequests(callable_, args_list, callback=None,
 83         exc_callback=_handle_thread_exception):
 84     """Create several work requests for same callable with different arguments.
 85 
 86     Convenience function for creating several work requests for the same
 87     callable where each invocation of the callable receives different values
 88     for its arguments.
 89 
 90     ``args_list`` contains the parameters for each invocation of callable.
 91     Each item in ``args_list`` should be either a 2-item tuple of the list of
 92     positional arguments and a dictionary of keyword arguments or a single,
 93     non-tuple argument.
 94 
 95     See docstring for ``WorkRequest`` for info on ``callback`` and
 96     ``exc_callback``.
 97 
 98     """
 99     requests = []
100     for item in args_list:
101         if isinstance(item, tuple):
102             requests.append(
103                 WorkRequest(callable_, item[0], item[1], callback=callback,
104                     exc_callback=exc_callback)
105             )
106         else:
107             requests.append(
108                 WorkRequest(callable_, [item], None, callback=callback,
109                     exc_callback=exc_callback)
110             )
111     return requests
112 
113 
114 # classes
115 class WorkerThread(threading.Thread):
116     """Background thread connected to the requests/results queues.
117 
118     A worker thread sits in the background and picks up work requests from
119     one queue and puts the results in another until it is dismissed.
120 
121     """
122 
123     def __init__(self, requests_queue, results_queue, poll_timeout=5, **kwds):
124         """Set up thread in daemonic mode and start it immediatedly.
125 
126         ``requests_queue`` and ``results_queue`` are instances of
127         ``Queue.Queue`` passed by the ``ThreadPool`` class when it creates a
128         new worker thread.
129 
130         """
131         threading.Thread.__init__(self, **kwds)
132         self.setDaemon(1)
133         self._requests_queue = requests_queue
134         self._results_queue = results_queue
135         self._poll_timeout = poll_timeout
136         self._dismissed = threading.Event()
137         self.start()
138 
139     def run(self):
140         """Repeatedly process the job queue until told to exit."""
141         while True:
142             if self._dismissed.isSet():
143                 # we are dismissed, break out of loop
144                 break
145             # get next work request. If we don't get a new request from the
146             # queue after self._poll_timout seconds, we jump to the start of
147             # the while loop again, to give the thread a chance to exit.
148             try:
149                 request = self._requests_queue.get(True, self._poll_timeout)
150             except Queue.Empty:
151                 continue
152             else:
153                 if self._dismissed.isSet():
154 
155                     # we are dismissed, put back request in queue and exit loop
156                     self._requests_queue.put(request)
157                     break
158                 try:
159                     result = request.callable(*request.args, **request.kwds)
160                     self._results_queue.put((request, result))
161                 except:
162                     request.exception = True
163                     self._results_queue.put((request, sys.exc_info()))
164 
165     def dismiss(self):
166         print '**********dismiss***********'
167         """Sets a flag to tell the thread to exit when done with current job.
168         """
169         self._dismissed.set()
170 
171 
172 class WorkRequest:
173     """A request to execute a callable for putting in the request queue later.
174 
175     See the module function ``makeRequests`` for the common case
176     where you want to build several ``WorkRequest`` objects for the same
177     callable but with different arguments for each call.
178 
179     """
180 
181     def __init__(self, callable_, args=None, kwds=None, requestID=None,
182             callback=None, exc_callback=_handle_thread_exception):
183         """Create a work request for a callable and attach callbacks.
184 
185         A work request consists of the a callable to be executed by a
186         worker thread, a list of positional arguments, a dictionary
187         of keyword arguments.
188 
189         A ``callback`` function can be specified, that is called when the
190         results of the request are picked up from the result queue. It must
191         accept two anonymous arguments, the ``WorkRequest`` object and the
192         results of the callable, in that order. If you want to pass additional
193         information to the callback, just stick it on the request object.
194 
195         You can also give custom callback for when an exception occurs with
196         the ``exc_callback`` keyword parameter. It should also accept two
197         anonymous arguments, the ``WorkRequest`` and a tuple with the exception
198         details as returned by ``sys.exc_info()``. The default implementation
199         of this callback just prints the exception info via
200         ``traceback.print_exception``. If you want no exception handler
201         callback, just pass in ``None``.
202 
203         ``requestID``, if given, must be hashable since it is used by
204         ``ThreadPool`` object to store the results of that work request in a
205         dictionary. It defaults to the return value of ``id(self)``.
206 
207         """
208         #__init__(  callable_, args=None,  kwds=None,  callback=None,    exc_callback=_handle_thread_exception)
209         #WorkRequest(callable_, item[0],   item[1],    callback=callback,exc_callback=exc_callback)
210         if requestID is None:
211             self.requestID = id(self)
212         else:
213             try:
214                 self.requestID = hash(requestID)
215             except TypeError:
216                 raise TypeError("requestID must be hashable.")
217         self.exception = False
218         self.callback = callback
219         self.exc_callback = exc_callback
220         self.callable = callable_
221         self.args = args or []
222         self.kwds = kwds or {}
223 
224     def __str__(self):
225         return "<WorkRequest id=%s args=%r kwargs=%r exception=%s>" % \
226             (self.requestID, self.args, self.kwds, self.exception)
227 
228 class ThreadPool:
229     """A thread pool, distributing work requests and collecting results.
230 
231     See the module docstring for more information.
232 
233     """
234 
235     def __init__(self, num_workers, q_size=0, resq_size=0, poll_timeout=5):
236         """Set up the thread pool and start num_workers worker threads.
237 
238         ``num_workers`` is the number of worker threads to start initially.
239 
240         If ``q_size > 0`` the size of the work *request queue* is limited and
241         the thread pool blocks when the queue is full and it tries to put
242         more work requests in it (see ``putRequest`` method), unless you also
243         use a positive ``timeout`` value for ``putRequest``.
244 
245         If ``resq_size > 0`` the size of the *results queue* is limited and the
246         worker threads will block when the queue is full and they try to put
247         new results in it.
248 
249         .. warning:
250             If you set both ``q_size`` and ``resq_size`` to ``!= 0`` there is
251             the possibilty of a deadlock, when the results queue is not pulled
252             regularly and too many jobs are put in the work requests queue.
253             To prevent this, always set ``timeout > 0`` when calling
254             ``ThreadPool.putRequest()`` and catch ``Queue.Full`` exceptions.
255 
256         """
257         self._requests_queue = Queue.Queue(q_size)
258         self._results_queue = Queue.Queue(resq_size)
259         self.workers = []
260         self.dismissedWorkers = []
261         self.workRequests = {}
262         self.createWorkers(num_workers, poll_timeout)
263 
264     def createWorkers(self, num_workers, poll_timeout=5):
265         """Add num_workers worker threads to the pool.
266 
267         ``poll_timout`` sets the interval in seconds (int or float) for how
268         ofte threads should check whether they are dismissed, while waiting for
269         requests.
270 
271         """
272         for i in range(num_workers):
273             self.workers.append(WorkerThread(self._requests_queue,
274                 self._results_queue, poll_timeout=poll_timeout))
275 
276     def dismissWorkers(self, num_workers, do_join=False):
277         """Tell num_workers worker threads to quit after their current task."""
278         dismiss_list = []
279         for i in range(min(num_workers, len(self.workers))):
280             worker = self.workers.pop()
281             worker.dismiss()
282             dismiss_list.append(worker)
283 
284         if do_join:
285             for worker in dismiss_list:
286                 worker.join()
287         else:
288             self.dismissedWorkers.extend(dismiss_list)
289 
290     def joinAllDismissedWorkers(self):
291         """Perform Thread.join() on all worker threads that have been dismissed.
292         """
293         for worker in self.dismissedWorkers:
294             worker.join()
295         self.dismissedWorkers = []
296 
297     def putRequest(self, request, block=True, timeout=None):
298         """Put work request into work queue and save its id for later."""
299         assert isinstance(request, WorkRequest)
300         # don't reuse old work requests
301         assert not getattr(request, 'exception', None)
302         import time
303         self._requests_queue.put(request, block, timeout)
304         self.workRequests[request.requestID] = request
305 
306     def poll(self, block=False):
307         """Process any new results in the queue."""
308         while True:
309             # still results pending?
310             if not self.workRequests:
311                 raise NoResultsPending
312             # are there still workers to process remaining requests?
313             elif block and not self.workers:
314                 raise NoWorkersAvailable
315             try:
316                 # get back next results
317 
318                 request, result = self._results_queue.get(block=block)
319 
320                 # has an exception occured?
321                 if request.exception and request.exc_callback:
322 
323                     request.exc_callback(request, result)
324 
325                 # hand results to callback, if any
326                 if request.callback and not \
327                        (request.exception and request.exc_callback):
328                     request.callback(request, result)
329                 del self.workRequests[request.requestID]
330             except Queue.Empty:
331                 break
332 
333     def wait(self):
334         """Wait for results, blocking until all have arrived."""
335         while 1:
336             try:
337                 self.poll(True)
338             except NoResultsPending:
339                 break
340 
341 
342 ################
343 # USAGE EXAMPLE
344 ################
345 
346 if __name__ == '__main__':
347     import random
348     import time
349 
350     # the work the threads will have to do (rather trivial in our example)
351     def do_something(data):
352         time.sleep(random.randint(1,5))
353         result = round(random.random() * data, 5)
354         # just to show off, we throw an exception once in a while
355         if result > 5:
356             raise RuntimeError("Something extraordinary happened!")
357         return result
358 
359     # this will be called each time a result is available
360     def print_result(request, result):
361         print("**** Result from request #%s: %r" % (request.requestID, result))
362 
363     # this will be called when an exception occurs within a thread
364     # this example exception handler does little more than the default handler
365     def handle_exception(request, exc_info):
366         if not isinstance(exc_info, tuple):
367             # Something is seriously wrong...
368             print(request)
369             print(exc_info)
370             raise SystemExit
371         print("**** Exception occured in request #%s: %s" % \
372           (request.requestID, exc_info))
373 
374     # assemble the arguments for each job to a list...
375     data = [random.randint(1,10) for i in range(20)]
376     # ... and build a WorkRequest object for each item in data
377     requests = makeRequests(do_something, data, print_result, handle_exception)
378     # to use the default exception handler, uncomment next line and comment out
379     # the preceding one.
380     #requests = makeRequests(do_something, data, print_result)
381 
382     # or the other form of args_lists accepted by makeRequests: ((,), {})
383     data = [((random.randint(1,10),), {}) for i in range(20)]
384     requests.extend(
385         makeRequests(do_something, data, print_result, handle_exception)
386         #makeRequests(do_something, data, print_result)
387         # to use the default exception handler, uncomment next line and comment
388         # out the preceding one.
389     )
390 
391     # we create a pool of 3 worker threads
392     print("Creating thread pool with 3 worker threads.")
393     main = ThreadPool(3)
394 
395     # then we put the work requests in the queue...
396     for req in requests:
397         main.putRequest(req)
398         print("Work request #%s added." % req.requestID)
399     # or shorter:
400     # [main.putRequest(req) for req in requests]
401 
402     # ...and wait for the results to arrive in the result queue
403     # by using ThreadPool.wait(). This would block until results for
404     # all work requests have arrived:
405     # main.wait()
406 
407     # instead we can poll for results while doing something else:
408     i = 0
409     while True:
410         try:
411             time.sleep(0.5)
412             main.poll()
413             print("Main thread working...")
414             print("(active worker threads: %i)" % (threading.activeCount()-1, ))
415             if i == 10:
416 
417                 main.createWorkers(3)
418             if i == 20:
419 
420                 main.dismissWorkers(2)
421             i += 1
422         except KeyboardInterrupt:
423             print("**** Interrupted!")
424             break
425         except NoResultsPending:
426             print("**** No pending results.")
427             break
428     if main.dismissedWorkers:
429         print("Joining all dismissed worker threads...")
430         main.joinAllDismissedWorkers()
threadpool Code

  首先我们来看这个线程池的大致原理。在初始化中,它会根据我们的需求,启动相应数量的线程,这些线程是初始化好的,一直到程序结束,不会停止,它们从任务队列中获取任务,在没有任务的时候就阻塞,他们当我们有任务的时候,对任务进行初始化,放入任务队列,拿到任务的线程结束了自己的阻塞人生,欢欢喜喜的拿回去执行,并在执行完毕之后,将结果放入结果队列,继续到任务队列中取任务,如果没有任务就进入阻塞状态。看了一整天的源码竟让我三两句话解释清楚了,我到底是表达能力强还是理解能力差!!!我想静静~~~附上类图如下:

    我的线程池

  下面就来介绍我写的线程池了,上面的线程池有一个问题,那就是一开始创建了多少个线程,这些线程就一直存在内存中,即使没有工作,也不会销毁。于是我有了一个想法,就像其他语言中的线程池一样,写一个拥有最大线程数和最小线程数限制的线程池。

  程序启动之初只将最小线程数的线程放在池中,并将线程设置为阻塞状态,用守护线程来查看任务队列,当任务队列中有任务时,则停止线程的阻塞状态,让它们到队列中去获取任务,执行,如果需要返回结果,则将结果返回结果队列。当任务很多,线程池中没有闲置的线程且当前线程数小于线程池最大线程数时,将创建新的线程(这里使用了yield)来接收新的任务,线程执行完毕后,则回到阻塞状态,长期闲置的线程会自动销毁,但池中线程永远不小于在最小线程数。当最小线程数和最大线程数相等的时候,内部就基本和野生线程相同啦~~~

  在参考了野生threadpool模块之后,我也学着继承原生的threading.Thread类,并重写了run方法,了解了给一个线程注入新方法的过程。并用到了Event方法和yield。如果不要返回值的话,我想效率还是很高的。尽管我在返回值方面还做了优化,哎~~~

  银角大王的线程池:

 1 from Queue import Queue 
 2 import contextlib 
 3 import threading 
 4    
 5 WorkerStop = object() 
 6    
 7    
 8 class ThreadPool: 
 9    
10     workers = 0
11    
12     threadFactory = threading.Thread 
13     currentThread = staticmethod(threading.currentThread) 
14    
15     def __init__(self, maxthreads=20, name=None): 
16    
17         self.q = Queue(0) 
18         self.max = maxthreads 
19         self.name = name 
20         self.waiters = [] 
21         self.working = [] 
22    
23     def start(self): 
24         needsiZe = self.q.qsize() 
25         while self.workers < min(self.max, needSize): 
26             self.startAWorker() 
27    
28     def startAWorker(self): 
29         self.workers += 1
30         name = "PoolThread-%s-%s" % (self.name or id(self), self.workers) 
31         newThread = self.threadFactory(target=self._worker, name=name) 
32         newThread.start() 
33    
34     def callInThread(self, func, *args, **kw): 
35         self.callInThreadWithCallback(None, func, *args, **kw) 
36    
37     def callInThreadWithCallback(self, onResult, func, *args, **kw): 
38         o = (func, args, kw, onResult) 
39         self.q.put(o) 
40    
41    
42     @contextlib.contextmanager 
43     def _workerState(self, stateList, workerThread): 
44         stateList.append(workerThread) 
45         try: 
46             yield
47         finally: 
48             stateList.remove(workerThread) 
49    
50     def _worker(self): 
51         ct = self.currentThread() 
52         o = self.q.get() 
53         while o is not WorkerStop: 
54             with self._workerState(self.working, ct): 
55                 function, args, kwargs, onResult = o 
56                 del o 
57                 try: 
58                     result = function(*args, **kwargs) 
59                     success = True
60                 except: 
61                     success = False
62                     if onResult is None: 
63                         pass
64    
65                     else: 
66                         pass
67    
68                 del function, args, kwargs 
69    
70                 if onResult is not None: 
71                     try: 
72                         onResult(success, result) 
73                     except: 
74                         #context.call(ctx, log.err) 
75                         pass
76    
77                 del onResult, result 
78    
79             with self._workerState(self.waiters, ct): 
80                 o = self.q.get() 
81    
82     def stop(self): 
83         while self.workers: 
84             self.q.put(WorkerStop) 
85             self.workers -= 1
86  
threadpool Code
 1 def show(arg): 
 2     import time 
 3     time.sleep(1) 
 4     print arg 
 5    
 6    
 7 pool = ThreadPool(20) 
 8    
 9 for i in range(500): 
10     pool.callInThread(show, i) 
11    
12 pool.start() 
13 pool.stop()
use example Code

  这里安利下我男神,哈哈哈~武sir的方法和上面的例子中不同的是,自定义了线程的start方法,当启动线程的时候才初始化线程池,并根据线程池定义的数量和任务数量取min,而不是先开启定义的线程数等待命令,在一定程度上避免了空线程对内存的消耗。

with知识点

  这里要介绍一个知识点。我们在做上下文管理的时候,用到过with。

  我们如何自定义一个with方法呢?

  

  如此一来,我们便可以实现对线程状态的监控和管理了。将正在运行中的线程,加入到一个列表中,并使用yield返回,当线程执行完之后,再从这个列表中移除,就可以知道哪些线程是正在运行的啦。

  

posted @ 2016-01-08 11:10 Eva_J 阅读(...) 评论(...) 编辑 收藏