Welcome, guest | Sign In | My Account | Store | Cart

There are probably <write your guess here>s of recipes presenting how to implement a pool of threads. Now that multiprocessing is becoming mainstream, this recipe takes multiprocessing.Pool as a model and re-implements it entirely with threads. Even the comments should look familiar... This recipe also adds 2 new methods: imap_async() and imap_unordered_async().

Python, 787 lines
  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
#
# @brief Pool of threads similar to multiprocessing.Pool
#
# Copyright (c) 2008,2016 david decotigny (this file)
# Copyright (c) 2006-2008, R Oudkerk (multiprocessing.Pool)
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
#
# 1. Redistributions of source code must retain the above copyright
#    notice, this list of conditions and the following disclaimer.
# 2. Redistributions in binary form must reproduce the above copyright
#    notice, this list of conditions and the following disclaimer in the
#    documentation and/or other materials provided with the distribution.
# 3. Neither the name of author nor the names of any contributors may be
#    used to endorse or promote products derived from this software
#    without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS "AS IS" AND
# ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
# ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
# OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
# HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
# LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
# OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
# SUCH DAMAGE.
#

# See http://docs.python.org/dev/library/multiprocessing.html
# Differences: added imap_async and imap_unordered_async, and terminate()
# has to be called explicitly (it's not registered by atexit).
#
# The general idea is that we submit works to a workqueue, either as
# single Jobs (one function to call), or JobSequences (batch of
# Jobs). Each Job is associated with an ApplyResult object which has 2
# states: waiting for the Job to complete, or Ready. Instead of
# waiting for the jobs to finish, we wait for their ApplyResult object
# to become ready: an event mechanism is used for that.
# When we apply a function to several arguments in "parallel", we need
# a way to wait for all/part of the Jobs to be processed: that's what
# "collectors" are for; they group and wait for a set of ApplyResult
# objects. Once a collector is ready to be used, we can use a
# CollectorIterator to iterate over the result values it's collecting.
#
# The methods of a Pool object use all these concepts and expose
# them to their caller in a very simple way.

import sys, threading, Queue, traceback


## Item pushed on the work queue to tell the worker threads to terminate
SENTINEL = "QUIT"
def is_sentinel(obj):
    """Predicate to determine whether an item from the queue is the
    signal to stop"""
    return type(obj) is str and obj == SENTINEL


class TimeoutError(Exception):
    """Raised when a result is not available within the given timeout"""
    pass


class PoolWorker(threading.Thread):
    """Thread that consumes WorkUnits from a queue to process them"""
    def __init__(self, workq, *args, **kwds):
        """\param workq: Queue object to consume the work units from"""
        threading.Thread.__init__(self, *args, **kwds)
        self._workq = workq

    def run(self):
        """Process the work unit, or wait for sentinel to exit"""
        while 1:
            workunit = self._workq.get()
            if is_sentinel(workunit):
                # Got sentinel
                break

            # Run the job / sequence
            workunit.process()


class Pool(object):
    """
    The Pool class represents a pool of worker threads. It has methods
    which allows tasks to be offloaded to the worker processes in a
    few different ways
    """

    def __init__(self, nworkers, name="Pool"):
        """
        \param nworkers (integer) number of worker threads to start
        \param name (string) prefix for the worker threads' name
        """
        self._workq   = Queue.Queue()
        self._closed  = False
        self._workers = []
        for idx in xrange(nworkers):
            thr = PoolWorker(self._workq, name="Worker-%s-%d" % (name, idx))
            try:
                thr.start()
            except:
                # If one thread has a problem, undo everything
                self.terminate()
                raise
            else:
                self._workers.append(thr)

    def apply(self, func, args=(), kwds=dict()):
        """Equivalent of the apply() builtin function. It blocks till
        the result is ready."""
        return self.apply_async(func, args, kwds).get()

    def map(self, func, iterable, chunksize=None):
        """A parallel equivalent of the map() builtin function. It
        blocks till the result is ready.

        This method chops the iterable into a number of chunks which
        it submits to the process pool as separate tasks. The
        (approximate) size of these chunks can be specified by setting
        chunksize to a positive integer."""
        return self.map_async(func, iterable, chunksize).get()

    def imap(self, func, iterable, chunksize=1):
        """
        An equivalent of itertools.imap().

        The chunksize argument is the same as the one used by the
        map() method. For very long iterables using a large value for
        chunksize can make make the job complete much faster than
        using the default value of 1.

        Also if chunksize is 1 then the next() method of the iterator
        returned by the imap() method has an optional timeout
        parameter: next(timeout) will raise processing.TimeoutError if
        the result cannot be returned within timeout seconds.
        """
        collector = OrderedResultCollector(as_iterator=True)
        self._create_sequences(func, iterable, chunksize, collector)
        return iter(collector)

    def imap_unordered(self, func, iterable, chunksize=1):
        """The same as imap() except that the ordering of the results
        from the returned iterator should be considered
        arbitrary. (Only when there is only one worker process is the
        order guaranteed to be "correct".)"""
        collector = UnorderedResultCollector()
        self._create_sequences(func, iterable, chunksize, collector)
        return iter(collector)

    def apply_async(self, func, args=(), kwds=dict(), callback=None):
        """A variant of the apply() method which returns an
        ApplyResult object.

        If callback is specified then it should be a callable which
        accepts a single argument. When the result becomes ready,
        callback is applied to it (unless the call failed). callback
        should complete immediately since otherwise the thread which
        handles the results will get blocked."""
        assert not self._closed # No lock here. We assume it's atomic...
        apply_result = ApplyResult(callback=callback)
        job = Job(func, args, kwds, apply_result)
        self._workq.put(job)
        return apply_result

    def map_async(self, func, iterable, chunksize=None, callback=None):
        """A variant of the map() method which returns a ApplyResult
        object.

        If callback is specified then it should be a callable which
        accepts a single argument. When the result becomes ready
        callback is applied to it (unless the call failed). callback
        should complete immediately since otherwise the thread which
        handles the results will get blocked."""
        apply_result = ApplyResult(callback=callback)
        collector    = OrderedResultCollector(apply_result, as_iterator=False)
        if not self._create_sequences(func, iterable, chunksize, collector):
          apply_result._set_value([])
        return apply_result

    def imap_async(self, func, iterable, chunksize=None, callback=None):
        """A variant of the imap() method which returns an ApplyResult
        object that provides an iterator (next method(timeout)
        available).

        If callback is specified then it should be a callable which
        accepts a single argument. When the resulting iterator becomes
        ready, callback is applied to it (unless the call
        failed). callback should complete immediately since otherwise
        the thread which handles the results will get blocked."""
        apply_result = ApplyResult(callback=callback)
        collector    = OrderedResultCollector(apply_result, as_iterator=True)
        if not self._create_sequences(func, iterable, chunksize, collector):
          apply_result._set_value(iter([]))
        return apply_result

    def imap_unordered_async(self, func, iterable, chunksize=None,
                             callback=None):
        """A variant of the imap_unordered() method which returns an
        ApplyResult object that provides an iterator (next
        method(timeout) available).

        If callback is specified then it should be a callable which
        accepts a single argument. When the resulting iterator becomes
        ready, callback is applied to it (unless the call
        failed). callback should complete immediately since otherwise
        the thread which handles the results will get blocked."""
        apply_result = ApplyResult(callback=callback)
        collector    = UnorderedResultCollector(apply_result)
        if not self._create_sequences(func, iterable, chunksize, collector):
          apply_result._set_value(iter([]))
        return apply_result

    def close(self):
        """Prevents any more tasks from being submitted to the
        pool. Once all the tasks have been completed the worker
        processes will exit."""
        # No lock here. We assume it's sufficiently atomic...
        self._closed = True

    def terminate(self):
        """Stops the worker processes immediately without completing
        outstanding work. When the pool object is garbage collected
        terminate() will be called immediately."""
        self.close()

        # Clearing the job queue
        try:
            while 1:
                self._workq.get_nowait()
        except Queue.Empty:
            pass

        # Send one sentinel for each worker thread: each thread will die
        # eventually, leaving the next sentinel for the next thread
        for thr in self._workers:
            self._workq.put(SENTINEL)

    def join(self):
        """Wait for the worker processes to exit. One must call
        close() or terminate() before using join()."""
        for thr in self._workers:
            thr.join()

    def _create_sequences(self, func, iterable, chunksize, collector = None):
        """
        Create the WorkUnit objects to process and pushes them on the
        work queue. Each work unit is meant to process a slice of
        iterable of size chunksize. If collector is specified, then
        the ApplyResult objects associated with the jobs will notify
        collector when their result becomes ready.

        \return the list of WorkUnit objects (basically: JobSequences)
        pushed onto the work queue, empty if iterable itself was empty
        """
        assert not self._closed # No lock here. We assume it's atomic...
        sequences = []
        results   = []
        it_ = iter(iterable)
        exit_loop = False
        while not exit_loop:
            seq = []
            for i in xrange(chunksize or 1):
                try:
                    arg = it_.next()
                except StopIteration:
                    exit_loop = True
                    break
                apply_result = ApplyResult(collector)
                job = Job(func, (arg,), {}, apply_result)
                seq.append(job)
                results.append(apply_result)
            if seq:
              sequences.append(JobSequence(seq))

        for seq in sequences:
          self._workq.put(seq)

        return sequences


class WorkUnit(object):
    """ABC for a unit of work submitted to the worker threads. It's
    basically just an object equipped with a process() method"""
    def process(self):
        """Do the work. Shouldn't raise any exception"""
        raise NotImplementedError("Children must override Process")


class Job(WorkUnit):
    """A work unit that corresponds to the execution of a single function"""
    def __init__(self, func, args, kwds, apply_result):
        """
        \param func/args/kwds used to call the function
        \param apply_result ApplyResult object that holds the result
        of the function call
        """
        WorkUnit.__init__(self)
        self._func   = func
        self._args   = args
        self._kwds   = kwds
        self._result = apply_result

    def process(self):
        """
        Call the function with the args/kwds and tell the ApplyResult
        that its result is ready. Correctly handles the exceptions
        happening during the execution of the function
        """
        try:
            result = self._func(*self._args, **self._kwds)
        except:
            self._result._set_exception()
        else:
            self._result._set_value(result)


class JobSequence(WorkUnit):
    """A work unit that corresponds to the processing of a continuous
    sequence of Job objects"""
    def __init__(self, jobs):
        WorkUnit.__init__(self)
        self._jobs = jobs

    def process(self):
        """
        Call process() on all the Job objects that have been specified
        """
        for job in self._jobs:
            job.process()


class ApplyResult(object):
    """An object associated with a Job object that holds its result:
    it's available during the whole life the Job and after, even when
    the Job didn't process yet. It's possible to use this object to
    wait for the result/exception of the job to be available.

    The result objects returns by the Pool::*_async() methods are of
    this type"""
    def __init__(self, collector = None, callback = None):
        """
        \param collector when not None, the notify_ready() method of
        the collector will be called when the result from the Job is
        ready
        \param callback when not None, function to call when the
        result becomes available (this is the paramater passed to the
        Pool::*_async() methods.
        """
        self._success   = False
        self._event     = threading.Event()
        self._data      = None
        self._collector = None
        self._callback  = callback

        if collector is not None:
            collector.register_result(self)
            self._collector = collector

    def get(self, timeout = None):
        """
        Returns the result when it arrives. If timeout is not None and
        the result does not arrive within timeout seconds then
        TimeoutError is raised. If the remote call raised an exception
        then that exception will be reraised by get().
        """
        if not self.wait(timeout):
            raise TimeoutError("Result not available within %fs" % timeout)
        if self._success:
            return self._data
        raise self._data[0], self._data[1], self._data[2]

    def wait(self, timeout = None):
        """Waits until the result is available or until timeout
        seconds pass."""
        self._event.wait(timeout)
        return self._event.isSet()

    def ready(self):
        """Returns whether the call has completed."""
        return self._event.isSet()

    def successful(self):
        """Returns whether the call completed without raising an
        exception. Will raise AssertionError if the result is not
        ready."""
        assert self.ready()
        return self._success

    def _set_value(self, value):
        """Called by a Job object to tell the result is ready, and
        provides the value of this result. The object will become
        ready and successful. The collector's notify_ready() method
        will be called, and the callback method too"""
        assert not self.ready()
        self._data    = value
        self._success = True
        self._event.set()
        if self._collector is not None:
            self._collector.notify_ready(self)
        if self._callback is not None:
            try:
                self._callback(value)
            except:
                traceback.print_exc()

    def _set_exception(self):
        """Called by a Job object to tell that an exception occured
        during the processing of the function. The object will become
        ready but not successful. The collector's notify_ready()
        method will be called, but NOT the callback method"""
        # traceback.print_exc()
        assert not self.ready()
        self._data    = sys.exc_info()
        self._success = False
        self._event.set()
        if self._collector is not None:
            self._collector.notify_ready(self)


class AbstractResultCollector(object):
    """ABC to define the interface of a ResultCollector object. It is
    basically an object which knows whuich results it's waiting for,
    and which is able to get notify when they get available. It is
    also able to provide an iterator over the results when they are
    available"""

    def __init__(self, to_notify):
        """
        \param to_notify ApplyResult object to notify when all the
        results we're waiting for become available. Can be None.
        """
        self._to_notify = to_notify

    def register_result(self, apply_result):
        """Used to identify which results we're waiting for. Will
        always be called BEFORE the Jobs get submitted to the work
        queue, and BEFORE the __iter__ and _get_result() methods can
        be called
        \param apply_result ApplyResult object to add in our collection
        """
        raise NotImplementedError("Children classes must implement it")

    def notify_ready(self, apply_result):
        """Called by the ApplyResult object (already registered via
        register_result()) that it is now ready (ie. the Job's result
        is available or an exception has been raised).
        \param apply_result ApplyResult object telling us that the job
        has been processed
        """
        raise NotImplementedError("Children classes must implement it")

    def _get_result(self, idx, timeout = None):
        """Called by the CollectorIterator object to retrieve the
        result's values one after another (order defined by the
        implementation)
        \param idx The index of the result we want, wrt collector's order
        \param timeout integer telling how long to wait (in seconds)
        for the result at index idx to be available, or None (wait
        forever)
        """
        raise NotImplementedError("Children classes must implement it")

    def __iter__(self):
        """Return a new CollectorIterator object for this collector"""
        return CollectorIterator(self)


class CollectorIterator(object):
    """An iterator that allows to iterate over the result values
    available in the given collector object. Equipped with an extended
    next() method accepting a timeout argument. Created by the
    AbstractResultCollector::__iter__() method"""
    def __init__(self, collector):
        """\param AbstractResultCollector instance"""
        self._collector = collector
        self._idx       = 0

    def __iter__(self):
        return self

    def next(self, timeout = None):
        """Return the next result value in the sequence. Raise
        StopIteration at the end. Can raise the exception raised by
        the Job"""
        try:
            apply_result = self._collector._get_result(self._idx, timeout)
        except IndexError:
            # Reset for next time
            self._idx = 0
            raise StopIteration
        except:
            self._idx = 0
            raise
        self._idx += 1
        assert apply_result.ready()
        return apply_result.get(0)


class UnorderedResultCollector(AbstractResultCollector):
    """An AbstractResultCollector implementation that collects the
    values of the ApplyResult objects in the order they become ready. The
    CollectorIterator object returned by __iter__() will iterate over
    them in the order they become ready"""

    def __init__(self, to_notify = None):
        """
        \param to_notify ApplyResult object to notify when all the
        results we're waiting for become available. Can be None.
        """
        AbstractResultCollector.__init__(self, to_notify)
        self._cond       = threading.Condition()
        self._collection = []
        self._expected   = 0

    def register_result(self, apply_result):
        """Used to identify which results we're waiting for. Will
        always be called BEFORE the Jobs get submitted to the work
        queue, and BEFORE the __iter__ and _get_result() methods can
        be called
        \param apply_result ApplyResult object to add in our collection
        """
        self._expected += 1

    def _get_result(self, idx, timeout = None):
        """Called by the CollectorIterator object to retrieve the
        result's values one after another, in the order the results have
        become available.
        \param idx The index of the result we want, wrt collector's order
        \param timeout integer telling how long to wait (in seconds)
        for the result at index idx to be available, or None (wait
        forever)
        """
        self._cond.acquire()
        try:
            if idx >= self._expected:
                raise IndexError
            elif idx < len(self._collection):
                return self._collection[idx]
            elif idx != len(self._collection):
                # Violation of the sequence protocol
                raise IndexError()
            else:
                self._cond.wait(timeout=timeout)
                try:
                    return self._collection[idx]
                except IndexError:
                    # Still not added !
                    raise TimeoutError("Timeout while waiting for results")
        finally:
            self._cond.release()

    def notify_ready(self, apply_result = None):
        """Called by the ApplyResult object (already registered via
        register_result()) that it is now ready (ie. the Job's result
        is available or an exception has been raised).
        \param apply_result ApplyResult object telling us that the job
        has been processed
        """
        first_item = False
        self._cond.acquire()
        try:
            self._collection.append(apply_result)
            first_item = (len(self._collection) == 1)

            self._cond.notifyAll()
        finally:
            self._cond.release()

        if first_item and self._to_notify is not None:
            self._to_notify._set_value(iter(self))


class OrderedResultCollector(AbstractResultCollector):
    """An AbstractResultCollector implementation that collects the
    values of the ApplyResult objects in the order they have been
    submitted. The CollectorIterator object returned by __iter__()
    will iterate over them in the order they have been submitted"""

    def __init__(self, to_notify = None, as_iterator = True):
        """
        \param to_notify ApplyResult object to notify when all the
        results we're waiting for become available. Can be None.
        \param as_iterator boolean telling whether the result value
        set on to_notify should be an iterator (available as soon as 1
        result arrived) or a list (available only after the last
        result arrived)
        """
        AbstractResultCollector.__init__(self, to_notify)
        self._results     = []
        self._lock        = threading.Lock()
        self._remaining   = 0
        self._as_iterator = as_iterator

    def register_result(self, apply_result):
        """Used to identify which results we're waiting for. Will
        always be called BEFORE the Jobs get submitted to the work
        queue, and BEFORE the __iter__ and _get_result() methods can
        be called
        \param apply_result ApplyResult object to add in our collection
        """
        self._results.append(apply_result)
        self._remaining += 1

    def _get_result(self, idx, timeout = None):
        """Called by the CollectorIterator object to retrieve the
        result's values one after another (order defined by the
        implementation)
        \param idx The index of the result we want, wrt collector's order
        \param timeout integer telling how long to wait (in seconds)
        for the result at index idx to be available, or None (wait
        forever)
        """
        res = self._results[idx]
        res.wait(timeout)
        return res

    def notify_ready(self, apply_result):
        """Called by the ApplyResult object (already registered via
        register_result()) that it is now ready (ie. the Job's result
        is available or an exception has been raised).
        \param apply_result ApplyResult object telling us that the job
        has been processed
        """
        got_first = False
        got_last  = False
        self._lock.acquire()
        try:
            assert self._remaining > 0
            got_first = (len(self._results) == self._remaining)
            self._remaining -= 1
            got_last  = (self._remaining == 0)
        finally:
            self._lock.release()

        if self._to_notify is not None:
            if self._as_iterator and got_first:
                self._to_notify._set_value(iter(self))
            elif not self._as_iterator and got_last:
                try:
                    lst = [r.get(0) for r in self._results]
                except:
                    self._to_notify._set_exception()
                else:
                    self._to_notify._set_value(lst)


def _test():
    """Some tests"""
    import thread, time

    def return42():
        return 42

    def f(x):
        return x*x

    def work(seconds):
        print "[%d] Start to work for %fs..." % (thread.get_ident(), seconds)
        time.sleep(seconds)
        print "[%d] Work done (%fs)." % (thread.get_ident(), seconds)
        return str(seconds)

    ### Test copy/pasted from multiprocessing
    pool = Pool(9)                # start 4 worker threads

    # edge cases
    assert pool.apply_async(return42, []).get() == 42
    assert pool.apply(return42, []) == 42
    assert pool.map(return42, []) == []
    assert list(pool.imap(return42, iter([]))) == []
    assert list(pool.imap_unordered(return42, iter([]))) == []
    assert pool.map_async(return42, []).get() == []
    assert list(pool.imap_async(return42, iter([])).get()) == []
    assert list(pool.imap_unordered_async(return42, iter([])).get()) == []

    # basic tests
    result = pool.apply_async(f, (10,))   # evaluate "f(10)" asynchronously
    assert result.get(timeout=1) == 100   # ... unless slow computer

    assert pool.map(f, range(10)) == map(f, range(10))

    it = pool.imap(f, range(10))
    assert it.next() == 0
    assert it.next() == 1
    assert it.next(timeout=1) == 4  # ... unless slow computer

    # Test apply_sync exceptions
    result = pool.apply_async(time.sleep, (3,))
    try:
        print result.get(timeout=1)           # raises `TimeoutError`
    except TimeoutError:
        print "Good. Got expected timeout exception."
    else:
        assert False, "Expected exception !"
    assert result.get() == None  # sleep() returns None

    def cb(s):
        print "Result ready: %s" % s

    # Test imap()
    assert list(pool.imap(work, xrange(10, 3, -1), chunksize=4)) == map(
        str, range(10, 3, -1))

    # Test imap_unordered()
    assert sorted(pool.imap_unordered(work, xrange(10, 3, -1))) == sorted(map(
        str, range(10, 3, -1)))

    # Test map_async()
    result = pool.map_async(work, xrange(10), callback=cb)
    try:
        result.get(timeout=1)           # raises `TimeoutError`
    except TimeoutError:
        print "Good. Got expected timeout exception."
    else:
        assert False, "Expected exception !"
    print result.get()

    # Test imap_async()
    result = pool.imap_async(work, xrange(3, 10), callback=cb)
    try:
        result.get(timeout=1)           # raises `TimeoutError`
    except TimeoutError:
        print "Good. Got expected timeout exception."
    else:
        assert False, "Expected exception !"
    for i in result.get():
        print "Item:", i
    print "### Loop again:"
    for i in result.get():
        print "Item2:", i

    # Test imap_unordered_async()
    result = pool.imap_unordered_async(work, xrange(10, 3, -1), callback=cb)
    try:
        print result.get(timeout=1)           # raises `TimeoutError`
    except TimeoutError:
        print "Good. Got expected timeout exception."
    else:
        assert False, "Expected exception !"
    for i in result.get():
        print "Item1:", i
    for i in result.get():
        print "Item2:", i
    r = result.get()
    for i in r:
        print "Item3:", i
    for i in r:
        print "Item4:", i
    for i in r:
        print "Item5:", i

    #
    # The case for the exceptions
    #

    # Exceptions in imap_unordered_async()
    result = pool.imap_unordered_async(work, xrange(2, -10, -1), callback=cb)
    time.sleep(3)
    try:
        for i in result.get():
            print "Got item:", i
    except IOError:
        print "Good. Got expected exception:"
        traceback.print_exc()

    # Exceptions in imap_async()
    result = pool.imap_async(work, xrange(2, -10, -1), callback=cb)
    time.sleep(3)
    try:
        for i in result.get():
            print "Got item:", i
    except IOError:
        print "Good. Got expected exception:"
        traceback.print_exc()

    # Stop the test: need to stop the pool !!!
    pool.terminate()
    print "End of tests"

if __name__ == "__main__":
    _test()

Be careful to call Pool::terminate() explicitly because the worker threads are not "daemon" threads; otherwise your program will hang forever instead of terminating. This is the main difference in usage wrt multiprocessing::Pool (which registers most of its objects for deletion by atexit).

When we say that an ApplyResult becomes "ready" above, it means that its associated Job either completed without exception (in which case the ApplyResult is also "successful"), or with an exception. In the first case, calling get() on the ApplyResult object will return the value of the result. In the second, it will raise the exception and the backtrace goes up to the root cause in the worker thread.

This recipe has been tested with python 2.5 and 2.6b3.

For more information about the general API, refer to http://docs.python.org/dev/library/multiprocessing.html#module-multiprocessing.pool for example.

4 comments

Andres Riancho 11 years, 9 months ago  # | flag

Please note that there is a bug in this implementation, passing an empty iterator to map (haven't tested others but they might have the issue as well) will block for ever. How to reproduce:

    p = Pool(10)
    p.map(lambda x:x, [])
Oren Tirosh 11 years, 6 months ago  # | flag

It looks like this particular battery is already included in the standard library:

from multiprocessing.pool import ThreadPool
Anton Malakhov 8 years, 2 months ago  # | flag

David, thank you for this example. I'm having troubles with reusing this code due to legal issue: it does not include MIT license text. Could you please update the code along with the 'Copyright (c) <year> <copyright holders>' comment added or apply the new license for this code? Thanks

david decotigny (author) 8 years, 2 months ago  # | flag

Andres: should be fixed in latest snapshot.

Anton: likewise

Oren: thanks. I encourage users of python >= 2.7 to use multiprocessing.pool.ThreadPool, indeed.