| 1 | n/a | # Copyright 2009 Brian Quinlan. All Rights Reserved. |
|---|
| 2 | n/a | # Licensed to PSF under a Contributor Agreement. |
|---|
| 3 | n/a | |
|---|
| 4 | n/a | """Implements ThreadPoolExecutor.""" |
|---|
| 5 | n/a | |
|---|
| 6 | n/a | __author__ = 'Brian Quinlan (brian@sweetapp.com)' |
|---|
| 7 | n/a | |
|---|
| 8 | n/a | import atexit |
|---|
| 9 | n/a | from concurrent.futures import _base |
|---|
| 10 | n/a | import queue |
|---|
| 11 | n/a | import threading |
|---|
| 12 | n/a | import weakref |
|---|
| 13 | n/a | import os |
|---|
| 14 | n/a | |
|---|
| 15 | n/a | # Workers are created as daemon threads. This is done to allow the interpreter |
|---|
| 16 | n/a | # to exit when there are still idle threads in a ThreadPoolExecutor's thread |
|---|
| 17 | n/a | # pool (i.e. shutdown() was not called). However, allowing workers to die with |
|---|
| 18 | n/a | # the interpreter has two undesirable properties: |
|---|
| 19 | n/a | # - The workers would still be running during interpreter shutdown, |
|---|
| 20 | n/a | # meaning that they would fail in unpredictable ways. |
|---|
| 21 | n/a | # - The workers could be killed while evaluating a work item, which could |
|---|
| 22 | n/a | # be bad if the callable being evaluated has external side-effects e.g. |
|---|
| 23 | n/a | # writing to a file. |
|---|
| 24 | n/a | # |
|---|
| 25 | n/a | # To work around this problem, an exit handler is installed which tells the |
|---|
| 26 | n/a | # workers to exit when their work queues are empty and then waits until the |
|---|
| 27 | n/a | # threads finish. |
|---|
| 28 | n/a | |
|---|
| 29 | n/a | _threads_queues = weakref.WeakKeyDictionary() |
|---|
| 30 | n/a | _shutdown = False |
|---|
| 31 | n/a | |
|---|
| 32 | n/a | def _python_exit(): |
|---|
| 33 | n/a | global _shutdown |
|---|
| 34 | n/a | _shutdown = True |
|---|
| 35 | n/a | items = list(_threads_queues.items()) |
|---|
| 36 | n/a | for t, q in items: |
|---|
| 37 | n/a | q.put(None) |
|---|
| 38 | n/a | for t, q in items: |
|---|
| 39 | n/a | t.join() |
|---|
| 40 | n/a | |
|---|
| 41 | n/a | atexit.register(_python_exit) |
|---|
| 42 | n/a | |
|---|
| 43 | n/a | class _WorkItem(object): |
|---|
| 44 | n/a | def __init__(self, future, fn, args, kwargs): |
|---|
| 45 | n/a | self.future = future |
|---|
| 46 | n/a | self.fn = fn |
|---|
| 47 | n/a | self.args = args |
|---|
| 48 | n/a | self.kwargs = kwargs |
|---|
| 49 | n/a | |
|---|
| 50 | n/a | def run(self): |
|---|
| 51 | n/a | if not self.future.set_running_or_notify_cancel(): |
|---|
| 52 | n/a | return |
|---|
| 53 | n/a | |
|---|
| 54 | n/a | try: |
|---|
| 55 | n/a | result = self.fn(*self.args, **self.kwargs) |
|---|
| 56 | n/a | except BaseException as e: |
|---|
| 57 | n/a | self.future.set_exception(e) |
|---|
| 58 | n/a | else: |
|---|
| 59 | n/a | self.future.set_result(result) |
|---|
| 60 | n/a | |
|---|
| 61 | n/a | def _worker(executor_reference, work_queue): |
|---|
| 62 | n/a | try: |
|---|
| 63 | n/a | while True: |
|---|
| 64 | n/a | work_item = work_queue.get(block=True) |
|---|
| 65 | n/a | if work_item is not None: |
|---|
| 66 | n/a | work_item.run() |
|---|
| 67 | n/a | # Delete references to object. See issue16284 |
|---|
| 68 | n/a | del work_item |
|---|
| 69 | n/a | continue |
|---|
| 70 | n/a | executor = executor_reference() |
|---|
| 71 | n/a | # Exit if: |
|---|
| 72 | n/a | # - The interpreter is shutting down OR |
|---|
| 73 | n/a | # - The executor that owns the worker has been collected OR |
|---|
| 74 | n/a | # - The executor that owns the worker has been shutdown. |
|---|
| 75 | n/a | if _shutdown or executor is None or executor._shutdown: |
|---|
| 76 | n/a | # Notice other workers |
|---|
| 77 | n/a | work_queue.put(None) |
|---|
| 78 | n/a | return |
|---|
| 79 | n/a | del executor |
|---|
| 80 | n/a | except BaseException: |
|---|
| 81 | n/a | _base.LOGGER.critical('Exception in worker', exc_info=True) |
|---|
| 82 | n/a | |
|---|
| 83 | n/a | class ThreadPoolExecutor(_base.Executor): |
|---|
| 84 | n/a | def __init__(self, max_workers=None, thread_name_prefix=''): |
|---|
| 85 | n/a | """Initializes a new ThreadPoolExecutor instance. |
|---|
| 86 | n/a | |
|---|
| 87 | n/a | Args: |
|---|
| 88 | n/a | max_workers: The maximum number of threads that can be used to |
|---|
| 89 | n/a | execute the given calls. |
|---|
| 90 | n/a | thread_name_prefix: An optional name prefix to give our threads. |
|---|
| 91 | n/a | """ |
|---|
| 92 | n/a | if max_workers is None: |
|---|
| 93 | n/a | # Use this number because ThreadPoolExecutor is often |
|---|
| 94 | n/a | # used to overlap I/O instead of CPU work. |
|---|
| 95 | n/a | max_workers = (os.cpu_count() or 1) * 5 |
|---|
| 96 | n/a | if max_workers <= 0: |
|---|
| 97 | n/a | raise ValueError("max_workers must be greater than 0") |
|---|
| 98 | n/a | |
|---|
| 99 | n/a | self._max_workers = max_workers |
|---|
| 100 | n/a | self._work_queue = queue.Queue() |
|---|
| 101 | n/a | self._threads = set() |
|---|
| 102 | n/a | self._shutdown = False |
|---|
| 103 | n/a | self._shutdown_lock = threading.Lock() |
|---|
| 104 | n/a | self._thread_name_prefix = thread_name_prefix |
|---|
| 105 | n/a | |
|---|
| 106 | n/a | def submit(self, fn, *args, **kwargs): |
|---|
| 107 | n/a | with self._shutdown_lock: |
|---|
| 108 | n/a | if self._shutdown: |
|---|
| 109 | n/a | raise RuntimeError('cannot schedule new futures after shutdown') |
|---|
| 110 | n/a | |
|---|
| 111 | n/a | f = _base.Future() |
|---|
| 112 | n/a | w = _WorkItem(f, fn, args, kwargs) |
|---|
| 113 | n/a | |
|---|
| 114 | n/a | self._work_queue.put(w) |
|---|
| 115 | n/a | self._adjust_thread_count() |
|---|
| 116 | n/a | return f |
|---|
| 117 | n/a | submit.__doc__ = _base.Executor.submit.__doc__ |
|---|
| 118 | n/a | |
|---|
| 119 | n/a | def _adjust_thread_count(self): |
|---|
| 120 | n/a | # When the executor gets lost, the weakref callback will wake up |
|---|
| 121 | n/a | # the worker threads. |
|---|
| 122 | n/a | def weakref_cb(_, q=self._work_queue): |
|---|
| 123 | n/a | q.put(None) |
|---|
| 124 | n/a | # TODO(bquinlan): Should avoid creating new threads if there are more |
|---|
| 125 | n/a | # idle threads than items in the work queue. |
|---|
| 126 | n/a | num_threads = len(self._threads) |
|---|
| 127 | n/a | if num_threads < self._max_workers: |
|---|
| 128 | n/a | thread_name = '%s_%d' % (self._thread_name_prefix or self, |
|---|
| 129 | n/a | num_threads) |
|---|
| 130 | n/a | t = threading.Thread(name=thread_name, target=_worker, |
|---|
| 131 | n/a | args=(weakref.ref(self, weakref_cb), |
|---|
| 132 | n/a | self._work_queue)) |
|---|
| 133 | n/a | t.daemon = True |
|---|
| 134 | n/a | t.start() |
|---|
| 135 | n/a | self._threads.add(t) |
|---|
| 136 | n/a | _threads_queues[t] = self._work_queue |
|---|
| 137 | n/a | |
|---|
| 138 | n/a | def shutdown(self, wait=True): |
|---|
| 139 | n/a | with self._shutdown_lock: |
|---|
| 140 | n/a | self._shutdown = True |
|---|
| 141 | n/a | self._work_queue.put(None) |
|---|
| 142 | n/a | if wait: |
|---|
| 143 | n/a | for t in self._threads: |
|---|
| 144 | n/a | t.join() |
|---|
| 145 | n/a | shutdown.__doc__ = _base.Executor.shutdown.__doc__ |
|---|