如何在Python中为函数添加超时

Question

过去已经进行了许多尝试以在Python中添加超时功能,以便在指定的时间限制到期时,等待代码可以继续.不幸的是,以前的配方要么允许运行功能继续运行并消耗资源,要么使用特定于平台的线程终止方法来终止该功能.这个wiki的目的是为这个问题开发一个跨平台的答案,许多程序员不得不为各种编程项目解决这个问题.

#! /usr/bin/env python
"""Provide way to add timeout specifications to arbitrary functions.

There are many ways to add a timeout to a function, but no solution
is both cross-platform and capable of terminating the procedure. This
module use the multiprocessing module to solve both of those problems."""

################################################################################

__author__ = 'Stephen "Zero" Chappell <Noctis.Skytower@gmail.com>'
__date__ = '11 February 2010'
__version__ = '$Revision: 3 $'

################################################################################

import inspect
import sys
import time
import multiprocessing

################################################################################

def add_timeout(function, limit=60):
    """Add a timeout parameter to a function and return it.

    It is illegal to pass anything other than a function as the first
    parameter. If the limit is not given, it gets a default value equal
    to one minute. The function is wrapped and returned to the caller."""
    assert inspect.isfunction(function)
    if limit <= 0:
        raise ValueError()
    return _Timeout(function, limit)

class NotReadyError(Exception): pass

################################################################################

def _target(queue, function, *args, **kwargs):
    """Run a function with arguments and return output via a queue.

    This is a helper function for the Process created in _Timeout. It runs
    the function with positional arguments and keyword arguments and then
    returns the function's output by way of a queue. If an exception gets
    raised, it is returned to _Timeout to be raised by the value property."""
    try:
        queue.put((True, function(*args, **kwargs)))
    except:
        queue.put((False, sys.exc_info()[1]))

class _Timeout:

    """Wrap a function and add a timeout (limit) attribute to it.

    Instances of this class are automatically generated by the add_timeout
    function defined above. Wrapping a function allows asynchronous calls
    to be made and termination of execution after a timeout has passed."""

    def __init__(self, function, limit):
        """Initialize instance in preparation for being called."""
        self.__limit = limit
        self.__function = function
        self.__timeout = time.clock()
        self.__process = multiprocessing.Process()
        self.__queue = multiprocessing.Queue()

    def __call__(self, *args, **kwargs):
        """Execute the embedded function object asynchronously.

        The function given to the constructor is transparently called and
        requires that "ready" be intermittently polled. If and when it is
        True, the "value" property may then be checked for returned data."""
        self.cancel()
        self.__queue = multiprocessing.Queue(1)
        args = (self.__queue, self.__function) + args
        self.__process = multiprocessing.Process(target=_target,
                                                 args=args,
                                                 kwargs=kwargs)
        self.__process.daemon = True
        self.__process.start()
        self.__timeout = self.__limit + time.clock()

    def cancel(self):
        """Terminate any possible execution of the embedded function."""
        if self.__process.is_alive():
            self.__process.terminate()

    @property
    def ready(self):
        """Read-only property indicating status of "value" property."""
        if self.__queue.full():
            return True
        elif not self.__queue.empty():
            return True
        elif self.__timeout < time.clock():
            self.cancel()
        else:
            return False

    @property
    def value(self):
        """Read-only property containing data returned from function."""
        if self.ready is True:
            flag, load = self.__queue.get()
            if flag:
                return load
            raise load
        raise NotReadyError()

    def __get_limit(self):
        return self.__limit

    def __set_limit(self, value):
        if value <= 0:
            raise ValueError()
        self.__limit = value

    limit = property(__get_limit, __set_limit,
                     doc="Property for controlling the value of the timeout.")

---

编辑:此代码是为Python 3.x编写的,并不是为类方法设计的装饰.该multiprocessing模块不是为跨过程边界修改类实例而设计的.

Answer 1

您的代码的主要问题是在一个根本不打算子类化的类中过度使用双下划线命名空间冲突.

一般来说,self.__foo代码气味应该伴随着一条评论# This is a mixin and we don't want arbitrary subclasses to have a namespace conflict.

此外,此方法的客户端API将如下所示:

def mymethod(): pass

mymethod = add_timeout(mymethod, 15)

# start the processing    
timeout_obj = mymethod()
try:
    # access the property, which is really a function call
    ret = timeout_obj.value
except TimeoutError:
    # handle a timeout here
    ret = None

这根本不是pythonic,更好的客户端api将是:

@timeout(15)
def mymethod(): pass

try:
    my_method()
except TimeoutError:
    pass

您在类中使用@property作为状态变异访问器的东西,这不是一个好主意.例如,当.value被访问两次时会发生什么？看起来它会失败,因为queue.get()会返回垃圾,因为队列已经是空的.

完全删除@property.不要在这种情况下使用它,它不适合您的用例.使呼叫块调用时,返回的值或抛出异常本身.如果您确实必须稍后访问该值,请将其设置为.get()或.value()等方法.

_target的这段代码应该重写一下:

def _target(queue, function, *args, **kwargs):
    try:
        queue.put((True, function(*args, **kwargs)))
    except:
        queue.put((False, exc_info())) # get *all* the exec info, don't do exc_info[1]

# then later:
    raise exc_info[0], exc_info[1], exc_info[2]

这样,堆栈跟踪将被正确保存并且对程序员可见.

我认为你在编写一个有用的库时已经做了一个合理的第一次破解,我喜欢使用处理模块来实现目标.

Answer 2

这是如何获得Jerub提到的装饰器语法

def timeout(limit=None):
    if limit is None:
        limit = DEFAULT_TIMEOUT
    if limit <= 0:
        raise TimeoutError() # why not ValueError here?
    def wrap(function):
        return _Timeout(function,limit)
    return wrap

@timeout(15)
def mymethod(): pass

Answer 3

这个问题是 9 年前提出的，从那时起，Python 发生了很大的变化，我的经验也是如此。在审查了标准库中的其他 API 并希望部分复制特定的 API 后，编写了以下模块来实现与问题中发布的类似目的。

异步.py

#! /usr/bin/env python3
import _thread
import abc as _abc
import collections as _collections
import enum as _enum
import math as _math
import multiprocessing as _multiprocessing
import operator as _operator
import queue as _queue
import signal as _signal
import sys as _sys
import time as _time

__all__ = (
    'Executor',
    'get_timeout',
    'set_timeout',
    'submit',
    'map_',
    'shutdown'
)


class _Base(metaclass=_abc.ABCMeta):
    __slots__ = (
        '__timeout',
    )

    @_abc.abstractmethod
    def __init__(self, timeout):
        self.timeout = _math.inf if timeout is None else timeout

    def get_timeout(self):
        return self.__timeout

    def set_timeout(self, value):
        if not isinstance(value, (float, int)):
            raise TypeError('value must be of type float or int')
        if value <= 0:
            raise ValueError('value must be greater than zero')
        self.__timeout = value

    timeout = property(get_timeout, set_timeout)


def _run_and_catch(fn, args, kwargs):
    # noinspection PyPep8,PyBroadException
    try:
        return False, fn(*args, **kwargs)
    except:
        return True, _sys.exc_info()[1]


def _run(fn, args, kwargs, queue):
    queue.put_nowait(_run_and_catch(fn, args, kwargs))


class _State(_enum.IntEnum):
    PENDING = _enum.auto()
    RUNNING = _enum.auto()
    CANCELLED = _enum.auto()
    FINISHED = _enum.auto()
    ERROR = _enum.auto()


def _run_and_catch_loop(iterable, *args, **kwargs):
    exception = None
    for fn in iterable:
        error, value = _run_and_catch(fn, args, kwargs)
        if error:
            exception = value
    if exception:
        raise exception


class _Future(_Base):
    __slots__ = (
        '__queue',
        '__process',
        '__start_time',
        '__callbacks',
        '__result',
        '__mutex'
    )

    def __init__(self, timeout, fn, args, kwargs):
        super().__init__(timeout)
        self.__queue = _multiprocessing.Queue(1)
        self.__process = _multiprocessing.Process(
            target=_run,
            args=(fn, args, kwargs, self.__queue),
            daemon=True
        )
        self.__start_time = _math.inf
        self.__callbacks = _collections.deque()
        self.__result = True, TimeoutError()
        self.__mutex = _thread.allocate_lock()

    @property
    def __state(self):
        pid, exitcode = self.__process.pid, self.__process.exitcode
        return (_State.PENDING if pid is None else
                _State.RUNNING if exitcode is None else
                _State.CANCELLED if exitcode == -_signal.SIGTERM else
                _State.FINISHED if exitcode == 0 else
                _State.ERROR)

    def __repr__(self):
        root = f'{type(self).__name__} at {id(self)} state={self.__state.name}'
        if self.__state < _State.CANCELLED:
            return f'<{root}>'
        error, value = self.__result
        suffix = f'{"raised" if error else "returned"} {type(value).__name__}'
        return f'<{root} {suffix}>'

    def __consume_callbacks(self):
        while self.__callbacks:
            yield self.__callbacks.popleft()

    def __invoke_callbacks(self):
        self.__process.join()
        _run_and_catch_loop(self.__consume_callbacks(), self)

    def cancel(self):
        self.__process.terminate()
        self.__invoke_callbacks()

    def __auto_cancel(self):
        elapsed_time = _time.perf_counter() - self.__start_time
        if elapsed_time > self.timeout:
            self.cancel()
        return elapsed_time

    def cancelled(self):
        self.__auto_cancel()
        return self.__state is _State.CANCELLED

    def running(self):
        self.__auto_cancel()
        return self.__state is _State.RUNNING

    def done(self):
        self.__auto_cancel()
        return self.__state > _State.RUNNING

    def __handle_result(self, error, value):
        self.__result = error, value
        self.__invoke_callbacks()

    def __ensure_termination(self):
        with self.__mutex:
            elapsed_time = self.__auto_cancel()
            if not self.__queue.empty():
                self.__handle_result(*self.__queue.get_nowait())
            elif self.__state < _State.CANCELLED:
                remaining_time = self.timeout - elapsed_time
                if remaining_time == _math.inf:
                    remaining_time = None
                try:
                    result = self.__queue.get(True, remaining_time)
                except _queue.Empty:
                    self.cancel()
                else:
                    self.__handle_result(*result)

    def result(self):
        self.__ensure_termination()
        error, value = self.__result
        if error:
            raise value
        return value

    def exception(self):
        self.__ensure_termination()
        error, value = self.__result
        if error:
            return value

    def add_done_callback(self, fn):
        if self.done():
            fn(self)
        else:
            self.__callbacks.append(fn)

    def _set_running_or_notify_cancel(self):
        if self.__state is _State.PENDING:
            self.__process.start()
            self.__start_time = _time.perf_counter()
        else:
            self.cancel()


class Executor(_Base):
    __slots__ = (
        '__futures',
    )

    def __init__(self, timeout=None):
        super().__init__(timeout)
        self.__futures = set()

    def submit(self, fn, *args, **kwargs):
        future = _Future(self.timeout, fn, args, kwargs)
        self.__futures.add(future)
        future.add_done_callback(self.__futures.remove)
        # noinspection PyProtectedMember
        future._set_running_or_notify_cancel()
        return future

    @staticmethod
    def __cancel_futures(iterable):
        _run_and_catch_loop(map(_operator.attrgetter('cancel'), iterable))

    def map(self, fn, *iterables):
        futures = tuple(self.submit(fn, *args) for args in zip(*iterables))

        def result_iterator():
            future_iterator = iter(futures)
            try:
                for future in future_iterator:
                    yield future.result()
            finally:
                self.__cancel_futures(future_iterator)

        return result_iterator()

    def shutdown(self):
        self.__cancel_futures(frozenset(self.__futures))

    def __enter__(self):
        return self

    def __exit__(self, exc_type, exc_val, exc_tb):
        self.shutdown()
        return False


_executor = Executor()
get_timeout = _executor.get_timeout
set_timeout = _executor.set_timeout
submit = _executor.submit
map_ = _executor.map
shutdown = _executor.shutdown
del _executor