关闭异步任务的正确方法 [英] Proper way to shutdown asyncio tasks

问题描述

我正在编写一个工具，该工具连接到X个UNIX套接字，发送命令并将输出保存在本地文件系统中。它每X秒运行一次。为了在工具接收到终止信号时执行一些清理，我将函数(SHUTDOWN)注册到Signal.SIGHUP和Signal.SIGTERM信号。此函数取消所有任务，然后关闭事件循环。

我的问题是我得到了

运行错误：事件循环在将来完成之前停止

当我发送信号.SIGTERM(终止‘pid’)时。我已经阅读了两次有关取消任务的文档，但是我没有发现我在这里做错了什么。

我还注意到一些奇怪的事情，当我发送终止信号时，程序处于休眠模式，我在日志中看到它唤醒了ull_stats()协程，您可以在日志的前两行看到这一点。

日志：

21:53:44,194 [23031] [MainThread:supervisor  ] DEBUG    **sleeping for 9.805s secs**
21:53:45,857 [23031] [MainThread:pull_stats  ] INFO     pull statistics
21:53:45,858 [23031] [MainThread:get         ] DEBUG    connecting to UNIX socket /run/haproxy/admin1.sock
21:53:45,858 [23031] [MainThread:get         ] DEBUG    connecting to UNIX socket /run/haproxy/admin4.sock
21:53:45,858 [23031] [MainThread:get         ] DEBUG    connecting to UNIX socket /run/haproxy/admin3.sock
21:53:45,858 [23031] [MainThread:get         ] DEBUG    connecting to UNIX socket /run/haproxy/admin3.sock
21:53:45,858 [23031] [MainThread:get         ] DEBUG    connecting to UNIX socket /run/haproxy/admin2.sock
21:53:45,858 [23031] [MainThread:get         ] DEBUG    connecting to UNIX socket /run/haproxy/admin2.sock
21:53:45,858 [23031] [MainThread:get         ] DEBUG    connecting to UNIX socket /run/haproxy/admin4.sock
21:53:45,859 [23031] [MainThread:get         ] DEBUG    connecting to UNIX socket /run/haproxy/admin1.sock
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     received stop signal, cancelling tasks...
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,859 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,860 [23031] [MainThread:shutdown    ] INFO     True
21:53:45,860 [23031] [MainThread:shutdown    ] INFO     stopping event loop
21:53:45,860 [23031] [MainThread:shutdown    ] INFO     bye, exiting...
Traceback (most recent call last):
  File "./pull.py", line 249, in <module>
    main()
  File "./pull.py", line 245, in main
    supervisor(loop, config)
  File "./pull.py", line 161, in supervisor
    config['pull']['socket-dir'], storage_dir, loop))
  File "/usr/lib/python3.4/asyncio/base_events.py", line 274, in run_until_complete
    raise RuntimeError('Event loop stopped before Future completed.')
RuntimeError: Event loop stopped before Future completed.

代码如下：

def shutdown(loop):
    LOGGER.info('received stop signal, cancelling tasks...')
    for task in asyncio.Task.all_tasks():
        LOGGER.info(task.cancel())
    LOGGER.info('stopping event loop')
    loop.stop()
    LOGGER.info('bye, exiting...')


def write_file(filename, data):
    try:
        with open(filename, 'w') as file_handle:
            file_handle.write(data.decode())
    except OSError as exc:
        return False
    else:
        return True


@asyncio.coroutine
def get(socket_file, cmd, storage_dir, loop):
    connect = asyncio.open_unix_connection(socket_file)
    reader, writer = yield from asyncio.wait_for(connect, 1)

    writer.write('{c}
'.format(c=cmd).encode())
    data = yield from reader.read()
    writer.close()

    filename = os.path.basename(socket_file) + '_' + cmd.split()[1]
    filename = os.path.join(storage_dir, filename)
    result = yield from loop.run_in_executor(None, write_file, filename, data)

    return result


@asyncio.coroutine
def pull_stats(socket_dir, storage_dir, loop):
    socket_files = glob.glob(socket_dir + '/*sock*')
    coroutines = [get(socket_file, cmd, storage_dir, loop)
                  for socket_file in socket_files
                  for cmd in CMDS]
    status = yield from asyncio.gather(*coroutines)

    if len(set(status)) == 1 and True in set(status):
        return True
    else:
        return False


def supervisor(loop, config):
    dst_dir = config.get('pull', 'dst-dir')
    tmp_dst_dir = config.get('pull', 'tmp-dst-dir')

    while True:
        start_time = int(time.time())
        storage_dir = os.path.join(tmp_dst_dir, str(start_time))

        try:
            os.makedirs(storage_dir)
        except OSError as exc:
            msg = "failed to create directory {d}:{e}".format(d=storage_dir,
                                                              e=exc)
            LOGGER.critical(msg)

        # Launch all connections.
        result = loop.run_until_complete(pull_stats(
            config['pull']['socket-dir'], storage_dir, loop))

        if result:
            try:
                shutil.move(storage_dir, dst_dir)
            except OSError as exc:
                LOGGER.critical("failed to move %s to %s: %s", storage_dir,
                                dst_dir, exc)
                break
            else:
                LOGGER.info('statistics are saved in %s', os.path.join(
                    dst_dir, os.path.basename(storage_dir)))
        else:
            LOGGER.critical('failed to pull stats')
            shutil.rmtree(storage_dir)

        sleep = config.getint('pull', 'pull-interval') - (time.time() -
                                                          start_time)
        if 0 < sleep < config.getint('pull', 'pull-interval'):
            time.sleep(sleep)
    loop.close()
    sys.exit(1)


def main():
    args = docopt(__doc__, version=VERSION)
    config = ConfigParser(interpolation=ExtendedInterpolation())
    config.read_dict(copy.copy(DEFAULT_OPTIONS))
    config.read(args['--file'])

    loop = asyncio.get_event_loop()

    loop.add_signal_handler(signal.SIGHUP, functools.partial(shutdown, loop))
    loop.add_signal_handler(signal.SIGTERM, functools.partial(shutdown, loop))

    num_level = getattr(logging, config.get('pull', 'loglevel').upper(), None)
    LOGGER.setLevel(num_level)

    supervisor(loop, config)

# This is the standard boilerplate that calls the main() function.
if __name__ == '__main__':
    main()

推荐答案

取消不是立即进行的，需要运行IOLOOP才能解决异常CancelledError。将ioloop.stop从关闭状态中移除并处理管理程序中的异常，以使其正常工作。下面是简化的示例。

重要的是，但是您可以取消Task，它只停止观察/等待结束/结果，循环不会为它处理更多事件。但不会停止下面的请求/管道。

简化示例：

import asyncio
import functools
import logging
import signal
import sys
from concurrent.futures import CancelledError


def shutdown(loop):
    logging.info('received stop signal, cancelling tasks...')
    for task in asyncio.Task.all_tasks():
        task.cancel()
    logging.info('bye, exiting in a minute...')    


@asyncio.coroutine
def get(i):
    logging.info('sleep for %d', i)
    yield from asyncio.sleep(i)    


@asyncio.coroutine
def pull_stats():
    coroutines = [get(i) for i in range(10,20)]
    status = yield from asyncio.gather(*coroutines)


def supervisor(loop):
    try:
        while True:
            result = loop.run_until_complete(pull_stats())
    except CancelledError:
        logging.info('CancelledError')
    loop.close()
    sys.exit(1)


def main():
    logging.getLogger().setLevel(logging.INFO)
    loop = asyncio.get_event_loop()
    loop.add_signal_handler(signal.SIGHUP, functools.partial(shutdown, loop))
    loop.add_signal_handler(signal.SIGTERM, functools.partial(shutdown, loop))
    supervisor(loop)


if __name__ == '__main__':
    main()

请注意，如果仅取消gather's将来，则所有子项也将设置为已取消。

和睡眠

任何信号或中断的接收都会导致程序恢复执行。因此，当进程接收SIGTERM并设置了HANDLER时，Python允许您处理它，以恢复该线程并调用SIGHANDLER。由于Iloop的实现和信号处理，唤醒后继续运行。

这篇关于关闭异步任务的正确方法的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！