python 多处理/线程清理

2024-02-19

我有一个 python 工具,基本上有这样的设置:

main process (P1) -> spawns a process (P2) that starts a tcp connection
                  -> spawns a thread (T1) that starts a loop to receive 
                     messages that are sent from P2 to P1 via a Queue (Q1)

server process (P2) -> spawns two threads (T2 and T3) that start loops to
                       receive messages that are sent from P1 to P2 via Queues (Q2 and Q3)

我遇到的问题是,当我停止程序(使用 Ctrl+C)时,它不会退出。服务器进程已结束,但主进程只是挂在那里,我必须杀死它。

线程循环函数看起来都一样:

def _loop(self):
    while self.running:
        res = self.Q1.get()
        if res is None:
            break
        self._handle_msg(res)

所有线程都作为守护进程启动:

t = Thread(target=self._loop)
t.setDaemon(True)
t.start()

在我的主进程中,我使用 atexit 来执行清理任务:

atexit.register(self.on_exit)

这些清理任务基本上如下:

1) set self.running在 P1 至False并发送None到 Q1,以便线程 T1 应该完成

self.running = False
self.Q1.put(None)

2)通过Q2向P2发送消息,通知该进程即将结束

self.Q2.put("stop")

3)在P2中,对“停止”消息做出反应并执行我们在P1中所做的操作

self.running = False
self.Q2.put(None)
self.Q3.put(None)

就是这样,根据我的理解,这应该能让一切顺利关闭,但事实并非如此。

P1的主要代码还包含以下无限循环,否则程序将提前结束:

while running:
    sleep(1)

也许这与问题有关,但我不明白为什么会这样。

那么我做错了什么?我的设置是否存在重大设计缺陷?我是否忘记关闭某些东西?

EDIT

好的,我修改了代码并设法使其在大多数情况下都能正确关闭。不幸的是,从现在开始,它仍然卡住了。

我设法编写了我的代码的一个小型工作示例。为了演示会发生什么,您需要简单地启动脚本,然后使用Ctrl + C阻止它。如果您按,现在通常会出现此问题Ctrl + C启动该工具后尽快。

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import signal
import sys
import logging
from multiprocessing import Process, Queue
from threading import Thread
from time import sleep


logger = logging.getLogger("mepy-client")


class SocketClientProtocol(object):

    def __init__(self, q_in, q_out, q_binary):
        self.q_in = q_in
        self.q_out = q_out
        self.q_binary = q_binary
        self.running = True
        t = Thread(target=self._loop)
        #t.setDaemon(True)
        t.start()
        t = Thread(target=self._loop_binary)
        #t.setDaemon(True)
        t.start()

    def _loop(self):
        print "start of loop 2"
        while self.running:
            res = self.q_in.get()
            if res is None:
                break
            self._handle_msg(res)
        print "end of loop 2"

    def _loop_binary(self):
        print "start of loop 3"
        while self.running:
            res = self.q_binary.get()
            if res is None:
                break
            self._handle_binary(res)
        print "end of loop 3"

    def _handle_msg(self, msg):
        msg_type = msg[0]
        if msg_type == "stop2":
            print "STOP RECEIVED"
            self.running = False
            self.q_in.put(None)
            self.q_binary.put(None)

    def _put_msg(self, msg):
        self.q_out.put(msg)

    def _handle_binary(self, data):
        pass

    def handle_element(self):
        self._put_msg(["something"])

def run_twisted(q_in, q_out, q_binary):
    s = SocketClientProtocol(q_in, q_out, q_binary)
    while s.running:
        sleep(2)
        s.handle_element()


class MediatorSender(object):

    def __init__(self):
        self.q_in = None
        self.q_out = None
        self.q_binary = None
        self.p = None
        self.running = False

    def start(self):
        if self.running:
            return
        self.running = True
        self.q_in = Queue()
        self.q_out = Queue()
        self.q_binary = Queue()
        print "!!!!START"
        self.p = Process(target=run_twisted, args=(self.q_in, self.q_out, self.q_binary))
        self.p.start()
        t = Thread(target=self._loop)
        #t.setDaemon(True)
        t.start()

    def stop(self):
        print "!!!!STOP"
        if not self.running:
            return
        print "STOP2"
        self.running = False
        self.q_out.put(None)
        self.q_in.put(["stop2"])
        #self.q_in.put(None)
        #self.q_binary.put(None)

        try:
            if self.p and self.p.is_alive():
                self.p.terminate()
        except:
            pass

    def _loop(self):
        print "start of loop 1"
        while self.running:
            res = self.q_out.get()
            if res is None:
                break
            self._handle_msg(res)
        print "end of loop 1"

    def _handle_msg(self, msg):
        self._put_msg(msg)

    def _put_msg(self, msg):
        self.q_in.put(msg)

    def _put_binary(self, msg):
        self.q_binary.put(msg)

    def send_chunk(self, chunk):
        self._put_binary(chunk)

running = True
def signal_handler(signal, frame):
    global running
    if running:
        running = False
        ms.stop()
    else:
        sys.exit(0)

if __name__ == "__main__":
    signal.signal(signal.SIGINT, signal_handler)
    ms = MediatorSender()
    ms.start()
    for i in range(100):
        ms.send_chunk("some chunk of data")
    while running:
        sleep(1)

我认为你正在腐蚀你的multiprocessing.Queue通过致电p.terminate()在子进程上。文档对此有警告:

Warning:如果当关联进程正在使用此方法时使用 管道或队列,那么管道或队列很容易损坏并且 可能会被其他进程变得无法使用。类似地,如果该过程有 获取锁或信号量等,然后终止它可能会 导致其他进程死锁。

在某些情况下,它看起来像p在你之前终止MediatorSender._loop方法可以消耗您加载到其中的哨兵,以让它知道它应该退出。

此外,您正在安装一个信号处理程序,该处理程序预计仅在主进程中工作,但 SIGINT 实际上由父进程接收and子进程,这意味着signal_handler在两个进程中都被调用,可能会导致ms.stop由于处理设置方式的竞争条件,被调用两次ms.running to False

我建议仅利用两个进程都接收 SIGINT,并同时拥有父进程和子进程的句柄KeyboardInterrupt直接地。这样,每个人就可以干净地关闭自己,而不是让父母终止孩子。以下代码演示了这一点,并且在我的测试中从未挂起。我在几个地方简化了您的代码,但功能上是完全相同的:

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import logging
from multiprocessing import Process, Queue
from threading import Thread
from time import sleep

logger = logging.getLogger("mepy-client")

class SocketClientProtocol(object):

    def __init__(self, q_in, q_out, q_binary):
        self.q_in = q_in
        self.q_out = q_out
        self.q_binary = q_binary
        t = Thread(target=self._loop)
        t.start()
        t = Thread(target=self._loop_binary)
        t.start()

    def _loop(self):
        print("start of loop 2")
        for res in iter(self.q_in.get, None):
            self._handle_msg(res)
        print("end of loop 2")

    def _loop_binary(self):
        print("start of loop 3")
        for res in iter(self.q_binary.get, None):
            self._handle_binary(res)
        print("end of loop 3")

    def _handle_msg(self, msg):
        msg_type = msg[0]
        if msg_type == "stop2":
            self.q_in.put(None)
            self.q_binary.put(None)

    def _put_msg(self, msg):
        self.q_out.put(msg)

    def stop(self):
        print("STOP RECEIVED")
        self.q_in.put(None)
        self.q_binary.put(None)

    def _handle_binary(self, data):
        pass

    def handle_element(self):
        self._put_msg(["something"])

def run_twisted(q_in, q_out, q_binary):
    s = SocketClientProtocol(q_in, q_out, q_binary)
    try:
        while True:
            sleep(2)
            s.handle_element()
    except KeyboardInterrupt:
        s.stop()

class MediatorSender(object):

    def __init__(self):
        self.q_in = None
        self.q_out = None
        self.q_binary = None
        self.p = None
        self.running = False

    def start(self):
        if self.running:
            return
        self.running = True
        self.q_in = Queue()
        self.q_out = Queue()
        self.q_binary = Queue()
        print("!!!!START")
        self.p = Process(target=run_twisted, 
                         args=(self.q_in, self.q_out, self.q_binary))
        self.p.start()
        self.loop = Thread(target=self._loop)
        self.loop.start()

    def stop(self):
        print("!!!!STOP")
        if not self.running:
            return
        print("STOP2")
        self.running = False
        self.q_out.put(None)

    def _loop(self):
        print("start of loop 1")
        for res in iter(self.q_out.get, None):
            self._handle_msg(res)
        print("end of loop 1")

    def _handle_msg(self, msg):
        self._put_msg(msg)

    def _put_msg(self, msg):
        self.q_in.put(msg)

    def _put_binary(self, msg):
        self.q_binary.put(msg)

    def send_chunk(self, chunk):
        self._put_binary(chunk)

if __name__ == "__main__":
    ms = MediatorSender()
    try:
        ms.start()
        for i in range(100):
            ms.send_chunk("some chunk of data")
        # You actually have to join w/ a timeout in a loop on 
        # Python 2.7. If you just call join(), SIGINT won't be 
        # received by the main process, and the program will 
        # hang. This is a bug, and is fixed in Python 3.x.
        while True:
            ms.loop.join()  
    except KeyboardInterrupt:
        ms.stop()

Edit:

如果您更喜欢使用信号处理程序而不是捕获KeyboardInterrupt,您只需要确保子进程使用自己的信号处理程序,而不是继承父进程的:

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import signal
import logging
from functools import partial
from multiprocessing import Process, Queue
from threading import Thread
from time import sleep

logger = logging.getLogger("mepy-client")

class SocketClientProtocol(object):

    def __init__(self, q_in, q_out, q_binary):
        self.q_in = q_in
        self.q_out = q_out
        self.q_binary = q_binary
        self.running = True
        t = Thread(target=self._loop)
        t.start()
        t = Thread(target=self._loop_binary)
        t.start()

    def _loop(self):
        print("start of loop 2")
        for res in iter(self.q_in.get, None):
            self._handle_msg(res)
        print("end of loop 2")

    def _loop_binary(self):
        print("start of loop 3")
        for res in iter(self.q_binary.get, None):
            self._handle_binary(res)
        print("end of loop 3")

    def _handle_msg(self, msg):
        msg_type = msg[0]
        if msg_type == "stop2":
            self.q_in.put(None)
            self.q_binary.put(None)

    def _put_msg(self, msg):
        self.q_out.put(msg)

    def stop(self):
        print("STOP RECEIVED")
        self.running = False
        self.q_in.put(None)
        self.q_binary.put(None)

    def _handle_binary(self, data):
        pass

    def handle_element(self):
        self._put_msg(["something"])

def run_twisted(q_in, q_out, q_binary):
    s = SocketClientProtocol(q_in, q_out, q_binary)
    signal.signal(signal.SIGINT, partial(signal_handler_child, s))
    while s.running:
        sleep(2)
        s.handle_element()

class MediatorSender(object):

    def __init__(self):
        self.q_in = None
        self.q_out = None
        self.q_binary = None
        self.p = None
        self.running = False

    def start(self):
        if self.running:
            return
        self.running = True
        self.q_in = Queue()
        self.q_out = Queue()
        self.q_binary = Queue()
        print("!!!!START")
        self.p = Process(target=run_twisted, 
                         args=(self.q_in, self.q_out, self.q_binary))
        self.p.start()
        self.loop = Thread(target=self._loop)
        self.loop.start()

    def stop(self):
        print("!!!!STOP")
        if not self.running:
            return
        print("STOP2")
        self.running = False
        self.q_out.put(None)

    def _loop(self):
        print("start of loop 1")
        for res in iter(self.q_out.get, None):
            self._handle_msg(res)
        print("end of loop 1")

    def _handle_msg(self, msg):
        self._put_msg(msg)

    def _put_msg(self, msg):
        self.q_in.put(msg)

    def _put_binary(self, msg):
        self.q_binary.put(msg)

    def send_chunk(self, chunk):
        self._put_binary(chunk)

def signal_handler_main(ms, *args):
    ms.stop()

def signal_handler_child(s, *args):
    s.stop()

if __name__ == "__main__":
    ms = MediatorSender()
    signal.signal(signal.SIGINT, partial(signal_handler_main, ms))
    ms.start()
    for i in range(100):
        ms.send_chunk("some chunk of data")
    while ms.loop.is_alive():
        ms.loop.join(9999999)  
    print('done main')
本文内容由网友自发贡献,版权归原作者所有,本站不承担相应法律责任。如您发现有涉嫌抄袭侵权的内容,请联系:hwhale#tublm.com(使用前将#替换为@)

python 多处理/线程清理 的相关文章

随机推荐