bas*_*kum 5 python multithreading multiprocessing
我有一个python工具,基本上具有这种设置:
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)
Run Code Online (Sandbox Code Playgroud)
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,但主要进程仅挂在那里,我必须将其杀死。
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)
Run Code Online (Sandbox Code Playgroud)
所有线程均作为守护程序启动:
t = Thread(target=self._loop)
t.setDaemon(True)
t.start()
Run Code Online (Sandbox Code Playgroud)
在我的主要过程中,我使用atexit来执行清理任务:
atexit.register(self.on_exit)
Run Code Online (Sandbox Code Playgroud)
这些清理任务实质上是以下这些:
1)self.running在P1中设置False并发None送到Q1,以便线程T1应该完成
self.running = False
self.Q1.put(None)
Run Code Online (Sandbox Code Playgroud)
2)通过Q2向P2发送消息,以通知该过程即将结束
self.Q2.put("stop")
Run Code Online (Sandbox Code Playgroud)
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)
Run Code Online (Sandbox Code Playgroud)
就是这样,据我所知,应该可以使一切正常关闭,但事实并非如此。
P1的主要代码还包含以下无限循环,因为否则程序将过早结束:
while running:
sleep(1)
Run Code Online (Sandbox Code Playgroud)
也许与问题有关,但我不知道为什么会这样。
那我做错了什么?我的安装程序是否存在重大设计缺陷?我忘了关闭某些东西吗?
编辑
好的,我修改了我的代码并设法使它在大多数情况下正确关闭了。不幸的是,从那以后,它仍然卡住了。
我设法写了一个小的工作示例代码。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停止脚本。如果您Ctrl + C在启动工具后尽快按一下,通常看起来现在出现问题。
#!/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)
Run Code Online (Sandbox Code Playgroud)
我认为你multiprocessing.Queue通过调用p.terminate()子进程来破坏你的进程。文档对此有警告:
警告:如果在关联进程正在使用管道或队列时使用此方法,则管道或队列可能会损坏并且可能无法被其他进程使用。类似地,如果进程已获取锁或信号量等,则终止它可能会导致其他进程死锁。
在某些情况下,它看起来像是p在您之前终止的MediatorSender._loop在您的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之前
此外,您正在安装一个预计仅在主进程中工作的信号处理程序,但 SIGINT 实际上是由父进程和子进程同时接收的,这意味着signal_handler在两个进程中都会被调用,可能会导致ms.stop被调用两次,因为以您处理设置ms.running的方式进行竞争条件False
我建议只利用两个进程都接收 SIGINT,并KeyboardInterrupt直接拥有父进程和子进程的句柄。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()
Run Code Online (Sandbox Code Playgroud)
编辑:
如果您更喜欢使用信号处理程序而不是 catch KeyboardInterrupt,则只需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')
Run Code Online (Sandbox Code Playgroud)