从父进程访问Python Multiprocessing.Process子类的状态

Question

我正在创建一个简单的TCP服务器作为存根，因此我可以测试运行一个测试设备的脚本，而不必在那里安装设备。 服务器应坐在那里等待连接，然后根据收到的命令维护和更新状态变量（仅由6个整数组成的列表）。 然后，父进程（例如，单元测试类）应该能够查询状态。

服务器的接口应尽可能简单：

server = StubServer()
server.start()
'''
the client script connects with the server and
some stuff happens to change the state
'''
newState = server.getState() # newState = [93,93,93,3,3,45] for example
server.terminate()

我将Multiprocessing.Process子类化，可以做到这一点，并且我可以毫无问题地启动服务器。 第一次测试时，在getState（）方法中，我只是返回了实例变量_state，但我发现这始终只是初始状态。 经过一番挖掘，我找不到任何类似的例子。 关于子类化处理的很多东西，但不是这个特定的问题。 最终，我将下面的代码放在一起，它使用内部Queue（）来存储状态，但是对我来说这看起来很凌乱和笨拙。 有一个更好的方法吗？

import socket
from multiprocessing import Process, Queue

class StubServer(Process):

    _port = 4001
    _addr = '' # all addresses 0.0.0.0
    _sock = None
    _state = []
    _queue = None

    def __init__(self, initState=[93,93,93,93,93,93]):
        super(StubServer, self).__init__()
        self._queue = Queue()
        self._state = initState

    def run(self):
        # Put the state into the queue
        self._queue.put(self._state)
        self._sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
        self._sock.bind((self._addr, self._port))
        self._sock.listen(1)

        waitingForConnection = True
        '''
        Main loop will continue until the connection is terminated. if a connection is closed, the loop returns
        to the start and waits for a new connection. This means multiple tests can be run with the same server
        '''
        while 1:
            # Wait for a connection, or go back and wait for a new message (if a connection already exists)
            if waitingForConnection:
                waitingForConnection = False
                conn, addr = self._sock.accept()
            chunk = ''
            chunks = []
            while '\x03' not in chunk: # '\x03' is terminating character for a message
                chunk = conn.recv(8192)
                if not chunk: # Connection terminated, start the loop again and wait for a new connection
                    waitingForConnection = True
                    break
                chunks.append(chunk)
            message = ''.join(chunks)
            # Now do some stuff to parse the message, and update the state if we received a command
            if isACommand(message):
                _updateState(message)
        conn.close()
        return

    def getState(self):
        # This is called from the parent process, so return the object on the queue
        state = self._queue.get()
        # But put the state back in the queue again so it's there if this method is called again before a state update
        self._queue.put(state)
        return state

    def _updateState(self, message):
        # Do some stuff to figure out what to update then update the state
        self._state[updatedElementIndex] = updatedValue
        # Now empty the queue and put the new state in the queue
        while not self._queue.empty():
            self._queue.get()
        self._queue.put(self._state)
        return

Answer 1

顾名思义， multiprocessing使用不同的进程。 在某个时候， fork()被调用，子进程会复制父进程的内存，并且子进程将保留自己的内存，而不与父进程共享。

不幸的是，您必须使用可用的工具在进程之间共享内存，从而导致您提到的代码开销。

您可以寻找其他使用共享内存进行并行处理的方法，但是请注意，在线程/进程/节点/等之间共享内存绝非易事。

Answer 2

您可以随时将存根服务器的状态转储到文件中，并从单元测试中读取它。 这是满足测试需求的非常简单的解决方案。

您需要做的所有事情：

• 将filename作为参数传递给构造函数
• 用初始化值调用_updateState
• 重写_updateState以将状态写入filename 。 最好在文件filename附近创建一个新文件并替换它。 如果您担心原子性。

Answer 3

谢谢费利佩，我的问题主要是“像使用问题一样，有没有比使用队列更好的方法”。 经过更多研究（提示您提到共享内存）后，我发现对于这种情况，共享阵列要好得多：

import socket
from multiprocessing import Process, Array

class StubServer(Process):

    _port = 4001
    _addr = '' # all addresses 0.0.0.0
    _sock = None
    _state = None
    _queue = None

    def __init__(self, initState=[93,93,93,93,93,93]):
        super(StubServer, self).__init__()
        self._state = Array('i', initState) # Is always a 6 element array

    def run(self):
        self._sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
        self._sock.bind((self._addr, self._port))
        self._sock.listen(1)

        waitingForConnection = True
        '''
        Main loop will continue until process is terminated. if a connection is closed, the loop returns
        to the start and waits for a new connection. This means multiple tests can be run with the same server
        '''
        while 1:
            # Wait for a connection, or go back and wait for a new message (if a connection already exists)
            if waitingForConnection:
                waitingForConnection = False
                conn, addr = self._sock.accept()
            chunk = ''
            chunks = []
            while '\x03' not in chunk: # '\x03' is terminating character for a message
                chunk = conn.recv(8192)
                if not chunk: # Connection terminated, start the loop again and wait for a new connection
                    waitingForConnection = True
                    break
                chunks.append(chunk)
            message = ''.join(chunks)
            # Now do some stuff to parse the message, and update the state if we received a command
            if isACommand(message):
                _updateState(message)
        conn.close()
        return

    def getState(self):
        # Aquire the lock return the contents of the shared array
        with self._state.get_lock():
            return self._state[:6] # This is OK because we know it is always a 6 element array
        return state

    def _updateState(self, message):
        # Do some stuff to figure out what to update then..
        # Aquire the lock and update the appropriate element in the shared array
        with self._state.get_lock():
            self._state[updatedElementIndex] = updatedValue
        return

这可以使人愉悦，并且更加优雅。 谢谢你的帮助