Python multiprocessing.Queue修改对象

Question

我有一个应用程序,它实现了像Python中的责任链.有一个进程通过multiprocessing.Queue()将对象传递给其他进程,然后对该对象执行操作.对于被跟踪的对象的最后修改时间也很重要,因此只有在修改对象时才能执行操作.

我遇到的问题是,从队列中提取后,对象中的_modified属性似乎会随机变化.但是,_mtime属性始终是正确的.下面的示例将运行并(有意)随机修改DummyObject,然后将其放在每个处理程序进程的Queue上.然后,每个处理程序将打印它们在对象中收到的_modified和_mtime值.我希望_modified值在command_func和handler函数中都是相同的,但通常情况并非如此.如果我从DummyObject中删除Object_w_mtime继承,那么我看不到发送和接收对象的任何差异.

我对python比较陌生.据我所知,应该发生的事情是每次将一个对象放在一个队列上,它被腌制,然后通过一个管道发送到接收进程,该进程将对象取消.那是对的吗？当对象被pickle/unpickled时,有没有任何方法可以搞砸对象继承？

我在Ubuntu 11.10上用Python 2.7.2和2.6.7以及Ubuntu 11.04上的python 2.7.1测试了这个.有时你必须让它运行一分钟左右才能看到行为,因为它似乎是随机的.

在这里抓住稻草,提前谢谢.

import multiprocessing
import time
import traceback
import os
import random

class Object_w_mtime(object):
    '''
    Parent object that tracks the last time an attribute was modified
    '''
    def __setattr__(self,a_name,a_value):
        if ((a_name not in ('_mtime','_modified')) and
            (a_value != getattr(self,a_name,None))
        ):
            object.__setattr__(self, '_modified', True)
            object.__setattr__(self, '_mtime', time.time())
        object.__setattr__(self, a_name, a_value)
        return True
    #END def

    def reset(self):
        self._modified = False
#END class

class DummyObject(Object_w_mtime):
    def __init__(self):
        self.value = 10

def handler(in_queue = None, handler_id = None):
    print 'PID:' + str(os.getpid()) + ':handler{0}:<RUN>'.format(handler_id)
    while True:
        try:
            obj = in_queue.get(True,61)
            print 'handler{} - _modified'.format(handler_id), obj._modified, ' \t_mtime', obj._mtime
        except multiprocessing.queues.Empty:
            break
        except KeyboardInterrupt:
            break
        except Exception as e:
            print traceback.format_exc()
            break
    return True
#END def

def command_func(next_links = None):
    print 'PID:' + str(os.getpid()) + ':command_func:<RUN>'
    obj = DummyObject()
    while True:
        try:
            # randomly assign a different value to test with a modified and unmodified object
            obj.value = random.randint(0,1)
            print '**************** obj.value = {0} ***************'.format(obj.value)
            print 'command_ - _modified', obj._modified, ' \t_mtime', obj._mtime
            for each in next_links:
                each.put(obj,False)
        except multiprocessing.queues.Empty:
            break
        except KeyboardInterrupt:
            break
        except Exception as e:
            print e
            print traceback.format_exc()
            break
        obj.reset()
        time.sleep(3)
    return True
#END def


if __name__ == '__main__':
    handler_queues = list()
    handler_processes = list()
    # Create a queue and process object for each command handler
    for handler_id in range(1,4):
        queue = multiprocessing.Queue()
        process = multiprocessing.Process(target=handler, args=(queue, handler_id))
        handler_queues.append(queue)
        handler_processes.append(process)

    try:
        # spawn handler processes
        for process in handler_processes:
            process.start()
        # Start sending commands to handlers
        command_func(handler_queues)

    # exit on keyboard interrupt
    except KeyboardInterrupt:
        for process in handler_processes:
            process.join()
    except Exception:
        traceback.print_exc()

Answer 1

简而言之,您obj在将其放入队列后进行修改.

查看http://svn.python.org/view/python/trunk/Lib/multiprocessing/queues.py?revision=76434&view=markup第285行,put()仅将对象放在内部队列中,如果还没有运行,启动后台线程来处理该队列中的对象.因此,代码之间each.put(obj,False)和obj.reset()代码之间存在竞争.

您应该只使用具有不可变(副本)对象的队列.