Python:多处理和c_char_p数组

时间:2013-01-08 14:36:42

标签: python multiprocessing ctypes

我正在启动3个进程,我希望他们将一个字符串放入一个共享数组,在与进程(i)对应的索引处。

查看下面的代码,生成的输出是:

['test 0', None, None]
['test 1', 'test 1', None]
['test 2', 'test 2', 'test 2']

为什么'test 0'被test 1覆盖,test 1test 2覆盖?

我想要的是(顺序并不重要):

['test 0', None, None]
['test 0', 'test 1', None]
['test 0', 'test 1', 'test 2']

代码:

#!/usr/bin/env python

import multiprocessing
from multiprocessing import Value, Lock, Process, Array
import ctypes
from ctypes import c_int, c_char_p

class Consumer(multiprocessing.Process):
    def __init__(self, task_queue, result_queue, arr, lock):
            multiprocessing.Process.__init__(self)
            self.task_queue = task_queue
            self.result_queue = result_queue
            self.arr = arr
            self.lock = lock

    def run(self):
            proc_name = self.name
            while True:
                next_task = self.task_queue.get()
                if next_task is None:
                    self.task_queue.task_done()
                    break            
                answer = next_task(arr=self.arr, lock=self.lock)
                self.task_queue.task_done()
                self.result_queue.put(answer)
            return

class Task(object):
    def __init__(self, i):
        self.i = i

    def __call__(self, arr=None, lock=None):
        with lock:
            arr[self.i] = "test %d" % self.i
            print arr[:]

    def __str__(self):
        return 'ARC'

    def run(self):
        print 'IN'

if __name__ == '__main__':
   tasks = multiprocessing.JoinableQueue()
   results = multiprocessing.Queue()

   arr = Array(ctypes.c_char_p, 3)

   lock = multiprocessing.Lock()

   num_consumers = multiprocessing.cpu_count() * 2
   consumers = [Consumer(tasks, results, arr, lock) for i in xrange(num_consumers)]

   for w in consumers:
      w.start()

   for i in xrange(3):
      tasks.put(Task(i))

   for i in xrange(num_consumers):
      tasks.put(None)

我正在运行Python 2.7.3(Ubuntu)

1 个答案:

答案 0 :(得分:5)

此问题与this one类似。在那里,J.F。Sebastian推测将arr[i]arr[i]分配给一个只对进行赋值的子进程有意义的内存地址。其他子进程在查看该地址时检索垃圾。

至少有两种方法可以避免这个问题。一种是使用multiprocessing.manager列表:

import multiprocessing as mp

class Consumer(mp.Process):
    def __init__(self, task_queue, result_queue, lock, lst):
            mp.Process.__init__(self)
            self.task_queue = task_queue
            self.result_queue = result_queue
            self.lock = lock
            self.lst = lst

    def run(self):
            proc_name = self.name
            while True:
                next_task = self.task_queue.get()
                if next_task is None:
                    self.task_queue.task_done()
                    break            
                answer = next_task(lock = self.lock, lst = self.lst)
                self.task_queue.task_done()
                self.result_queue.put(answer)
            return

class Task(object):
    def __init__(self, i):
        self.i = i

    def __call__(self, lock, lst):
        with lock:
            lst[self.i] = "test {}".format(self.i)
            print([lst[i] for i in range(3)])

if __name__ == '__main__':
   tasks = mp.JoinableQueue()
   results = mp.Queue()
   manager = mp.Manager()
   lst = manager.list(['']*3)

   lock = mp.Lock()
   num_consumers = mp.cpu_count() * 2
   consumers = [Consumer(tasks, results, lock, lst) for i in xrange(num_consumers)]

   for w in consumers:
      w.start()

   for i in xrange(3):
      tasks.put(Task(i))

   for i in xrange(num_consumers):
      tasks.put(None)

   tasks.join()

另一种方法是使用具有固定大小的共享数组,例如mp.Array('c', 10)

import multiprocessing as mp

class Consumer(mp.Process):
    def __init__(self, task_queue, result_queue, arr, lock):
            mp.Process.__init__(self)
            self.task_queue = task_queue
            self.result_queue = result_queue
            self.arr = arr
            self.lock = lock

    def run(self):
            proc_name = self.name
            while True:
                next_task = self.task_queue.get()
                if next_task is None:
                    self.task_queue.task_done()
                    break            
                answer = next_task(arr = self.arr, lock = self.lock)
                self.task_queue.task_done()
                self.result_queue.put(answer)
            return

class Task(object):
    def __init__(self, i):
        self.i = i

    def __call__(self, arr, lock):
        with lock:
            arr[self.i].value = "test {}".format(self.i)
            print([a.value for a in arr])

if __name__ == '__main__':
   tasks = mp.JoinableQueue()
   results = mp.Queue()
   arr = [mp.Array('c', 10) for i in range(3)]

   lock = mp.Lock()
   num_consumers = mp.cpu_count() * 2
   consumers = [Consumer(tasks, results, arr, lock) for i in xrange(num_consumers)]

   for w in consumers:
      w.start()

   for i in xrange(3):
      tasks.put(Task(i))

   for i in xrange(num_consumers):
      tasks.put(None)

   tasks.join()

我推测,当mp.Array(ctypes.c_char_p, 3)没有时,这是有效的,因为mp.Array('c', 10)具有固定大小,因此内存地址永远不会改变,而mp.Array(ctypes.c_char_p, 3)具有可变大小,所以将arr[i]分配给更大的字符串时,内存地址可能会发生变化。

也许这就是the docs在发表声明时发出的警告,

  

虽然可以将指针存储在共享内存中,但请记住   这将引用特定地址空间中的位置   处理。但是,指针很可能在无效中   第二个进程的上下文,并尝试取消引用指针   第二个过程可能会导致崩溃。