python多处理队列没有正确并行化

Question

我很难让我的代码正确并行运行。 我需要的是以下内容：

• 我有 4 个磁盘代码，我需要在一定的时间内以离散的时间步长进行演化
• 我想并行发展这些代码中的每一个

由于代码和我需要存储的数据的性质，在尝试了进程池之后，我决定最好的方法是使用队列。 我尝试了两种不同的代码。 对于这个例子， disk_codes只是数字，但在我的实际代码中，它们是另一个代码的单独实例，用于求解磁盘中的运动方程。

代码 1：

import multiprocessing

try:
    import queue
except:
    import Queue as queue

def evolve_single_disk(queue, dt):
    print "Empty queue? ", queue.empty()
    code = queue.get()
    print "Evolving disk {0} in {1}".format(code, multiprocessing.current_process().name)
    queue.task_done()

if __name__ == '__main__':
    ncores = 4
    code_queue = queue.Queue()
    disk_codes = range(ncores)

    for disk in disk_codes:
        code_queue.put(disk)

    dt = 1
    t_end = 10
    t = 0

    # Evolve!
    while t < t_end:
        print "t=", t

        processes = multiprocessing.Process(target=evolve_single_disk, args=(code_queue, dt, ))

        processes.start()
        processes.join()

        disk_codes = code_queue.get()
        print "disk codes: ", disk_codes


        t += dt

这导致：

t= 0
Empty queue?  False
Evolving disk 0 in Process-1
disk codes:  0

t= 1
Empty queue?  False
Evolving disk 1 in Process-2
disk codes:  1

t= 2
Empty queue?  False
Evolving disk 2 in Process-3
disk codes:  2

t= 3
Empty queue?  False
Evolving disk 3 in Process-4
disk codes:  3

t= 4
Empty queue?  True

因此，在每个时间步长中，其中一个磁盘都会“进化”。 这不是我想要的，因为我希望所有四个磁盘在同一时间步长中并行演化。

然后我尝试了这个：

代码 2：

import multiprocessing

try:
    import queue
except:
    import Queue as queue


def evolve_single_disk(queue, dt):
    print "Empty queue? ", queue.empty()
    code = queue.get()  
    print "Evolving disk {0} in {1}".format(code, multiprocessing.current_process().name)
    queue.task_done()


if __name__ == '__main__':
    ncores = 4
    code_queue = queue.Queue()
    disk_codes = range(ncores)

    for disk in disk_codes:
        code_queue.put(disk)

    dt = 1
    t_end = 10
    t = 0

    # Evolve!
    while t < t_end:
        print ""
        print "t=", t

        processes = [multiprocessing.Process(target=evolve_single_disk, args=(code_queue, dt, )) for x in range(ncores)]

        for p in processes:
            p.start()
            p.join()

        disk_codes = [code_queue.get() for p in processes]
        print "disk codes: ", disk_codes

        t += dt

结果是：

t= 0
Empty queue?  False
Evolving disk 0 in Process-1
Empty queue?  False
Evolving disk 0 in Process-2
Empty queue?  False
Evolving disk 0 in Process-3
Empty queue?  False
Evolving disk 0 in Process-4
disk codes:  [0, 1, 2, 3]

t= 1
Empty queue?  True

...然后代码就挂了。 所以在这里我在每个时间步启动 4 个进程，但每个进程都接收完全相同的磁盘。

我怎样才能正确地做到这一点，以便在每个时间步长有 4 个进程，并且每个进程都演化为 1 个单个磁盘？ 我已经阅读了文档和许多教程以及 SO 问题/答案，但我仍然很困惑。

编辑：

我尝试使用multiprocessing队列，但是当我尝试将磁盘代码放入队列时出现 TypeError。 可悲的是，磁盘代码也不能pickle。 使用带有磁盘代码的multiprocessing队列时的回溯：

t= 0
<multiprocessing.queues.Queue object at 0x7fee4a2d3950>
Traceback (most recent call last):
  File "/home/fran/anaconda2/lib/python2.7/multiprocessing/queues.py", line 268, in _feed
    send(obj)
TypeError: expected string or Unicode object, NoneType found
Traceback (most recent call last):
  File "/home/fran/anaconda2/lib/python2.7/multiprocessing/queues.py", line 268, in _feed
    send(obj)
TypeError: expected string or Unicode object, NoneType found
Traceback (most recent call last):
  File "/home/fran/anaconda2/lib/python2.7/multiprocessing/queues.py", line 268, in _feed
    send(obj)
TypeError: expected string or Unicode object, NoneType found
Traceback (most recent call last):
  File "/home/fran/anaconda2/lib/python2.7/multiprocessing/queues.py", line 268, in _feed
    send(obj)
TypeError: expected string or Unicode object, NoneType found
Empty queue?  True

Answer 1

将其移动到 while 循环中：

    for disk in disk_codes:
        code_queue.put(disk)

这是完整的代码：

import multiprocessing

def evolve_single_disk(queue, result, dt):
    print "Empty queue? ", queue.empty()
    code = queue.get()
    print "Evolving disk {0} in {1}".format(code, multiprocessing.current_process().name)
    result.put(code)


if __name__ == '__main__':
    ncores = 4
    code_queue = multiprocessing.Queue()
    result_queue = multiprocessing.Queue()
    disk_codes = range(ncores)

    dt = 1
    t_end = 10
    t = 0

    # Evolve!
    while t < t_end:
        for disk in disk_codes:
            code_queue.put(disk)

        print ""
        print "t=", t

        process_list = list()
        for x in range(ncores):
            process = multiprocessing.Process(target=evolve_single_disk, args=(code_queue, result_queue, dt))
            process_list.append(process)

        for p in process_list:
            p.start()
            p.join()

        disk_codes = [result_queue.get() for p in process_list]
        print "disk codes: ", disk_codes

        t += dt

输出

t= 0
Empty queue?  False
Evolving disk 0 in Process-1
Empty queue?  False
Evolving disk 1 in Process-2
Empty queue?  False
Evolving disk 2 in Process-3
Empty queue?  False
Evolving disk 3 in Process-4
disk codes:  [0, 1, 2, 3]

t= 1
Empty queue?  False
Evolving disk 0 in Process-5
Empty queue?  False
Evolving disk 1 in Process-6
Empty queue?  False
Evolving disk 2 in Process-7
Empty queue?  False
Evolving disk 3 in Process-8
disk codes:  [0, 1, 2, 3]

...

t= 9
Empty queue?  False
Evolving disk 0 in Process-37
Empty queue?  False
Evolving disk 1 in Process-38
Empty queue?  False
Evolving disk 2 in Process-39
Empty queue?  False
Evolving disk 3 in Process-40
disk codes:  [0, 1, 2, 3]