Python multiprocessing.Pool.map以静默方式死亡

Question

我试图并行放置一个for循环来加速一些代码。 考虑一下：

from multiprocessing import Pool

results = []

def do_stuff(str):
    print str
    results.append(str)

p = Pool(4)
p.map(do_stuff, ['str1','str2','str3',...]) # many strings here ~ 2000
p.close()

print results

我有一些调试消息显示来自do_stuff以跟踪程序在死亡前的程度。 它似乎每次都在不同的点死亡。 例如，它将打印'str297'，然后它将停止运行，我将看到所有的CPU停止工作，程序就在那里。 应该发生一些错误，但没有显示错误消息。 有谁知道如何调试这个问题？

UPDATE

我尝试了一点点重新编写代码。 而不是使用map函数我尝试了apply_async函数，如下所示：

        pool = Pool(5)
        results = pool.map(do_sym, underlyings[0::10])
        results = []
        for sym in underlyings[0::10]:
            r = pool.apply_async(do_sym, [sym])
            results.append(r)

        pool.close()
        pool.join()

        for result in results:
            print result.get(timeout=1000)

这与map功能一样好，但最终以同样的方式悬挂。 它永远不会到达打印结果的for循环。

在进行了更多的工作之后，尝试了一些调试日志，就像在unutbu的回答中提到的那样，我将在这里提供更多信息。 问题很奇怪。 似乎游泳池只是挂在那里，无法关闭并继续该计划。 我使用PyDev环境来测试我的程序，但我想我会尝试在控制台中运行python。 在控制台中我得到了相同的行为，但是当我按下control + C来杀死程序时，我得到了一些输出，可以解释问题所在：

> KeyboardInterrupt ^CProcess PoolWorker-47: Traceback (most recent call
> last):   File "/usr/lib/python2.7/multiprocessing/process.py", line
> 258, in _bootstrap Process PoolWorker-48: Traceback (most recent call
> last):   File "/usr/lib/python2.7/multiprocessing/process.py", line
> 258, in _bootstrap Process PoolWorker-45: Process PoolWorker-46:
> Process PoolWorker-44:
>     self.run()   File "/usr/lib/python2.7/multiprocessing/process.py", line 114, in run
>     self._target(*self._args, **self._kwargs)   File "/usr/lib/python2.7/multiprocessing/pool.py", line 102, in worker
> Traceback (most recent call last): Traceback (most recent call last):
> Traceback (most recent call last):   File
> "/usr/lib/python2.7/multiprocessing/process.py", line 258, in
> _bootstrap   File "/usr/lib/python2.7/multiprocessing/process.py", line 258, in _bootstrap   File
> "/usr/lib/python2.7/multiprocessing/process.py", line 258, in
> _bootstrap
>     task = get()   File "/usr/lib/python2.7/multiprocessing/queues.py", line 374, in get
>     self.run()   File "/usr/lib/python2.7/multiprocessing/process.py", line 114, in run
>     racquire()
>     self._target(*self._args, **self._kwargs)   File "/usr/lib/python2.7/multiprocessing/pool.py", line 102, in worker
> KeyboardInterrupt
>     task = get()   File "/usr/lib/python2.7/multiprocessing/queues.py", line 374, in get
>     self.run()
>     self.run()   File "/usr/lib/python2.7/multiprocessing/process.py", line 114, in run
>     self.run()   File "/usr/lib/python2.7/multiprocessing/process.py", line 114, in run   File
> "/usr/lib/python2.7/multiprocessing/process.py", line 114, in run
>     self._target(*self._args, **self._kwargs)   File "/usr/lib/python2.7/multiprocessing/pool.py", line 102, in worker
>     self._target(*self._args, **self._kwargs)
>     self._target(*self._args, **self._kwargs)   File "/usr/lib/python2.7/multiprocessing/pool.py", line 102, in worker
>     racquire()   File "/usr/lib/python2.7/multiprocessing/pool.py", line 102, in worker KeyboardInterrupt
>     task = get()   File "/usr/lib/python2.7/multiprocessing/queues.py", line 374, in get
>     task = get()
>     task = get()   File "/usr/lib/python2.7/multiprocessing/queues.py", line 376, in get  
> File "/usr/lib/python2.7/multiprocessing/queues.py", line 374, in get
>     racquire()
>     return recv()
>     racquire() KeyboardInterrupt KeyboardInterrupt KeyboardInterrupt

然后实际上程序永远不会死。 我最终不得不关闭终端窗口来杀死它。

更新2

我缩小了池中运行的函数内部的问题，这是导致问题的MySQL数据库事务。 我之前使用的是MySQLdb包。 我为事务切换了一个pandas.read_sql函数，它现在正在运行。

Answer 1

pool.map在列表中返回结果。 因此，不要在并发进程中调用results.append （由于每个进程都有自己独立的results副本，因此无法工作）， pool.map将results分配给主进程中pool.map返回的值：

import multiprocessing as mp

def do_stuff(text):
    return text

if __name__ == '__main__':
    p = mp.Pool(4)
    tasks = ['str{}'.format(i) for i in range(2000)]
    results = p.map(do_stuff, tasks)
    p.close()

    print(results)

产量

['str0', 'str1', 'str2', 'str3', ...]

---

调试使用多处理的脚本的一种方法是添加日志记录语句。 mp.log_to_stderr ， multiprocessing模块提供辅助函数mp.log_to_stderr 。 例如，

import multiprocessing as mp
import logging

logger = mp.log_to_stderr(logging.DEBUG)

def do_stuff(text):
    logger.info('Received {}'.format(text))
    return text

if __name__ == '__main__':
    p = mp.Pool(4)
    tasks = ['str{}'.format(i) for i in range(2000)]
    results = p.map(do_stuff, tasks)
    p.close()

    logger.info(results)

产生日志输出，如：

[DEBUG/MainProcess] created semlock with handle 139824443588608
[DEBUG/MainProcess] created semlock with handle 139824443584512
[DEBUG/MainProcess] created semlock with handle 139824443580416
[DEBUG/MainProcess] created semlock with handle 139824443576320
[DEBUG/MainProcess] added worker
[INFO/PoolWorker-1] child process calling self.run()
[DEBUG/MainProcess] added worker
[INFO/PoolWorker-2] child process calling self.run()
[DEBUG/MainProcess] added worker
[INFO/PoolWorker-3] child process calling self.run()
[DEBUG/MainProcess] added worker
[INFO/PoolWorker-4] child process calling self.run()
[INFO/PoolWorker-1] Received str0
[INFO/PoolWorker-2] Received str125
[INFO/PoolWorker-3] Received str250
[INFO/PoolWorker-4] Received str375
[INFO/PoolWorker-3] Received str251
...
[INFO/PoolWorker-4] Received str1997
[INFO/PoolWorker-4] Received str1998
[INFO/PoolWorker-4] Received str1999
[DEBUG/MainProcess] closing pool
[INFO/MainProcess] ['str0', 'str1', 'str2', 'str3', ...]
[DEBUG/MainProcess] worker handler exiting
[DEBUG/MainProcess] task handler got sentinel
[INFO/MainProcess] process shutting down
[DEBUG/MainProcess] task handler sending sentinel to result handler
[DEBUG/MainProcess] running all "atexit" finalizers with priority >= 0
[DEBUG/MainProcess] finalizing pool
[DEBUG/MainProcess] task handler sending sentinel to workers
[DEBUG/MainProcess] helping task handler/workers to finish
[DEBUG/MainProcess] result handler got sentinel
[DEBUG/PoolWorker-3] worker got sentinel -- exiting
[DEBUG/MainProcess] removing tasks from inqueue until task handler finished
[DEBUG/MainProcess] ensuring that outqueue is not full
[DEBUG/MainProcess] task handler exiting
[DEBUG/PoolWorker-3] worker exiting after 2 tasks
[INFO/PoolWorker-3] process shutting down
[DEBUG/MainProcess] result handler exiting: len(cache)=0, thread._state=0
[DEBUG/PoolWorker-3] running all "atexit" finalizers with priority >= 0
[DEBUG/MainProcess] joining worker handler
[DEBUG/MainProcess] terminating workers
[DEBUG/PoolWorker-3] running the remaining "atexit" finalizers
[DEBUG/MainProcess] joining task handler
[DEBUG/MainProcess] joining result handler
[DEBUG/MainProcess] joining pool workers
[DEBUG/MainProcess] cleaning up worker 4811
[DEBUG/MainProcess] running the remaining "atexit" finalizers

请注意，每行指示发出记录记录的进程。 因此，输出在某种程度上序列化了并发进程中的事件顺序。

通过明智地放置logging.info调用，你应该能够缩小你的脚本“静静地死”的地方，也许也就是为什么（或者，至少它不会像死的那样沉默）。