multiprocessing.Pool 在 Linux/Python2.7 上终止（）后生成新的子进程？

我有一个可执行文件，我需要经常使用不同的参数运行它。为此，我使用多处理模块编写了一个小型 Python (2.7) 包装器，遵循给定的模式here.

我的代码如下所示：

try:
     logging.info("starting pool runs")
     pool.map(run_nlin, params)
     pool.close()
 except KeyboardInterrupt:
     logging.info("^C pressed")
     pool.terminate()
 except Exception, e:
     logging.info("exception caught: ", e)
     pool.terminate()
 finally:
     time.sleep(5)
     pool.join()
     logging.info("done")

我的工作人员功能在这里：

class KeyboardInterruptError(Exception): pass

def run_nlin((path_config, path_log, path_nlin, update_method)):
    try:
        with open(path_log, "w") as log_:
            cmdline = [path_nlin, path_config]
            if update_method:
                cmdline += [update_method, ]
            sp.call(cmdline, stdout=log_, stderr=log_)
    except KeyboardInterrupt:
        time.sleep(5)
        raise KeyboardInterruptError()
    except:
        raise

path_config是二进制程序的配置文件的路径；那里有例如运行程序的日期。

当我启动包装时，一切看起来都很好。但是，当我按^C，包装脚本似乎启动了一个额外的numproc在终止之前从池中删除进程。例如，当我启动第 1-10 天的脚本时，我可以在ps aux输出表明二进制程序的两个实例正在运行（通常在第 1 天和第 3 天）。现在，当我按下^C，包装脚本退出，第 1 天和第 3 天的二进制程序消失了，但第 5 天和第 7 天有新的二进制程序运行。

所以对我来说似乎Pool启动另一个numproc最终死亡之前的过程。

你知道这里发生了什么吗？我能做些什么？

On 这一页，多处理模块的作者 Jesse Noller 展示了处理的正确方法KeyboardInterrupt是让子进程返回——而不是重新引发异常。这允许主进程终止池。

然而，正如下面的代码所示，主进程并没有到达except KeyboardInterrupt阻塞直到after生成的所有任务pool.map已运行。这就是为什么（我相信）您会看到对工作函数的额外调用，run_nlin, after Ctrl-C已被按下。

一种可能的解决方法是测试所有工作函数是否存在multiprocessing.Event已经设置好了。如果事件已定，则让工人尽早退出，否则继续进行长期计算。

import logging
import multiprocessing as mp
import time

logger = mp.log_to_stderr(logging.WARNING)

def worker(x):
    try:
        if not terminating.is_set():
            logger.warn("Running worker({x!r})".format(x = x))
            time.sleep(3)
        else:
            logger.warn("got the message... we're terminating!")
    except KeyboardInterrupt:
        logger.warn("terminating is set")        
        terminating.set()
    return x

def initializer(terminating_):
    # This places terminating in the global namespace of the worker subprocesses.
    # This allows the worker function to access `terminating` even though it is
    # not passed as an argument to the function.
    global terminating
    terminating = terminating_

def main():
    terminating = mp.Event()    
    result = []
    pool = mp.Pool(initializer=initializer, initargs=(terminating, ))
    params = range(12)
    try:
         logger.warn("starting pool runs")
         result = pool.map(worker, params)
         pool.close()
    except KeyboardInterrupt:
        logger.warn("^C pressed")
        pool.terminate()
    finally:
        pool.join()
        logger.warn('done: {r}'.format(r = result))

if __name__ == '__main__':
    main()

运行脚本会产生：

% test.py
[WARNING/MainProcess] starting pool runs
[WARNING/PoolWorker-1] Running worker(0)
[WARNING/PoolWorker-2] Running worker(1)
[WARNING/PoolWorker-3] Running worker(2)
[WARNING/PoolWorker-4] Running worker(3)

这里按下 Ctrl-C；每个工人设定terminating事件。我们实际上只需要一个来设置它，但是尽管效率有点低，但这仍然有效。

  C-c C-c[WARNING/PoolWorker-4] terminating is set
[WARNING/PoolWorker-2] terminating is set
[WARNING/PoolWorker-3] terminating is set
[WARNING/PoolWorker-1] terminating is set

现在所有其他任务都在排队pool.map正在运行：

[WARNING/PoolWorker-4] got the message... we're terminating!
[WARNING/PoolWorker-2] got the message... we're terminating!
[WARNING/PoolWorker-1] got the message... we're terminating!
[WARNING/PoolWorker-2] got the message... we're terminating!
[WARNING/PoolWorker-4] got the message... we're terminating!
[WARNING/PoolWorker-2] got the message... we're terminating!
[WARNING/PoolWorker-1] got the message... we're terminating!
[WARNING/PoolWorker-3] got the message... we're terminating!

最终主流程到达except KeyboardInterrupt block.

[WARNING/MainProcess] ^C pressed
[WARNING/MainProcess] done: []