问题描述

multiprocessing.Pool 快把我逼疯了...
我想升级许多软件包，并且对于每个软件包，我都必须检查是否有更高版本.这是由 check_one 函数完成的.
主要代码在 Updater.update 方法中:在那里我创建了 Pool 对象并调用 map() 方法.

multiprocessing.Pool is driving me crazy...
I want to upgrade many packages, and for every one of them I have to check whether there is a greater version or not. This is done by the check_one function.
The main code is in the Updater.update method: there I create the Pool object and call the map() method.

代码如下:

def check_one(args):
    res, total, package, version = args
    i = res.qsize()
    logger.info('[{0:.1%} - {1}, {2} / {3}]',
        i / float(total), package, i, total, addn=False)
    try:
        json = PyPIJson(package).retrieve()
        new_version = Version(json['info']['version'])
    except Exception as e:
        logger.error('Error: Failed to fetch data for {0} ({1})', package, e)
        return
    if new_version > version:
        res.put_nowait((package, version, new_version, json))

class Updater(FileManager):

    # __init__ and other methods...

    def update(self):    
        logger.info('Searching for updates')
        packages = Queue.Queue()
        data = ((packages, self.set_len, dist.project_name, Version(dist.version)) 
            for dist in self.working_set)
        pool = multiprocessing.Pool()
        pool.map(check_one, data)
        pool.close()
        pool.join()
        while True:
            try:
                package, version, new_version, json = packages.get_nowait()
            except Queue.Empty:
                break
            txt = 'A new release is avaiable for {0}: {1!s} (old {2}), update'.format(package,
                                                                                      new_version,
                                                                                      version)
            u = logger.ask(txt, bool=('upgrade version', 'keep working version'), dont_ask=self.yes)
            if u:
                self.upgrade(package, json, new_version)
            else:
                logger.info('{0} has not been upgraded', package)
        self._clean()
        logger.success('Updating finished successfully')

当我运行它时，我得到了这个奇怪的错误:

When I run it I get this weird error:

Searching for updates
Exception in thread Thread-1:
Traceback (most recent call last):
  File "/usr/lib/python2.7/threading.py", line 552, in __bootstrap_inner
    self.run()
  File "/usr/lib/python2.7/threading.py", line 505, in run
    self.__target(*self.__args, **self.__kwargs)
  File "/usr/local/lib/python2.7/dist-packages/multiprocessing/pool.py", line 225, in _handle_tasks
    put(task)
PicklingError: Can't pickle <type 'thread.lock'>: attribute lookup thread.lock failed

推荐答案

multiprocessing 通过 将任务(包括 check_one 和 data)传递给工作进程mp.SimpleQueue.与 Queue.Queue 不同，放在 mp.SimpleQueue 中的所有内容都必须是可选择的.Queue.Queues 是不可挑选的:

multiprocessing passes tasks (which include check_one and data) to the worker processes through a mp.SimpleQueue. Unlike Queue.Queues, everything put in the mp.SimpleQueue must be pickable. Queue.Queues are not pickable:

import multiprocessing as mp
import Queue

def foo(queue):
    pass

pool=mp.Pool()
q=Queue.Queue()

pool.map(foo,(q,))

产生此异常:

UnpickleableError: Cannot pickle <type 'thread.lock'> objects

您的 data 包括 packages，这是一个 Queue.Queue.这可能是问题的根源.

Your data includes packages, which is a Queue.Queue. That might be the source of the problem.

这是一个可能的解决方法:Queue 用于两个目的:

Here is a possible workaround: The Queue is being used for two purposes:

1. 找出近似大小(通过调用qsize)
2. 存储结果以供日后检索.

我们可以使用 mp.Value，而不是调用 qsize，以便在多个进程之间共享一个值.

Instead of calling qsize, to share a value between multiple processes, we could use a mp.Value.

我们可以(并且应该)只返回来自对 check_one 的调用的值，而不是将结果存储在队列中.pool.map 将结果收集到自己制作的队列中，并将结果作为 pool.map 的返回值返回.

Instead of storing results in a queue, we can (and should) just return values from calls to check_one. The pool.map collects the results in a queue of its own making, and returns the results as the return value of pool.map.

例如:

import multiprocessing as mp
import Queue
import random
import logging

# logger=mp.log_to_stderr(logging.DEBUG)
logger = logging.getLogger(__name__)


qsize = mp.Value('i', 1)
def check_one(args):
    total, package, version = args
    i = qsize.value
    logger.info('[{0:.1%} - {1}, {2} / {3}]'.format(
        i / float(total), package, i, total))
    new_version = random.randrange(0,100)
    qsize.value += 1
    if new_version > version:
        return (package, version, new_version, None)
    else:
        return None

def update():    
    logger.info('Searching for updates')
    set_len=10
    data = ( (set_len, 'project-{0}'.format(i), random.randrange(0,100))
             for i in range(set_len) )
    pool = mp.Pool()
    results = pool.map(check_one, data)
    pool.close()
    pool.join()
    for result in results:
        if result is None: continue
        package, version, new_version, json = result
        txt = 'A new release is avaiable for {0}: {1!s} (old {2}), update'.format(
            package, new_version, version)
        logger.info(txt)
    logger.info('Updating finished successfully')

if __name__=='__main__':
    logging.basicConfig(level=logging.DEBUG)
    update()