在多处理期间保持统一计数?

2022-01-12 00:00:00 python multithreading multiprocessing

问题描述

我有一个 python 程序，它运行蒙特卡罗模拟来寻找概率问题的答案.我正在使用多处理，这里是伪代码

I have a python program that runs a Monte Carlo simulation to find answers to probability questions. I am using multiprocessing and here it is in pseudo code

import multiprocessing

def runmycode(result_queue):
    print "Requested..."
    while 1==1:
       iterations +=1
    if "result found (for example)":
        result_queue.put("result!")

    print "Done"

processs = []
result_queue = multiprocessing.Queue()

for n in range(4): # start 4 processes
    process = multiprocessing.Process(target=runmycode, args=[result_queue])
    process.start()
    processs.append(process)

print "Waiting for result..."

result = result_queue.get() # wait

for process in processs: # then kill them all off
    process.terminate()

print "Got result:", result

我想对此进行扩展，以便统一计算已运行的迭代次数.就像如果线程 1 已经运行了 100 次，线程 2 已经运行了 100 次，那么我想总共显示 200 次迭代，作为控制台的打印.我指的是线程进程中的 iterations 变量.如何确保所有线程都添加到同一个变量?我认为使用 iterations 的 Global 版本会起作用，但事实并非如此.

I'd like to extend this so that I can keep a unified count of the number of iterations that have been run. Like if thread 1 has run 100 times and thread 2 has run 100 times then I want to show 200 iterations total, as a print to the console. I am referring to the iterations variable in the thread process. How can I make sure that ALL threads are adding to the same variable? I thought that using a Global version of iterations would work but it does not.

解决方案

正常的全局变量在进程之间的共享方式与线程之间的共享方式不同.您需要使用流程感知数据结构.对于您的用例，multiprocessing.Value 应该可以正常工作:

Normal global variables are not shared between processes the way they are shared between threads. You need to use a process-aware data structure. For your use-case, a multiprocessing.Value should work fine:

import multiprocessing

def runmycode(result_queue, iterations):
   print("Requested...")
   while 1==1: # This is an infinite loop, so I assume you want something else here
       with iterations.get_lock(): # Need a lock because incrementing isn't atomic
           iterations.value += 1
   if "result found (for example)":
       result_queue.put("result!")

   print("Done")


if __name__ == "__main__":
    processs = []
    result_queue = multiprocessing.Queue()

    iterations = multiprocessing.Value('i', 0)
    for n in range(4): # start 4 processes
        process = multiprocessing.Process(target=runmycode, args=(result_queue, iterations))
        process.start()
        processs.append(process)

    print("Waiting for result...")

    result = result_queue.get() # wait

    for process in processs: # then kill them all off
        process.terminate()

    print("Got result: {}".format(result))
    print("Total iterations {}".format(iterations.value))

几点说明:

  1. 我明确地将 Value 传递给孩子，以保持代码与 Windows 兼容，Windows 无法在父子之间共享读/写全局变量.
  2. 我用锁保护了增量，因为它不是原子操作，并且容易受到竞争条件的影响.
  3. 我添加了一个 if __name__ == "__main__": 保护，再次帮助提高 Windows 兼容性，并作为一般最佳实践.
  1. I explicitly passed the Value to the children, to keep the code compatible with Windows, which can't share read/write global variables between parent and children.
  2. I protected the increment with a lock, because its not an atomic operation, and is susceptible to race conditions.
  3. I added an if __name__ == "__main__": guard, again to help with Windows compatibility, and just as a general best practice.

相关文章