在ProcessPoolExecutor中打开pandas数据框时，永远不会释放内存

Question

以下是我正在使用的代码的一个非常简单的示例...

from concurrent.futures import ProcessPoolExecutor
import pandas


if __name__ == "__main__":

    def i_use_lots_of_memory():
        print 'doing something that uses a lot of memory'
        data = pandas.read_csv('large_txt_file.txt')
        del data
        # do other things here as soon as I've solved mem usage issues
        print 'ha ha  I used up a ton of memory.'

    def simplest_callback_ever(future):
        _ = future.result()
        print 'callback was run'


    class ManagesFileReading(object):
        def __init__(self):
            self.pool = ProcessPoolExecutor(max_workers=24)

        def add_job(self, callback=None):
            future = self.pool.submit(i_use_lots_of_memory)
            if callback:
                future.add_done_callback(callback)


    mfr = ManagesFileReading()
    mfr.add_job(simplest_callback_ever)

在这个例子中，我打开一个800MB的文本文件，占用大约2GB的内存。输出是......

doing something that uses a lot of memory
ha ha  I used up a ton of memory.
callback function was run. Task is complete.

所以任务完成，问题是内存永远不会释放。即使未来已经完成，它也永远不会释放内存。我可以释放它的唯一方法是通过运行self._pool.shutdown（）

来关闭进程池

除非我误解了ProcessPoolExecutor是如何工作的，否则当回调函数完成时意味着任务已经完成，对吧？为什么未来被删除以及内存被释放？有什么想法吗？