共享只读数据是否复制到不同的进程以进行多处

问题描述

我的这段代码看起来像这样:

The piece of code that I have looks some what like this:

glbl_array = # a 3 Gb array

def my_func( args, def_param = glbl_array):
    #do stuff on args and def_param

if __name__ == '__main__':
  pool = Pool(processes=4)
  pool.map(my_func, range(1000))

有没有办法确保(或鼓励)不同的进程不会获得 glbl_array 的副本而是共享它.如果无法停止复制，我将使用 memmapped 数组，但我的访问模式不是很规律，所以我希望 memmapped 数组更慢.以上似乎是首先要尝试的.这是在 Linux 上.我只是想从 Stackoverflow 获得一些建议，不想惹恼系统管理员.如果第二个参数是真正的不可变对象，如 glbl_array.tostring()，您认为会有帮助吗?

Is there a way to make sure (or encourage) that the different processes does not get a copy of glbl_array but shares it. If there is no way to stop the copy I will go with a memmapped array, but my access patterns are not very regular, so I expect memmapped arrays to be slower. The above seemed like the first thing to try. This is on Linux. I just wanted some advice from Stackoverflow and do not want to annoy the sysadmin. Do you think it will help if the the second parameter is a genuine immutable object like glbl_array.tostring().

推荐答案

您可以很容易地将 multiprocessing 中的共享内存内容与 Numpy 一起使用:

You can use the shared memory stuff from multiprocessing together with Numpy fairly easily:

import multiprocessing
import ctypes
import numpy as np

shared_array_base = multiprocessing.Array(ctypes.c_double, 10*10)
shared_array = np.ctypeslib.as_array(shared_array_base.get_obj())
shared_array = shared_array.reshape(10, 10)

#-- edited 2015-05-01: the assert check below checks the wrong thing
#   with recent versions of Numpy/multiprocessing. That no copy is made
#   is indicated by the fact that the program prints the output shown below.
## No copy was made
##assert shared_array.base.base is shared_array_base.get_obj()

# Parallel processing
def my_func(i, def_param=shared_array):
    shared_array[i,:] = i

if __name__ == '__main__':
    pool = multiprocessing.Pool(processes=4)
    pool.map(my_func, range(10))

    print shared_array

打印

[[ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.]
 [ 1.  1.  1.  1.  1.  1.  1.  1.  1.  1.]
 [ 2.  2.  2.  2.  2.  2.  2.  2.  2.  2.]
 [ 3.  3.  3.  3.  3.  3.  3.  3.  3.  3.]
 [ 4.  4.  4.  4.  4.  4.  4.  4.  4.  4.]
 [ 5.  5.  5.  5.  5.  5.  5.  5.  5.  5.]
 [ 6.  6.  6.  6.  6.  6.  6.  6.  6.  6.]
 [ 7.  7.  7.  7.  7.  7.  7.  7.  7.  7.]
 [ 8.  8.  8.  8.  8.  8.  8.  8.  8.  8.]
 [ 9.  9.  9.  9.  9.  9.  9.  9.  9.  9.]]