Example of how to use a anonymous mmap block with the multiprocessing module

mmap-mp.py

from multiprocessing import Pool
import mmap

mm = mmap.mmap(-1, 12, flags=mmap.MAP_SHARED|mmap.MAP_ANONYMOUS)
n = mm.write(b"lots of data")
mm.seek(0)

print(mm.read().decode('ascii'))
mm.seek(0)

def f(x):
    offset, char = x
    mm[offset:offset+1] = str(char).encode('ascii')

with Pool(3) as p:
    p.map(f, [(0, "f"), (1, "o"), (2, "o")])

print(mm.read().decode('ascii'))

The mmap block is shared by the parent and children processes and is writeable by all. No data needs to be serialised in order to be sent between processes. Synchronisation of access is left as an exercise for the reader :)

The mmap could also be file backed if persistence was required (remove MAP_ANONYMOUS and pass a fileno).

Oh! Very nice! I guess that article was wrong. I'd much rather use mmap than the shared_memory module. Thanks!