python multiprocessing is loosing values

Question

I tried to speed up a calculation using Pool from the multiprocessing package. While I did get a significant speedup I'm missing more and more values as I increase the core/worker count.

I share my variables with all processes through the mp.value() class.

Where did i go wrong and how can i fix this?

poss = [x+1 for x in range(20)]
all_rolls = itertools.product(poss, repeat=6)

win = mp.Value('i', 0)
draw = mp.Value('i', 0)
loose = mp.Value('i', 0)

def some_func(roll):
  if(comparison on rolls):
    win.value += 1
  elif(other comparison):
    draw.value +=1
  else:
    loose.value +=1

with Pool(8) as p:
  p.map(some_func, all_rolls)

On 16 cores i got 55,923,638 values instead of 64,000,000

Answer 1

In addition to what @jfowkes answered, note that you can use each Value with its own lock, which might make things a bit faster:

win = mp.Value('i', lock = True)
draw = mp.Value('i', lock = True)
loose = mp.Value('i', lock = True)

def some_func(roll):
    if(comparison on rolls):
        with win.get_lock() : 
            win.value += 1
    elif(other comparison):
        with draw.get_lock():
            draw.value +=1
    else:
        with loose.get_lock():
            loose.value +=1

Answer 2

You need to protect the modification of your values with Lock (see this article).

from multiprocessing import Lock

lock = Lock()

def some_func(roll):
    with lock:
        if(comparison on rolls):
            win.value += 1
        elif(other comparison):
            draw.value +=1
        else:
            loose.value +=1