How to do parallel programming in Python?

Question

For C++, we can use OpenMP to do parallel programming; however, OpenMP will not work for Python. What should I do if I want to parallel some parts of my python program?

The structure of the code may be considered as:

solve1(A)
solve2(B)

Where solve1 and solve2 are two independent function. How to run this kind of code in parallel instead of in sequence in order to reduce the running time? The code is:

def solve(Q, G, n):
    i = 0
    tol = 10 ** -4

    while i < 1000:
        inneropt, partition, x = setinner(Q, G, n)
        outeropt = setouter(Q, G, n)

        if (outeropt - inneropt) / (1 + abs(outeropt) + abs(inneropt)) < tol:
            break
            
        node1 = partition[0]
        node2 = partition[1]
    
        G = updateGraph(G, node1, node2)

        if i == 999:
            print "Maximum iteration reaches"
    print inneropt

Where setinner and setouter are two independent functions. That's where I want to parallel...

Answer 1

You can use the multiprocessing module. For this case I might use a processing pool:

from multiprocessing import Pool
pool = Pool()
result1 = pool.apply_async(solve1, [A])    # evaluate "solve1(A)" asynchronously
result2 = pool.apply_async(solve2, [B])    # evaluate "solve2(B)" asynchronously
answer1 = result1.get(timeout=10)
answer2 = result2.get(timeout=10)

This will spawn processes that can do generic work for you. Since we did not pass processes, it will spawn one process for each CPU core on your machine. Each CPU core can execute one process simultaneously.

If you want to map a list to a single function you would do this:

args = [A, B]
results = pool.map(solve1, args)

Don't use threads because the GIL locks any operations on python objects.

Answer 2

This can be done very elegantly with Ray.

To parallelize your example, you'd need to define your functions with the @ray.remote decorator, and then invoke them with .remote.

import ray

ray.init()

# Define the functions.

@ray.remote
def solve1(a):
    return 1

@ray.remote
def solve2(b):
    return 2

# Start two tasks in the background.
x_id = solve1.remote(0)
y_id = solve2.remote(1)

# Block until the tasks are done and get the results.
x, y = ray.get([x_id, y_id])

There are a number of advantages of this over the multiprocessing module.

1. The same code will run on a multicore machine as well as a cluster of machines.
2. Processes share data efficiently through shared memory and zero-copy serialization.
3. Error messages are propagated nicely.

4. These function calls can be composed together, e.g.,

   @ray.remote
   def f(x):
       return x + 1
   
   x_id = f.remote(1)
   y_id = f.remote(x_id)
   z_id = f.remote(y_id)
   ray.get(z_id)  # returns 4
   

5. In addition to invoking functions remotely, classes can be instantiated remotely as actors.

Note that Ray is a framework I've been helping develop.