Goal: a function get_free_port() -> port number that supports multi-thread and multi-process race condition.

That means you can call get_free_port() no matter when and where and you should be able to bind that port number care-free.

Actually, the goal is not logically attainable, because get_free_port() will have no idea whether ports output from it have done being used or not. So, the best it can do is to remember every port output, and make sure not to use the one in the list.

It should be more logical if we add another part to the function which goes release_port() or another way that will do this automatically.

We know that calling socket.bind(('', 0)) will automatically bind a free port for us. In a sense, we can do something like this:

import socket 

def get_free_port():
	sock = socket.socket()
	sock.bind(('', 0))
	ip, port = sock.getnameinfo()
	return port

It seems … working but not quite. Because this approach let the port unbound for a short time, and thus subjective to race condition.

So far, we know that getting a free port is not so hard, but transferring a port safely to a caller is much much harder.

In thinking about this, we should come up with an idea that the port, being unbound, should first have another mechanism to prevent any mischievous binding. I have come up with something like this:

get a free port -> look at dictionary (and lock file) -> bind a free port -> write a dictionary (and lock file) -> release port -> return the port

From above, you will see that I invoke the use of lock file by which I can be certain that even the port is left unbound it will not be reacquired by any process in any possible race conditions. (I use lib fasteners here)

I have translated this into this code (with testing):