Multiprocessing Pool - too many file descriptors using Pytorch

Hello I'm trying to use the multiprocessing pool to try and play games simultaneously in a program I'm writing.

Here is a code snipet of what i'm doing:

self_play_pool = get_context("spawn").Pool(processes=2)
results = [self_play_pool.apply_async(self.play_game, (False,)) for _ in range(200)]
#self_play_pool.close()
				
for res in results:
     replay_buffer.save_game(res.get())


self_play_pool.terminate()

I have tried running it with the ".close()" method on the third line and without it. I also tried the concurrent.futures pool but it gave me similar errors.

What happens is that after roughly 100 games it crashes giving me either a "broken pipe" error, or a "too many files open" error, like this one:
BrokenPipeError: [Errno 32] Broken pipe

For what I was able to gather online, this seems to happen when there are too many processes running since each requires a file descriptor... However I only have 2 processes running simultaneasly.
I also tried to use the "with" context manager but the same happened.

Any help would be apreciated.
Thank you.

I found the solution for my problem, if anyone finds this post and wants to know... If you change the pytorch's sharing strategy with the function torch.multiprocessing.set_sharing_strategy('file_system'), it won't create extra file descriptors.
It is very important to note that you must run the function both inside the father process and inside each child. I did it by simply putting it as the first line of my play_game function, but I think you can achieve the same with the multiprocessing.pool "initializer" argument.