Defining the amount of threads in use for sync functions

[ivergara]

How is the number of threads launched for an asynchronous function being decorated with unsync? Is there a possibility to define that number? If it's not possible out of the box, how could I start approaching that?

[alex-sherman]

The async loop runs in a single thread and "regular" functions decorated with @unsync without setting cpu_bound = True will run in a ThreadPoolExecutor.

Assuming your referring to the later, without any arguments ThreadPoolExecutor will overprovision (see max_workers) in Python 3.5 and later, creating 5 threads per core. If that's insufficient I think you can just overwrite unsync.thread_executor = concurrent.futures.ThreadPoolExecutor(max_workers = 9000).

I'm curious what your workload is that would you would notice this, is it really doing hundreds of IO bound operations?

[ivergara]

Thanks for the answer. Indeed it was the later.

I'm not noticing it at the moment, but I was preemptively wondering how to control the number of threads, particularly at different steps. For example, due to a limitation on external resources, I could be limited by the number of threads accessing that resource but then local downstream processing could benefit from more threads.

[alex-sherman]

Makes sense. I'll close this for now since I think just assigning over unsync.thread_executor will cover your scenario.

Feel free to reopen, or open a PR if you'd like to add a cleaner API or something for setting the number of threads. My own opinion is that a function like unsync.set_max_thread_workers(n) is less clean than just assigning to thread_executor.