asyncio on a second thread

[bjpirt]

Hi all,
I'm currently trying to run a second thread on an ESP32 running asyncio (but with no asyncio on the main thread currently). I've seen references on the internet about how asyncio doesn't work on threads - before I start debugging why my code isn't working I just wanted to check if there are still issues with this arrangement?
Thanks,
Ben

[andrewleech]

Currently in micropython there's only one global asyncio event loop, I don't think it will be possible to have a separate loop on a second thread.

[jimmo]

@bjpirt As long as you only run a single event loop, there's no problem. (@andrewleech -- asyncio doesn't know about threads and doesn't have any thread-local storage, it doesn't care which thread it's running on).

That said, you need to be very careful that you do not access/modify any asyncio internal state concurrently (e.g. from another thread or interrupt handler). e.g. you may not call create_task from other threads, or Event.set, while asyncio is running (on any thread). MicroPython has asyncio.ThreadSafeFlag which you can use to safely signal into asyncio from another thread or interrupt handler.

[andrewleech]

Thanks @jimmo for the extra detail. I assumed the question was about being able to run two separate loops concurrently in separate threads; this is possible on cpython but not on micropython.

The single event loop will only be serviced / async functions run on the thread it was started on.

[peterhinch]

@bjpirt This discussion was about running asyncio on core 1 and delegating work to core 0. While clever, I'm unconvinced that it is an effective way to design applications.

There can only be one instance of asyncio. Core 0 has privileged access to hardware. ISR's run on core 0. In my opinion the best design is to run asyncio on core 0, delegating activities to core 1 only where they would otherwise stall core 0. This takes advantage of the excellent capability of asyncio to handle concurrency in hardware interfaces while using core 1 to do those things where it would fail, namely dealing with blocking methods and demanding computational loads.

[bjpirt]

Thanks @peterhinch - that's good info about the Core 0 priority

[bjpirt]

Thanks for the replies all - so I should be good to run a single instance of uasyncio on Core 1, but it's better to use Core 0 if I need to access hardware.

[andrewleech]

There's also some previous discussion about separate thread loops here: #8340