aioble peripheral may need throttling to avoid random disconnects

[PBrunot]

I am using AIOBLE (latest version as per December 2022) on ESP32-C3 with Micropython 1.19 (LOLIN D32 board).
My app uses 3 BT LE services (peripheral role),:

• a custom service for requests (with 1 write-only and 1notify-only characteristics)
• a battery_service
• a device_information service

I observed frequent client disconnects to client (I tested with another ESP32 and with NRFConnect app on Android).
The problem seems to happen more frequently when many requests were done in a rapid succession. In such conditions, the peripheral stopped responding, and disconnected the clients after a while (10-20s). Once the clients reconnected, everything worked again.

I was able to solve the problem after one week of testing, by imposing a minimum delay between all characteristics notifications (in my case 50 ms) : now I don't have any disconnects and everything is stable.

Concept code:

            since_last = time.ticks_ms() - __last_notification_ms
            while _MIN_BLUETOOTH_DELAY_MS > since_last > 0 and not force:
                await asyncio.sleep_ms(_MIN_BLUETOOTH_DELAY_MS - since_last)
                # Need to recheck because another transmission may have happened during await period
                since_last = time.ticks_ms() - __last_notification_ms
            data = __get_notification_data()
            __status_characteristic.write(data, True)
            __last_notification_ms = time.ticks_ms()

I guess there is a minimum delay for physical transmission over bluetooth and handling by RTOS and if we push too many notifications in a short period, something goes wrong. I wrote this issue for informing other AIOBLE users and for suggesting defining a rate limit inside the aioble library, or a warning, because this was quite difficult to troubleshoot.

[jimmo]

Thanks @PBrunot

It would be worth getting to the root cause of why this causes disconnects, but yes this seems to be becoming a bit of a common gotcha. It might be possible to make the server-side char.write and char.notify detect this case and raise a "busy" exception. (i.e. it should be impossible to write Python code that causes a failure like this).

FWIW, using an L2CAP connection-oriented-channel for streaming data is a much better alternative to "streaming" notifications, because flow control is built-in.

[andrewleech]

Is L2CAP available on esp32?

[jimmo]

Is L2CAP available on esp32?

Ah, good point. Now that we've moved to synchronous events on the ESP32 there should be nothing blocking enabling it though, just needs testing.

[andrewleech]

Awesome, yeah I thought you'd done some work in this area recently :-D

[RantyDave]

I'm also getting these frequent (after 5-10 minutes) disconnections. I'm pushing a reasonable quantity of traffic, maybe 50-100 notifications/sec over 7 characteristics, using aioble, and receive an OSError exception (22 EINVAL). Throttling makes no difference. After some quick perusal of the code I see some references to a ring buffer - could it be that it's not dealing with this filling up correctly?