Disconnect dispatcher at cleanup

[Marenz]

fixes #145

[llucax]

I think this is a safer/more complete/future-proof approach. Didn't think too much about the order, but I think it should be fine.

[llucax]

As @ela-kotulska-frequenz pointed out this also misses disconnecting at stop. If we implement it there, it should come for free to __aexit__() as the default implementation of __aexit__() for BackgroundService is to call stop() (stop() calls self.cancel() + await self.wait()).

For now a "simple" solution when we need to do something async in cancel() (like the call to self._client.disconnect() would be to store the future, like:

def cancel():
    self._disconnecting_future = self._client.disconnect()

def wait():
    if self._disconnecting_future is not None:
        await self._disconnecting_future
    await super(self).wait()

Not super nice, as the disconnection won't really happen until the await, I think, but it should work for most cases. To do it right we would need to spawn a new task just to do the disconnection :-/

[Marenz]

Still missing the stop() part

[llucax]

Once you "fixed" cancel() and wait() there should be no need to override stop() too. In this case I guess you should move the actor stopping to cancel() too. And once you do that, maybe it makes sense to actually spawn a new task to do the cleanup. not sure, I guess you can also have a list of futures to wait for in the wait(), but you will not actually initiate the cleanup upon calling cancel(), only when calling wait().

[Marenz]

Hmm so basically just remove stop() now? The actor stopping is already in cancel (using .cancel)..

[llucax]

Then I guess so, I hope it works. Stopping/cleanup is pretty difficult/complicated at the moment, I hope we can improve things soon.

[llucax]

I praise the lord that this works 🙏 😆