backends/bluezdbus: use an a new reactor for each client

[mickael9]

This makes things much cleaner :

• The default asyncio event loop is not modified
• No default twisted reactor is installed (a new one will be created every time)
• The reactor will use the provided event loop

Related to: #93 #125

[hbldh]

This. This is how it should have been from the start. Thank you very much, this I will merge immediately!

[mickael9]

Thanks. I haven't got much experience with asyncio and multiple event loops. I'm still wondering exactly how useful the "loop" argument is, since we're already in an async function, there must already be a running event loop, and I don't think you can await on tasks from another event loop

[hbldh]

If no loop is specified, the default loop will be used. There are multiple people you have asked about sending in non-default loops, This enables running multiple BleakClients in the same process, which is highly desireable.

[JoeHut]

actually this is problematic... It looks like the reactor leaks file descriptors (2-3 for each connected client). So if you are connecting and disconnecting devices continuously you will run out of file descriptors and crash. I am not really familiar with twisted, but as I understand it, there are wakers installed represented by file descriptors and neither closed by calling stop(), disconnectAll() or removeAll() nor affected by the garbage collector... So I guess the only way is to try and instantiate as few reactors as possible and pass them. If there is someone who can explain me more, I would be happy to help

[JoeHut]

The only reference I found is this really old ticket: https://twistedmatrix.com/trac/ticket/3063
But I don't quite understand whether it was resolved or what the general development state of twisted is