bug: thread starvation #10
Description
On some occasions, the discovery thread seems to be starved, causing the issues below:
- Since broadcasts are not being made, the timeout is reached and API automatically removes the device from the discovered set; a supporting evidence for this is running API with
DNET_P2P_UDP_PEER_TIMEOUT=999999, which fixes the issue of device being timed-out. - The timed-out device occasionally pops up again, and then disappears back etc. However, when the timeout is large (as told above) but the device is closed (which triggers a removal request via UDP) this removal request is missed as well, and the device is stuck as "discovered" even though it is closed.
This bug does not seem to be happening much on 2-device settings, but we shall try with 3-device setting more as it happens randomly there.