forked from me-no-dev/AsyncTCP
Emil Muratov
7bfbb947d0
- do not let `_remove_events_with_arg()` call to block on queue pertubation, otherwise it could deadlock
- in any case do not block on adding LWIP_TCP_POLL event to the queue, it does not make much sense anyway
due to poll events are repetitive
- `_get_async_event()` will discard LWIP_TCP_POLL events when q gets filled up
this will work in combination with throttling poll events when adding to the queue
Poor designed apps and multiple parallel connections could flood the queue with interleaved poll events
that can't be properly throttled or coalesced. So we can discard it in the eviction task after coalescing
It will work in this way:
- queue is up to 1/4 full - all events are entering the queue and serviced
- queue is from 1/4 and up to 3/4 full - new poll events are throttled on enqueue with linear probability
- queue is from 3/4 up to full top - all new poll events are ignored on enqueue and existing poll events
already in the queue are discarded on eviction with linear probability giving away priority for all other
events to be serviced. It is expected that on a new poll timer connection polls could reenter the queue