Set default wait time to 0 (no wait) to enqueue tcp events

[mathieucarbou]

No description provided.

[vortigont]

do not think that it's a good idea here, those functions here are called by LWIP and it is more critical to try best not to miss those events by default. It could block and wait while asynctcp's task is running on evicting the events from a queue, so there should not be a deadlock condition. The exception cases for poll events are called with 0 timeout.
We could probably set some reasonable timeout of several seconds here instead of portMAX_DELAY in case watchdog is disabled for asynctcp task.

[mathieucarbou]

@vortigont : that is something we played around with @me-no-dev yesterday, with also the log pr (#24). We were able to see the events being discarded on a full queue of 64 and by adding a delay of 3 sec in the onBody to slow down request processing (data tcp event).

autocannon -c 32 -w 32 -a 96 -t 30 --renderStatusCodes -m POST -H "Content-Type: application/json" -b '{"game": "test"}' http://192.168.4.1/game_log

and doing some normal curl get at the same time.

Otherwise in normal case, all works fine, with a lot of concurrent requests also.

The idea being to remove any wait time and if queue is full, see when it happens and if it happens, maybe abort requests.
What we saw is that this is quite hard to produce a case that will exhaust the queue , even with size 64.

[me-no-dev]

those functions here are called by LWIP and it is more critical to try best not to miss those events by default.

at the same time we do not want to block LwIP. There are other ways to handle the missed events. You will block LwIP because another client can't push it's event and thus also prevent the running one (that is halting the thread for too long) to also not be able to complete