Remove double check for keepalive

[Dirk007]

(at least on MacOS) the call_later-target is executed some milliseconds before the set timeout and therefore the ping will not be sent in a valid time for the broker.

Explanation:
For example: if keepalive is set to 2 seconds timeout, the call will eventually be made after 1.91(+-) seconds. Thus, the first call comes in after 1.91 seconds and the comparison if time.monotonic() - self._last_data_in >= self._keepalive will evaluate to false. The second call will be made after 3.92 seconds - but its already too late to send the ping to the broker then, because the maximum tolerance on broker-side is 1.5 * keepalive.

As the timeout for call_later is already equal to keepalive, we can simply remove the double check inside the _keep_connection to have everything worky.

[Lenka42]

@Dirk007 thanks for the PR
the idea is, we do not need to send ping request if there is any other packages exchange between client and broket, so self._last_data_in is set in put_package method
But if such an issue occurs in MacOS, maybe check with some kind of coefficient could help? self._last_data_in >= 0.8 * self._keepalive maybe?

[Dirk007]

I understand. a coefficient on its own won't help as the second call will be made too late already in some cases I think. How about changing self._keep_connection_callback = asyncio.get_event_loop().call_later(self._keepalive, self._keep_connection) to self._keep_connection_callback = asyncio.get_event_loop().call_later(self._keepalive / 2, self._keep_connection) together with the coefficient?

[Lenka42]

@Dirk007 I see. Yes, sounds good to me.

[Dirk007]

ping :D

[Lenka42]

@Dirk007 sorry for delay 🙂
check version 0.6.3

[wdv4758h]

I'm literally digging the MQTT connection today. My broker keep saying the peer has keepalive timeout, but I do set the keepalive I want. After some checking, the version I'm using (0.6.2) will have chance to not sending ping request due to a small gap for condition. I see the time diff is like 39.993 when keepalive is 40.

I was about to create a PR, then notice you guys already fix it couple days ago!!!
My issue is fixed after upgrading to 0.6.3, thanks!

[wdv4758h]

Also this fix not only effects MacOS, my issue was happend on GKE with Debian container (python:3.8.2-slim-buster).