Perform busy waiting when the wait duration is less than MIN_SLEEP_DURATION

[byeonggiljun]

This fixes #513.

Add busy waiting when the wait duration is less than MIN_SLEEP_DURATION to prevent the reactor from advancing to a logical time greater than the current physical time.

[byeonggiljun]

@erlingrj I ended up creating a new PR instead of #403 because that PR is WIP. I have two questions for you.

1. What do you prefer between removing MIN_SLEEP_DURATION and adding busy waiting?
2. If we're going to add busy waiting, should we adjust the value of MIN_SLEEP_DURATION? In other words, is busy waiting for at most 10 usec too much?

[erlingrj]

Hmm, I think doing busy-waiting is fine, but I think it should be viewed as a real-time optimization, the purpose of it is to reduce lag. That might not be always what is wanted, however, we only do this a short period of time anyways. Another argument is that if the users are writing real-time programs they have to configure the underlying platform to achieve it. If they are using Linux, this means enabling PREEMPT_RT and using a real-time scheduling policy for the process that runs the LF program.

[byeonggiljun]

Surprisingly, either removing MIN_SLEEP_DURATION (CI-test#184b55f) or adding busy waiting (CI-test#fd7bd22) in wait_until led to test failures in window-latest while the error didn't occur when I revert the change.

[erlingrj]

My guess here is that we try going to sleep for a negative duration and the Windows platform implementation does not support that. I suggest that we verify that the sleep duration is greater than zero. (Basically setting MIN_SLEEP_DURATION to zero).

The first failures when adding the busy-wait seems to be different than the ones when you removed it altogether:

*** java.lang.instrument ASSERTION FAILED ***: "!errorOutstanding" with message can't create name string at s\src\java.instrument\share\native\libinstrument\JPLISAgent.c line: 830
*** java.lang.instrument ASSERTION FAILED ***: "!errorOutstanding" with message can't create name string at s\src\java.instrument\share\native\libinstrument\JPLISAgent.c line: 830

Could be something else completely

[erlingrj]

Very wierd Java errors, almost seems like its something with the network?

[byeonggiljun]

Yes, it is very weird. RomanIakovlev/timeshape#1 Maybe the busy waiting causes a stack or heap overflow?

I suggest that we verify that the sleep duration is greater than zero. (Basically setting MIN_SLEEP_DURATION to zero).

Anyway, I'll add your suggestion. Thanks!

[byeonggiljun]

Hmm... I found that there's already the code to handle negative time delays below in lf_windows_support.c. Let's first see whether the test passes now.

reactor-c/low_level_platform/impl/src/lf_windows_support.c

Lines 293 to 299 in 856e98a

	int _lf_cond_timedwait(lf_cond_t* cond, instant_t wakeup_time) {
	// Convert the absolute time to a relative time.
	interval_t wait_duration = wakeup_time - lf_time_physical();
	if (wait_duration <= 0) {
	// physical time has already caught up sufficiently and we do not need to wait anymore
	return 0;
	}

[byeonggiljun]

Why don't we set MIN_SLEEP_DURATION to 0 for now to avoid the flaky error? We can improve the design for real-time optimization in other PRs.

[erlingrj]

LGTM