I see that wait time granularity change from milli seconds to dozens of seconds, is there a reason? I think this will make the CI test cases spend much more time.

Overall, changing millis to seconds was due to avoid unstability of tests.

1. As far as I tried, the nginx process and the test process are not synced well in the ms order. Thus, when the test process sleeps with ngx.sleep(1-ms) to on-set ss.000 ms, the nginx process might be still at ss.998 ms.
   • Adding several jitters may relax this problem, but in my environment, it requires several hundred ms (WSL2+Docker)
2. Some of test takes several hundred ms to run, especially, iteration based-tests such as

   kong/spec/03-plugins/24-response-rate-limiting/04-access_spec.lua

   Line 483 in 4c3b9cd

   for i = n+1, ITERATIONS do

   must be completed 10 requests in 1secs, that is not stable.

In this change, wait(reset) function will only sleep if current time is at 50-59s or counter-reset is required. Since most of tests are separated into different limit counters, the CI will sleep only 1 to 3 times during tests.

That's a good effort on making the tests more stable. However, we are doing refinement on the test cases, which is not just waiting for a longer time, we will rely on some status endpoint to make the waiting more reasonable and waiting time as short as possible.
So could you just make the functional changes, and revert the test spec changes? So that I could approve and merge your code. Much appreciates on your effort!

OK, It's reverted. However, I'm not sure it passes the tests on CI environments. It never passes the tests in my local environment.

Thanks, I think you can leave the test name unchanged(keep the "flaky" in description), we will remove it after we make it stable.