Flaky FiberRefSpec Test

[adamgfraser]

@mlangc Can you take a look?

[info] - FiberRefSpec
[info]   - Create a new FiberRef with a specified value and check if:
[info]     - the value of the winner is inherited when racing two ZIOs with raceAll
[info]       (update2,update1) did not satisfy equalTo((update1,update1))

https://circleci.com/gh/zio/zio/32289?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link

[mlangc]

I'll look into it today or tomorrow evening.

[mlangc]

@adamgfraser can you reproduce this locally? I've tried really hard (including running the test thousands of times sequentially and in parallel - both while my workstation was idle and under high load using different JVMs) but I could not reproduce the failure.

[adamgfraser]

@mlangc I can't. It definitely happened per the CI logs and I double checked that this was on a branch with looseTimeAndCpu implemented with Live[Clock] but I am also unable to reproduce.

[mlangc]

Hmm.... I've experimented with different numbers in Schedule.recurs in looseTimeAndCpu, and the smallest number where I could reproduce the issue using nonFlaky(10000) was 32. That being said, I see three possible solutions:

1. Bump the value of recurs to 200 or even higher, and see if that helps.
2. Let the looser wait on a promise, that does not get succeeded before the race is over, as discussed on discord (aka cheating).
3. Document that the test is flaky by nature, as we cannot control the operation system process scheduler, and leave everything as is.