test: stabilize eventloopdelay test

[benjamingr]

This change makes the eventloopdelay test less flakey on some platforms
by replacing a blocking-wait with a busy-wait which means the eventloop
will always be positive increase.

This is a partial revert of #30787 that returns the test to the state it
had originally.

Example failure:

https://ci.nodejs.org/job/node-test-commit-linuxone/30460/nodes=rhel7-lto-s390x/testReport/junit/(root)/test/sequential_test_performance_eventloopdelay_/

AssertionError [ERR_ASSERTION]: The expression evaluated to a falsy value:

  assert(histogram.min > 0)

    at Timeout.spinAWhile [as _onTimeout] (/home/iojs/build/workspace/node-test-commit-linuxone/test/sequential/test-performance-eventloopdelay.js:65:7)
    at listOnTimeout (node:internal/timers:559:17)
    at processTimers (node:internal/timers:502:7) {
  generatedMessage: true,
  code: 'ERR_ASSERTION',
  actual: false,
  expected: true,
  operator: '=='
}

Node.js v18.0.0-pre

[Trott]

@nodejs/testing, especially @cjihrig

[Trott]

LGTM if CI is green

[targos]

should we do a stress test?

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/42188/

[cjihrig]

I'm not so sure about this fix. It's good if we can get the test passing, but I think either Node's event loop monitoring needs to be able to handle this scenario, or something else in this test is incorrect.

should we do a stress test?

I think so, yes.

[benjamingr]

I'm not so sure about this fix.

Me neither - if there is a way to synthetically get the histogram to report event loop delay that would be preferable.

[cjihrig]

I'm curious how long this test has been flaky for. I've definitely noticed it lately, but the change being partially reverted here landed over two years ago. Maybe I just missed the failures in the past, but is it possible that a recent change introduced the flakiness?

[richardlau]

I'm curious how long this test has been flaky for. I've definitely noticed it lately, but the change being partially reverted here landed over two years ago. Maybe I just missed the failures in the past, but is it possible that a recent change introduced the flakiness?

It seemed to start appearing around 22 December 2021 #41286 (comment).

[cjihrig]

It seemed to start appearing around 22 December 2021

23637e9 landed three days prior to that and could be worth investigating.

[richardlau]

It seemed to start appearing around 22 December 2021

23637e9 landed three days prior to that and could be worth investigating.

#41286 (comment)

Walking backwards from commits landed on 22 December 2021, I've started two stress runs around 23637e9:

665b404: https://ci.nodejs.org/view/Stress/job/node-stress-single-test/306/nodes=rhel7-s390x/ (✔️ no failures in 1000 runs) 23637e9: https://ci.nodejs.org/view/Stress/job/node-stress-single-test/307/nodes=rhel7-s390x/ (❌ 100 failures in 1000 runs)

#41286 (comment) added some debug and it looks like in the failing case we end up with histogram.min being 0 which trips an assertion in the test. And at that point I don't know if it's the test or the implementation that's at fault.

dismissing my review while this is discussed

[benjamingr]

Adding 'blocked' so this doesn't land before other avenues have been explored.