Repository navigation
investigate flaky test-timers-promisified #37395
Description
Activity
Here's the code that's failing. In this code,
setInterval()isrequire('timers/promises).setInterval().{ // Check that the timing is correct let pre = false; let post = false; setPromiseTimeout(1).then(() => pre = true); const iterable = setInterval(() => {}, 2); const iterator = iterable[Symbol.asyncIterator](); iterator.next().then(common.mustCall(() => { assert.ok(pre, 'interval ran too early'); assert.ok(!post, 'interval ran too late'); return iterator.next(); })).then(common.mustCall(() => { assert.ok(post, 'second interval ran too early'); return iterator.return(); })); setPromiseTimeout(3).then(() => post = true); }
I think we established already that Node makes no timing guarantees (at the moment) other than:
The only guarantee is that the timeout will not execute sooner than the declared timeout interval.
(I think this behaviour is very counterintuitive, but I also acknowledge I was just not aware of it for a pretty long time - and fixing this is hard)
Since the order of timers is not guaranteed - I would write these tests using mocks (mock out Timeout or something a-la @sinon/fake-timers).
I think any fix we will make to these sort of tests is inherently flakey if it relies on a guarantee timers don't make.
I'm not sure if this is the other bug, I think that there's timer "drift" happening here. Maybe relying on 1-2-3 ms timers wasn't such a great idea.
I agree with @benjamingr that it's difficult to fix this correctly, although with the current timer implementation I think that without the second check ('second interval ran too early') this test shouldn't fail.
FWIW I’ve had a bad run of consecutive failures so I ran a stress test to check this is still flaky and hasn’t become consistently failing. It is still flaky, failing 436 times out of 1000: https://ci.nodejs.org/job/node-stress-single-test/nodes=rhel7-s390x/217/
FWIW I’ve had a bad run of consecutive failures so I ran a stress test to check this is still flaky and hasn’t become consistently failing. It is still flaky, failing 436 times out of 1000: https://ci.nodejs.org/job/node-stress-single-test/nodes=rhel7-s390x/217/
@richardlau #37425 is an attempt to fix this if you want to give that a review.
Reacted by Richard Lau- added a commit that references this issue
on Feb 25, 2021 - added a commit that references this issue
on Feb 28, 2021
https://ci.nodejs.org/job/node-test-commit-linuxone/25660/nodes=rhel7-s390x/console