Investigate flaky parallel/test-async-hooks-http-parser-destroy on OS X

[sam-github]

Seems to be failing regularly:

• https://ci.nodejs.org/job/node-test-commit-osx/nodes=osx1011/27247/
• https://ci.nodejs.org/job/node-test-commit-osx/nodes=osx1011/27267/
• https://ci.nodejs.org/job/node-test-commit-osx/nodes=osx1011/27258/

For all above, failures are identical:

timeout
assert.js:89
  throw new AssertionError(obj);
  ^

AssertionError [ERR_ASSERTION]: Expected values to be strictly deep-equal:
+ actual - expected ... Lines skipped

  [
-   1288,
    156,

Perhaps it should be marked flaky? I'm not sure how common it has to be to deserve the marking.

• Version:
• Platform:
• Subsystem:

[Flarna]

The test duration in failed cases seems to be significant longer as in the pass case. Not sure if duration_ms: 120.95 is really in milli seconds - as a wild guess I would say it's in sec and therefore 2 minutes which would match the default http timeout used by node as far as I know.

I have no access to OS X so hard for me to work on this.

[Trott]

The test duration in failed cases seems to be significant longer as in the pass case. Not sure if duration_ms: 120.95 is really in milli seconds - as a wild guess I would say it's in sec and therefore 2 minutes which would match the default http timeout used by node as far as I know.

I have no access to OS X so hard for me to work on this.

duraion_ms is unfortunate, but it's a tap thing so we can't/shouldn't change it. It is duration in seconds.

It fails at 120 seconds because that's the timeout duration in the test runner.

So it gets caught in a deadlock or infinite loop or waiting for an event that never actually fires or (unlikely but possible) maybe just takes too darned long to run and would complete if it had more time or...

[Flarna]

Hmm, maybe add some more counters, e.g. one for HTTPINCOMINGMESSAGE and HTTPCLIENTREQUEST, one for client responses and then assert on them first before the diff of the list to get some hint what is hanging.

Or hope that some OS X users gives a try to reproduce this locally and debug.

[sam-github]

@Flarna Are you willing to PR that change, ☝️ ? It would be very helpful if you have time to start looking into this.

[Flarna]

Sure, will take a look in a free slot during next days.

created #28253

[Flarna]

Is this still valid? PR is open since quite a while now...

[lpinca]

I can't reproduce locally but I'm on macOS 10.14 not 10.11.

[Flarna]

Has #28253 fixed the flaky parallel/test-async-hooks-http-parser-destroy on OS X? Don't know how to query CI statistics.

[Trott]

Has #28253 fixed the flaky parallel/test-async-hooks-http-parser-destroy on OS X? Don't know how to query CI statistics.

Used ncu-ci from node-core-utils to check and couldn't find any async-hooks test failures other than test-statwatcher.js. So I think we can conclude that this isn't a problem and plan to re-open if that's wrong and someone notices it poppingup again.

Thanks, @Flarna!