test: add simple extra debugging to flaky test (async-wrap-uncaughtexception)

[addaleax]

Refs: #16210

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• commit message follows commit guidelines

Affected core subsystem(s)

test

just fyi, /cc @jasnell@refack@Trott

[addaleax]

CI: https://ci.nodejs.org/job/node-test-commit/13737/

[addaleax]

Doesn’t seem to do the trick. The “good” news here seems to be that this fails on linuxone as well, which should get us better stress testing I guess…?

So here’s a stress test to get an idea of the linuxone failure rate: https://ci.nodejs.org/job/node-stress-single-test/1540/nodes=rhel72-s390x/

[addaleax]

Okay, it’s failing 100 % of the time on linuxone now. Awesome? I guess? Let’s run some things and keep track in this comment so I don’t forget what I already tried out:

• Stress test, with hooks.disable() removed: https://ci.nodejs.org/job/node-stress-single-test/1541/nodes=rhel72-s390x/ (Still failing – yay)
• Stress test, with debugging to make sure one beforeExit call is resurrecting the loop: https://ci.nodejs.org/job/node-stress-single-test/1542/nodes=rhel72-s390x/
• Stress test, yet more debugging (listing all active handles): https://ci.nodejs.org/job/node-stress-single-test/1543/nodes=rhel72-s390x/ – Okay, yay! Something inside beforeExit is creating a new, refed, active uv_async_t. Awesome.
• Maybe it’s MarkGarbageCollectionEnd()? Let’s find out: https://ci.nodejs.org/job/node-stress-single-test/1544/nodes=rhel72-s390x/
• A fix, at last: https://ci.nodejs.org/job/node-stress-single-test/1545/nodes=rhel72-s390x

I’ll be closing this and opening a PR with what I think is a reasonable fix. Yay!

[refack]

So it's a combination of the pref_hooks added in 8.5.0 & V8 6.1 updated in 8.7.0... (my assumption since locally I can repro 100% with the Windows.64 official binary, and couldn't repro with 8.6.0)

[addaleax]

@refack Yup, I assume that’s what’s happening. My suggestion for a fix is in #16758, but if you think something else resolves the problem better, I’ll gladly take it.