src: unload native addons when the environment is destroyed

[gabrielschulhof]

This is an alternative to #23319
which attaches the loaded addons to the environment and closes them
when the environment is destroyed.

It includes the test/addons/worker-addon,
test/addons/hello-world/test-worker.js and the documentation update from the original PR written
by @addaleax in a separate commit.

Checklist

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

[nodejs-github-bot]

@gabrielschulhof build started: https://ci.nodejs.org/blue/organizations/jenkins/node-test-pull-request-lite-pipeline/detail/node-test-pull-request-lite-pipeline/1848/pipeline

[gabrielschulhof]

@joyeecheung I have addressed your review comments.

@bnoordhuis I have addressed your review comments. I put the body of DLib into a new node_binding-inl.h' which is included from the bottom of node_binding.h`.

[addaleax]

Thank you! 💙

[gabrielschulhof]

CI: https://ci.nodejs.org/job/node-test-pull-request/19417/

[gabrielschulhof]

Failure looks legit.

[gabrielschulhof]

Resume, just to see if it's 100% reproducible:

https://ci.nodejs.org/job/node-test-pull-request/19427/

[addaleax]

Resume CI: https://ci.nodejs.org/job/node-test-pull-request/19569/

[Trott]

Fresh CI rather than Resume Build so that it pulls in the fact that a bunch of tests are currently marked flaky: https://ci.nodejs.org/job/node-test-pull-request/19572/

[Trott]

Any chance that this AIX failure is related?

16:58:17 not ok 2208 addons/at-exit/test16:58:17 ---16:58:17 duration_ms: 0.72316:58:17 severity: crashed16:58:17 exitcode: -416:58:17 stack: |-16:58:17 ...

I'm guessing not, but you tell me. ¯\(ツ)/¯

In the meantime, Resume Build CI: https://ci.nodejs.org/job/node-test-pull-request/19573/

[addaleax]

@Trott I’m not sure whether we should wait for the outcome of the resume build (run stress tests if it comes back green?), but before/after stress tests with a hundred runs or so might be helpful?

[Trott]

before/after stress tests with a hundred runs or so might be helpful?

I guess if we're going to tie up an AIX machine for a few hours doing a stress test, this is the time of day/week to do it, so here we go:

Stress test on master: https://ci.nodejs.org/job/node-stress-single-test/2128/

Stress test on this PR: https://ci.nodejs.org/job/node-stress-single-test/2129/

[Trott]

CI Resume Build for AIX showed the same addons failure.

https://ci.nodejs.org/job/node-test-commit-aix/19762/nodes=aix61-ppc64/console

18:59:40 not ok 2208 addons/at-exit/test18:59:40 ---18:59:40 duration_ms: 0.91818:59:40 severity: crashed18:59:40 exitcode: -418:59:40 stack: |-18:59:40 ...

Still awaiting stress test for this PR to start, but stress test on master is clean so far after almost 900 runs. (It's set to do 1000.)

[Trott]

Stress test is crashing on addons/at-exit/test 100% of the time after 400+ runs on AIX. Impressed/weirded-out that this issue is AIX only. But I guess it needs to be sorted out before landing...

20:22:34 not ok 1 addons/at-exit/test20:22:34 ---20:22:34 duration_ms: 0.57320:22:34 severity: crashed20:22:34 exitcode: -420:22:34 stack: |-20:22:34 ...

[MylesBorins]

This change broke the node-report module

It seems that is was somehow related to nan?

nodejs/node-report#116

Fix landed in nan and awaiting a release nodejs/nan#831

Would someone be willing to open a backport so we can track the progress?

[addaleax]

This change broke the node-report module

@MylesBorins This seems very unlikely … can you double-check and maybe share how you tested this?

Fix landed in nan and awaiting a release nodejs/nan#831

That PR was opened before this one landed and the contents seem very much unrelated…

[BridgeAR]

This has been backported and released in v11.7.0 but it turned out that this actually breaks node-report. Therefore I am going to remove the backport again and release v11.7.1 without this.
Please have another look at this so it does not cause any issues with node-report.

[addaleax]

I’m sorry that I did not add a test case for the issue we’re seeing with node-report when I wrote the original PR at #23319 – I’ve opened #25577 to address it.

(This does not mean that this is not a bug in node-report, btw – it is. But there’s a lot of addons that run into similar issues with this PR, I assume.)

[richardlau]

(This does not mean that this is not a bug in node-report, btw – it is. But there’s a lot of addons that run into similar issues with this PR, I assume.)

If it's something we can improve, please report/submit a PR over in https://github.com/nodejs/node-report. Note that one of the issues with addons and master is that nan requires an update to work with the version of V8 in master (AFAIK nodejs/nan#831 has yet to be released) which means addons currently fail in citgm, which is obviously masking stuff like this.

[addaleax]

@richardlau I can try to do that, but given the work on integrating it into core in #22712, I don’t think I’d personally want to spend much time on the npm module… the first problem I encountered here is that nodereport_trigger_async is never removed from the event loop (e.g. in an Environment cleanup hook). Generally, in #22712 we had (and I still partly have) a hard time figuring out what lifetime management objects from node-report should have and how much global state is actually desirable and managable.

(Also, it’s absolutely understandable if you’re not interested in going through the nearly one thousand comments in #22712, but there’s a lot in there that applies to the npm module as well, including this issue iirc.)