Tracking issue for Fastify regressions

[mcollina]

Currently, Fastify tests are failing on v20 on CITGM.

https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/3224/nodes=fedora-latest-x64/testReport/junit/(root)/citgm/fastify_v4_22_2/

The tests that are failing to shut down the process are:

• https://github.com/fastify/fastify/blob/main/test/plugin.test.js
• https://github.com/fastify/fastify/blob/main/test/schema-examples.test.js
• https://github.com/fastify/fastify/blob/main/test/schema-validation.test.js
• https://github.com/fastify/fastify/blob/main/test/serial/logger.1.test.js
• https://github.com/fastify/fastify/blob/main/test/hooks.test.js
• https://github.com/fastify/fastify/blob/main/test/hooks-async.test.js

The test passes, but it times out when it's trying to shut down the process.

---

This might be due to changes in the keep-alive logic in v20. It's entirely possible that those changes unveiled some bugs elsewhere, but we are having a hard time investigating this because it happens only in CI.

The other option is that it's a bug in the testing harness (tap), or coverage collection.
This might be the case because we are seeing some flaky tests in CITGM too: https://github.com/nodejs/citgm/actions/runs/6143382259/job/16666721399#step:14:618.

[mcollina]

A few notes that we have already fixed a few problems that broke Fastify, so this is not relatively new:

• http: memory leak if a server is not closed #48604 (and its fix http: start connections checking interval on listen #48611).
• net: fix family autoselection timeout handling. #47860

[mcollina]

cc @ShogunPanda

[ShogunPanda]

I don't think this is related to http or net module but something is preventing Node to close. In couple of weeks once I finish traveling I'll look into this.

[ShogunPanda]

@mcollina This are caused by #49344, isn't it?

[mcollina]

Most of those issues are caused by #49344.

However even after disabling code coverage, CITGM is still failing: https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/3272/.

This might be partially due to some of our CITGM machines, but it's something that should be investigated.