GeezerVison

queued a stress test - https://ci.nodejs.org/job/node-stress-single-test/557/

I'm not sure I understand the difference. Why using a for loop reduces the flakiness?

Currently it is only one test that attempts to allocate 8GB of memory, the purpose of the for loop is to make them into 4 separate tests. I think it could have 2 benefits:

1. we could be able to to know which one timed out
2. potentially give more chances to gc

Actually - now I see my stress tests start to fail on ubuntu2204-arm64 with the same timeout, I think you might be right @lpinca 🤔 it wouldn't do anything useful, do you think this also relates to the deadlock? Maybe run a major gc is a better way to deflake the test 😅 wdyt?

Do you think this also relates to the deadlock?

Yes, probably.

I've attempted to add major gc after each allocation and queued another stress test - https://ci.nodejs.org/job/node-stress-single-test/558/. This test seems having a way too high failure rate.

Might be worth checking if distributing each of these into separate test files improves things

CI: https://ci.nodejs.org/job/node-test-pull-request/66133/

The stress tests seem ok, the CI failures relate to other flaky tests. Are we happy to do major gc or better off separate them into different files?

Any further actionable item should I take?

• Continue with landing this PR
• Separate the tests into multiple files

Happy to take either path 👍

I would try to separate into multiple files without manually triggering GC and see what happens. Manually calling the GC is not a fix.

what are the practical ways to verify if things've improved?

• have a few CI runs
• 1000 stress tests

Any other suggestions?

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 90.22%. Comparing base (1540fc6) to head (c76d93b).
Report is 242 commits behind head on main.

@@            Coverage Diff             @@
##             main   #57789      +/-   ##
==========================================
- Coverage   90.23%   90.22%   -0.02%     
==========================================
  Files         630      630              
  Lines      185288   185518     +230     
  Branches    36344    36380      +36     
==========================================
+ Hits       167203   167387     +184     
- Misses      11006    11027      +21     
- Partials     7079     7104      +25     

CI: https://ci.nodejs.org/job/node-test-pull-request/66169/

Landed in 795dd8e

Tests are failing on GHA with this change (https://github.com/nodejs/node/actions/runs/14865929007/job/41742567775), we'd need a manual backport PR if we want to port this change.

I will look into the backport

Hi @aduh95 I took a look and realised this PR was fixing the flaky test introduced in #51821 which 51821 is a semver-major PRs that contain breaking changes and should be released in the next major version. so, I think I should've added don't land on labels