test(ci): probe — drop mocha --exit on Windows backend tests

[JohnMcLear]

Summary

One-line workflow probe on a separate branch from #7855: remove the --exit flag from pnpm test on both Windows backend matrices.

Why

The current workflow has been invoking pnpm test -- --exit on Windows since PR #7762, with the comment:

--exit forces process.exit(failures) after the suite completes, closing the post-suite event-loop drain window where Windows + Node 24 hard-kills the process.

That's a direct admission of a known post-suite kill bug on Windows + Node 24, and --exit was added as a mitigation that hides it. The current silent-ELIFECYCLE flake fires mid-suite, but --exit is still in play after the dying test and may also be masking earlier signal.

Drop it on a separate probe branch. Either:

• Tests still complete cleanly via natural drain → the original kill from feat(metrics): 3 Prometheus counters for scaling dive (#7756) #7762 is no longer reproducible on current Node 24.x and we've been carrying the workaround for nothing.
• A fresh hard-kill appears at post-suite drain → second data point distinct from the mid-suite flake, comparable to it via the merged diagnostic infra (test(ci): heartbeat + running-test pointer to debug the silent backend-test ELIFECYCLE #7838 + test(ci): schedule a mid-test snapshot 150ms after every beforeEach #7842 + test(ci): OS-level sidecar watcher for the Windows silent ELIFECYCLE #7846).
• The mid-suite flake fires → does mocha emit additional output that --exit's process.exit() was suppressing?

Any of these is data we don't currently have.

Test plan

• Linux ± plugins must pass (no change there)
• Windows backend matrices: fail, pass, or unchanged — all three outcomes are informative
• Run alongside test(ci): probe — disable Defender + capture Event Log + fix tasklist sidecar #7855 in parallel to maximise data per CI window

🤖 Generated with Claude Code

[qodo-code-review]

Qodo reviews are paused for this user.

Troubleshooting steps vary by plan Learn more →

On a Teams plan?
Reviews resume once this user has a paid seat and their Git account is linked in Qodo.
Link Git account →

Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center?
These require an Enterprise plan - Contact us
Contact us →

[qodo-free-for-open-source-projects]

Review Summary by Qodo

Probe: remove mocha --exit flag from Windows backend tests

🧪 Tests

Walkthroughs

Description

• Remove mocha --exit flag from Windows backend test matrices
• Probe to diagnose post-suite process hard-kill on Windows + Node 24
• Investigate whether original mitigation from PR #7762 still needed
• Enable detection of additional mocha error output without early exit

Diagram

flowchart LR
  A["Windows Backend Tests"] -->|Remove --exit flag| B["Natural Process Drain"]
  B -->|Outcome 1| C["Clean exit - bug resolved"]
  B -->|Outcome 2| D["Post-suite hard-kill - new data point"]
  B -->|Outcome 3| E["Mid-suite flake - unchanged behavior"]

Loading

File Changes

1. .github/workflows/backend-tests.yml 🧪 Tests +2/-2

Remove --exit flag from Windows backend test runs

• Remove --exit flag from pnpm test command on line 259 (first Windows backend matrix)
• Remove --exit flag from pnpm test command on line 393 (second Windows backend matrix)
• Preserve existing diagnostic infrastructure and watcher process management
• Keep explanatory comments about the original mitigation rationale

.github/workflows/backend-tests.yml

---

[qodo-free-for-open-source-projects]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0)

1. Removed mitigation without guard ✓ Resolved 🐞 Bug ☼ Reliability

Description

The workflow comment says --exit was used as a Windows-only mitigation for a post-suite
natural-drain window where Windows + Node 24 hard-kills the process, but the Windows backend test
steps no longer pass --exit. This both risks reintroducing post-suite termination/hang behavior
(potentially preventing watcher cleanup) and will fail the vitest spec that asserts the workflow
contains exactly two pnpm test -- --exit invocations.

Code

.github/workflows/backend-tests.yml[395]

Evidence

In .github/workflows/backend-tests.yml, the inline documentation still describes --exit as the
Windows mitigation for the post-suite drain hard-kill and the steps rely on cleanup that only runs
after pnpm test returns, but the actual Windows commands now run pnpm test without --exit.
Separately, src/tests/backend-new/specs/backend-tests-flake-mitigation.test.ts matches the literal
string pnpm test -- --exit and asserts it appears exactly twice in the workflow, so with the
workflow no longer containing that string the match count becomes 0 and the test will fail, breaking
CI for runs that execute pnpm run test:vitest.

.github/workflows/backend-tests.yml[255-266]
.github/workflows/backend-tests.yml[389-400]
src/tests/backend-new/specs/backend-tests-flake-mitigation.test.ts[48-54]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The Windows backend test steps in `.github/workflows/backend-tests.yml` removed `--exit` even though the workflow comment documents it as a Windows+Node24 mitigation for post-suite drain/hard-kill behavior, and the steps rely on post-command cleanup for a background watcher. This change also breaks a vitest spec that asserts the workflow contains exactly two `pnpm test -- --exit` invocations.
## Issue Context
The workflow starts a background watcher loop and then runs `pnpm test`; cleanup happens only after `pnpm test` returns, so hangs/early termination can prevent watcher cleanup. A vitest assertion in `src/tests/backend-new/specs/backend-tests-flake-mitigation.test.ts` is implemented by searching the workflow file for the exact string `pnpm test -- --exit` and expecting two occurrences, which no longer exist.
## Fix Focus Areas
- .github/workflows/backend-tests.yml[241-266]
- .github/workflows/backend-tests.yml[375-400]
- .github/workflows/backend-tests.yml[255-266]
- .github/workflows/backend-tests.yml[389-400]
- src/tests/backend-new/specs/backend-tests-flake-mitigation.test.ts[48-58]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

[JohnMcLear]

Closing — this was a diagnostic probe while hunting the Windows backend silent-ELIFECYCLE flake. Root cause found and fixed: in-process handler gate + Windows pinned to Node 24.16.0 (#7866), investigation scaffolding removed (#7868), and the underlying libuv TCP-connect stack overrun filed upstream as nodejs/node#63620. Branch deleted.