[Bug] CI test job can hang for hours after tests stop producing output

[404-Page-Found]

Summary

The GitHub Actions test job can remain in npm test for roughly six hours and then be cancelled, even though the test suite itself can complete successfully in under a minute.

Observed behavior

The hang is not Ubuntu-specific. Recent CI history shows the same long-running npm test behavior on multiple runners, with successful runs on the other platforms.

Examples:

• Run 153: https://github.com/404-PF/commit-echo/actions/runs/35682640012 — Ubuntu cancelled after ~6 hours; macOS and Windows passed.
• Run 154: https://github.com/404-PF/commit-echo/actions/runs/35682696329 — Ubuntu cancelled after ~6 hours; macOS and Windows passed.
• Run 167: https://github.com/404-PF/commit-echo/actions/runs/35687668950 — Ubuntu cancelled after ~6 hours while this run otherwise had successful platform jobs.
• Run 35591288122 — macOS cancelled after ~6 hours while Ubuntu and Windows passed.
• Run 35556676049 — macOS and Ubuntu both cancelled after ~6 hours while Windows passed.
• Run 35542687267 — macOS cancelled after ~6 hours while Ubuntu and Windows passed.

For the cancelled jobs, the logs end with patterns such as:

The operation was canceled.
Cleaning up orphan processes
Terminate orphan process: pid (...) (npm test)
Terminate orphan process: pid (...) (sh)
Terminate orphan process: pid (...) (MainThread)
...

On macOS, the cleanup also shows orphan node processes being terminated.

The runner therefore still had the npm test process and child processes alive when the job was cancelled.

By comparison, a successful Ubuntu run completed:

tests 441
pass 441
fail 0
cancelled 0
skipped 0
duration_ms 38389.531839

Reproduction / characteristics

The CI workflow runs the same test command on Ubuntu, macOS, and Windows:

npm run build && node --test "tests/**/*.test.mjs"

The test suite contains multiple E2E tests that spawn the CLI as child Node processes and create local HTTP servers.

The hang appears to be intermittent and platform-dependent in manifestation, but cross-platform in scope: different runs can hang on different runners, while the other runners complete normally.

This points more toward a test-process lifecycle / resource cleanup issue (for example lingering child processes, servers, sockets, or timers) than an Ubuntu-only runner problem.

Expected behavior

After all tests have completed, the npm test process should terminate normally and the GitHub Actions job should finish promptly on every supported runner.

Impact

This can consume the full GitHub Actions job execution window before the run is cancelled, making CI appear hung and significantly delaying feedback.

Environment

• GitHub Actions
• ubuntu-latest, macos-latest, and windows-latest
• Node.js 24 (configured by the workflow)
• node:test
• 441-test suite on the observed successful run

[404-Page-Found]

Implemented in #324. The fix bounds E2E child shutdown after SIGINT (with a SIGKILL fallback), handles already-exited children, cleans up timed-out child processes, and awaits local HTTP server teardown while closing active connections. Added a regression test for a child that ignores SIGINT. The three-platform CI matrix is running now.