Fix Chromium supervisor stop signaling

[rgarcia]

Summary

• Signal the full Chromium process group when supervisord stops Chromium in both headful and headless images.
• Change stopwaitsecs from 0 to 1 so supervisord has a short window to observe SIGCHLD after the initial SIGKILL instead of immediately entering its final-kill loop.

Why

We observed supervisorctl stop chromium producing thousands of repeated killing 'chromium' (...) with SIGKILL log lines before reporting stopped.

Supervisor's STOPPING transition loop explains this behavior: when the process is STOPPING and delay - now <= 0, it logs the final SIGKILL warning and calls kill(SIGKILL) again. With stopwaitsecs=0, that deadline is expired immediately, so supervisord can repeat this until SIGCHLD processing catches up:

• Tight-loop SIGKILL path: https://github.com/Supervisor/supervisor/blob/abc60468ea4b78c446cf3a194f6fccb83d90f670/supervisor/process.py#L709-L718
• stopwaitsecs, stopasgroup, and killasgroup config parsing: https://github.com/Supervisor/supervisor/blob/abc60468ea4b78c446cf3a194f6fccb83d90f670/supervisor/options.py#L911-L914
• Process-group signaling sends to -pid when group signaling is enabled: https://github.com/Supervisor/supervisor/blob/abc60468ea4b78c446cf3a194f6fccb83d90f670/supervisor/process.py#L432-L467

stopasgroup=true / killasgroup=true also ensures Chromium renderer, utility, zygote, and browser descendants in the supervised process group receive the stop signal, rather than only signaling the direct supervised runuser process.

Test plan

• Rebuilt headful image locally: DOCKER_BUILDKIT=1 docker build -f images/chromium-headful/Dockerfile -t kernel-headful-test .
• Ran 40 /process/exec stop/start iterations against the rebuilt headful container: max stop latency was 181ms and no repeated killing 'chromium' ... SIGKILL lines appeared.
• Opened 80 Chromium pages via Playwright CDP, then stopped Chromium through /process/exec: stop completed in 229ms and live Chromium descendants were gone afterward.

Made with Cursor

---

Note

Low Risk
Container runtime/supervisor tuning only; no app auth, data, or API surface changes, with minimal behavioral impact beyond cleaner Chromium shutdown.

Overview
Updates supervisord’s Chromium program config in headful and headless images so stops target the whole process group and don’t spin on repeated SIGKILLs.

Both chromium.conf files now set stopasgroup=true and killasgroup=true, so stop/kill signals reach Chromium’s child processes (not only the launcher). stopwaitsecs changes from 0 to 1, giving supervisord a brief window after the initial KILL before its STOPPING loop re-issues SIGKILL—addressing noisy supervisorctl stop chromium logs and long stop times.

^{Reviewed by Cursor Bugbot for commit 4cb2357. Bugbot is set up for automated code reviews on this repo. Configure here.}

[firetiger-agent]

Firetiger deploy monitoring skipped

This PR didn't match the auto-monitor filter configured on your GitHub connection:

Any PR that changes the kernel API. Monitor changes to API endpoints (packages/api/cmd/api/) and Temporal workflows (packages/api/lib/temporal) in the kernel repo

Reason: This PR modifies Chromium supervisor configuration in Docker images, not the kernel API endpoints or Temporal workflows specified in the filter.

To monitor this PR anyway, reply with @firetiger monitor this.