[fix][R2-2346] Add terminationGracePeriodSeconds and preStop hook to agent-executor proxy - #354
Conversation
…agent-executor proxy Mirrors the lifecycle config from retool_development PR #82013, which only affects CI. The real values live here in retool-helm. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
| {{ toYaml $as.labels | indent 8 }} | ||
| {{- end }} | ||
| spec: | ||
| terminationGracePeriodSeconds: 30 |
There was a problem hiding this comment.
Add proxy shutdown rendering coverage
The agent-sandbox fixtures do not assert that these shutdown fields render only on the proxy deployment. Add a focused Helm rendering test covering both terminationGracePeriodSeconds and the preStop hook so future template changes cannot silently remove or misplace them.
Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!
|
| Filename | Overview |
|---|---|
| charts/retool/templates/deployment_agent_sandbox.yaml | Adds the proxy pod termination grace period and container preStop lifecycle hook at the intended deployment and container scopes. |
| charts/retool/Chart.yaml | Bumps the chart patch version to publish the template update. |
Reviews (2): Last reviewed commit: "[chore] bump chart version to 6.11.13" | Re-trigger Greptile
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Summary
terminationGracePeriodSeconds: 30to the agent-executor proxy pod speclifecycle.preStopsleep of 5s to the proxy containerWithout
terminationGracePeriodSeconds, kubelet's default 30s still applies but there's nopreStophook, so SIGTERM can arrive before kube-proxy finishes removing the endpoint — new connections still route to the terminating pod. ThepreStopsleep covers that gap, andterminationGracePeriodSeconds: 30ensures kubelet doesn't SIGKILL during the drain window (proxy has a 10sDRAIN_TIMEOUT_MS).Closes R2-2346.
Test Plan
helm templatediff shows only the two new fields on the proxy deployment, not the controller