Unpin Forgejo runners from kai-desktop-tower-wsl (remove obsolete SSH-era nodeSelector) #241

Closed
opened 2026-06-06 05:11:23 +00:00 by coilysiren · 0 comments
Owner

Both Forgejo Actions runner StatefulSets (deploy/forgejo-runner.yml, deploy/forgejo-runner-tap-writer.yml) carried nodeSelector: kubernetes.io/hostname: kai-desktop-tower-wsl, pinning the entire CI runner tier to a single WSL node on the desktop tower. When that node went NotReady (2d+, see coilysiren/inbox#65) all flight-deck CI went dark with no failover.

The pin originally existed to work around an SSH networking problem that is believed no longer to apply. Remove the nodeSelector from both so runners schedule on any Ready node (today: kai-server, the control-plane, which k3s leaves schedulable).

Follow-up: if SSH-dependent jobs break after this, the pin's original reason still holds and we hard-pin to kai-server instead. Always-on hardening for the WSL node tracked separately.

Both Forgejo Actions runner StatefulSets (`deploy/forgejo-runner.yml`, `deploy/forgejo-runner-tap-writer.yml`) carried `nodeSelector: kubernetes.io/hostname: kai-desktop-tower-wsl`, pinning the entire CI runner tier to a single WSL node on the desktop tower. When that node went `NotReady` (2d+, see coilysiren/inbox#65) all flight-deck CI went dark with no failover. The pin originally existed to work around an SSH networking problem that is believed no longer to apply. Remove the `nodeSelector` from both so runners schedule on any Ready node (today: `kai-server`, the control-plane, which k3s leaves schedulable). Follow-up: if SSH-dependent jobs break after this, the pin's original reason still holds and we hard-pin to `kai-server` instead. Always-on hardening for the WSL node tracked separately.
Sign in to join this conversation.
No description provided.