- Reduce PVC size from 20Gi to 5Gi
- Revert storageClass from longhorn to local-path due to storage constraints
- Keep Gitea on longhorn, other apps on local-path
Reduced CPU requests to prevent node-agent scheduling failures:
- code-server: 100m → 50m (actual usage ~5m)
- kubernetes-dashboard metrics-scraper: 100m → 50m (actual usage minimal)
This optimization frees up 100m CPU on worker-node-2, allowing
Velero node-agent (30m request) to be scheduled successfully.
Worker-node-2 CPU allocation before: 840m/1000m (84%)
Worker-node-2 CPU allocation after: 730m/1000m (73%)
- Added ingress.yaml for code-server, kubernetes-dashboard, and umami
- Updated kustomization files to include ingress resources
- Migrated from centralized ingress management to per-app architecture