Previous approaches (port-forward to Service/Pod) failed with 'connection
refused' — the runner cannot tunnel to pod IPs through the API server.
Switch to LoadBalancer service type:
- After rollout, poll kubectl get svc for status.loadBalancer.ingress[0].ip
- Once assigned, poll http://<lb-ip>:80 until reachable
- Write HEADLAMP_URL=http://<lb-ip>:80 to .env.e2e
The runner pod (in the cluster) can reach LoadBalancer IPs assigned
by the cloud controller or metallb.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Previous attempt used kubectl port-forward to a Service, which failed
with 'connection refused' — the API server could not reach pod IPs.
Switch to NodePort (30080) service type and use the node's InternalIP
for HEADLAMP_URL, reachable from the GitHub Actions runner pod.
- Change Service type from ClusterIP to NodePort with nodePort: 30080
- After rollout, get node InternalIP via kubectl get nodes
- Poll http://<node-ip>:30080 until reachable
- Write HEADLAMP_URL=http://<node-ip>:30080 to .env.e2e
- Remove port-forward leftover cleanup from teardown script
Co-Authored-By: Paperclip <noreply@paperclip.ing>
The browser runs outside the cluster and cannot resolve
headlamp-e2e.${E2E_NAMESPACE}.svc.cluster.local DNS names.
- Start kubectl port-forward in background after service rollout
- Poll until localhost:4466 is reachable before writing .env.e2e
- Write HEADLAMP_URL=http://localhost:4466 so Playwright browser can connect
- teardown: kill port-forward processes with pkill
Fixes PRI-752.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
- Adds e2e/auth.setup.ts, e2e/kube-vip.spec.ts with waitForSidebar helper
- Adds playwright.config.ts, scripts/deploy-e2e-headlamp.sh, scripts/teardown-e2e-headlamp.sh
- Adds .github/workflows/e2e.yaml
- Fixes plugin settings test to wait for list before searching
Chris Farhood