e2e leaves cloud-provider-kind running but only checks Deployment Available=True. That does not prove LoadBalancer / Ingress traffic reaches the pod the way deployah cluster up users expect.
On the shared Kind cluster: deploy a fixture that exposes a LoadBalancer (or Ingress), resolve the address, and HTTP-check it with a bounded timeout. No second Kind cycle.
Stay on Kind. Do not switch the suite to k3s just for a load balancer.
e2e leaves cloud-provider-kind running but only checks Deployment
Available=True. That does not prove LoadBalancer / Ingress traffic reaches the pod the waydeployah cluster upusers expect.On the shared Kind cluster: deploy a fixture that exposes a LoadBalancer (or Ingress), resolve the address, and HTTP-check it with a bounded timeout. No second Kind cycle.
Stay on Kind. Do not switch the suite to k3s just for a load balancer.