Exposing All Ray Cluster Nodes(Head and Worker) in K8s

1

So i have 3 nodes (one head and 2 workers)…All have the ray backend deployed via

serve.start(detached=True, http_options={‘location’:“EveryNode”,
“host”: “0.0.0.0”
})
When I try to access the endpoint of the ray backend via the service listening on port 8000. I get connection refused as if the process is not listening on port 8000 inside the container on all interfaces. but if I jump into the containers(any of them or all of them_). I can hit the endpoint and get a response.

apiVersion: v1
kind: Service
metadata:
  name: ml-cluster-aks
  namespace: ray
spec:
  type: LoadBalancer
  ports:
    - port: 8001
      targetPort: 8000
      protocol: TCP
  selector:
    ray-cluster-name: ml-cluster