I’m trying to setup Kubeflow setup in a Azure AKS cluste using charmed Kubeflow 1.7. I’ve created AKS cluster with Kubernetes 1.24.10 version. It has 2 nodes with 2 CPUs, 16 GB ram, 75 GB storage (Standard_E2ds_v4) each.
I have two issues while doing this set up.
Issue 1
tensorboard-controller is not starting and it’s showing Waiting for gateway info relation message. As many posts suggested I executed following,
juju run istio-pilot/0 "export JUJU_DISPATCH_PATH=hooks/config-changed; ./dispatch"
But it’s saying ERROR invalid unit or action name
. (previous --unit
option is not valid now)
Issue 2
I got following information by running : kubectl get service -n Kubeflow
istio-ingressgateway-workload LoadBalancer 10.0.157.53 20.XX.XX.XX 80:31821/TCP,443:30441/TCP
Then I used following commands (with http and https both)
juju config dex-auth public-url=http://20.XX.XX.XX
juju config oidc-gatekeeper public-url=http://20.XX.XX.XX
But still, I cannot access the Kubeflow dashboard by the EXTERNEL_IP (20.XX.XX.XX). Are there any additional steps to take here on AKS specifically?
I followed Install Kubeflow on Azure Kubernetes Service (AKS) | Ubuntu tutorial and
kubectl patch role -n kubeflow istio-ingressgateway-operator -p '{"apiVersion":"rbac.authorization.k8s.io/v1","kind":"Role","metadata":{"name":"istio-ingressgateway-operator"},"rules":[{"apiGroups":["*"],"resources":["*"],"verbs":["*"]}]}'
Command is not working as well (with the error –
Error from server (NotFound): roles.rbac.authorization.k8s.io "istio-ingressgateway-operator" not found
)