[Bug] Modification of nameOverride will cause label selector mismatch for head node

[kevin85421]

Why are these changes needed?

As mentioned in #495 and #530, the value of nameOverride needs to be hardcoded with the value "kuberay" to avoid the mismatch between head pod and head service.

The value of app.kubernetes.io/name in the head pod's labels is same as app.kubernetes.io/name specified in RayCluster.Spec.HeadGroupSpec.Template.ObjectMeta.Labels of RayCluster custom resource YAML file. See the function labelPod in pod.go for more details.

RayCluster.Spec.HeadGroupSpec.Template.ObjectMeta.Labels["app.kubernetes.io/name"] is controlled by the value of nameOverride in helm chart.

However, the value of app.kubernetes.io/name in the head pod service's selector is hardcoded with the constant variable ApplicationName (= "kuberay")

Related issue number

closes #571
#495
#530

Checks

• I've made sure the tests are passing.
• Testing Strategy
  • Unit tests
  • Manual tests
  • This PR is not tested :(

# Step1: Clone my repo and checkout to `nameOverride` branch
# Step2: Build Docker image
# Step3: Update image of kuberay-operator chart
# Step4: Update `nameOverride` in ray-cluster chart from "kuberay" to "aaaaaaaaaa"
# Step5: Install kuberay-operator
cd helm-chart/kuberay-operator
helm install kuberay-operator .

# Step6: Install raycluster custom resource
cd helm-chart/ray-cluster
helm install ray-cluster .

# Step7: Set up port-forwarding
kubectl port-forward svc/ray-cluster-aaaaaaaaaa-head-svc 8265:8265

# Step8: Submit a simple job
ray job submit --address http://localhost:8265 -- python -c "import ray; ray.init(); print(ray.cluster_resources())"

Check head node service

• kubectl describe svc/ray-cluster-aaaaaaaaaa-head-svc

• Without this PR:

Selector: app.kubernetes.io/created-by=kuberay-operator,app.kubernetes.io/name=kuberay,ray.io/cluster=ray-cluster-aaaaaaaaaa,ray.io/identifier=ray-cluster-aaaaaaaaaa-head,ray.io/node-type=head

• With this PR:

Selector: app.kubernetes.io/created-by=kuberay-operator,app.kubernetes.io/name=aaaaaaaaaa,ray.io/cluster=ray-cluster-aaaaaaaaaa,ray.io/identifier=ray-cluster-aaaaaaaaaa-head,ray.io/node-type=head

[DmitriGekhtman]

This makes sense to me -- I left some minor nits about test documentation.

If you haven't yet, could you test this out manually? You could do that by building an operator image with the changes, deploying with a non-default label, and making sure the worker pod connects to the cluster successfully.

[DmitriGekhtman]

This makes sense to me -- I left some minor nits about test documentation.

If you haven't yet, could you test this out manually? You could do that by building an operator image with the changes, deploying with a non-default label, and making sure the worker pod connects to the cluster successfully.

Sorry, I see you've already done that.