[Bug] Pod reconciliation fails if worker pod name is supplied

[kevin85421]

Why are these changes needed?

Deploy a RayCluster with a worker group with replicas >= 2 and supply a name in the pod template. The operator will fail to generate multiple worker replicas and will show an error creating the second pod. See this commit for more details. The following instructions can reproduce this bug.

# Step1: Clone my Kuberay repo (kevin85421/kuberay), and check out to the `replicate-worker-name-control-group` branch.
git clone git@github.com:kevin85421/kuberay.git
git checkout replicate-worker-name-control-group

# Step2: Install kuberay-operator
cd helm-chart/kuberay-operator
helm install kuberay-operator .

# Step3: Install ray-cluster
cd helm-chart/ray-cluster
helm install ray-cluster .

[Solution]: Set ObjectMeta.Name to an empty string, and use GenerateName to prevent name conflicts.

[Note]: Relationship between Name and GenerateName (Link)

Related issue number

Closes #582

Checks

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

I tested this PR with the same config for ray-cluster in kevin85421@fe80c75, but used the new image for kuberay-operator. As shown in the screenshot, two workers are created successfully.

[DmitriGekhtman]

Looks good. This is the correct action here. Could you add a unit test?

And then it would be great to mention in the Ray docs here https://docs.ray.io/en/latest/cluster/kubernetes/user-guides/config.html#pod-templates that metadata.name and metadata.generateName are ignored.

[kevin85421]

Looks good. This is the correct action here. Could you add a unit test?

And then it would be great to mention in the Ray docs here https://docs.ray.io/en/latest/cluster/kubernetes/user-guides/config.html#pod-templates that metadata.name and metadata.generateName are ignored.

I added a unit test. I opened an issue to update the document in the ray repository.

[DmitriGekhtman]

lgtm