[Bug] KubeRay does not work on M1 macs.

[kevin85421]

Why are these changes needed?

The nightly docker images for arm64 are available. See ray-project/ray#31537 (comment) for more details. Hence, users can run Ray in Docker containers on ARM machines (including Mac M1) and on ARM cloud instances (e.g. AWS Graviton).

Related issue number

Closes #523

Checks

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

# Step 1: Create a KinD cluster 
kind create cluster

# Step 2: Register a Helm chart repo
helm repo add kuberay https://ray-project.github.io/kuberay-helm/

# Step 3: Install both CRDs and KubeRay operator v0.4.0.
helm install kuberay-operator kuberay/kuberay-operator --version 0.4.0

# Step 4: Install a RayCluster custom resource
helm install raycluster kuberay/ray-cluster --version 0.4.0 --set image.tag=nightly-aarch64

# Step 5: Forward the port of Dashboard
kubectl port-forward --address 0.0.0.0 svc/raycluster-kuberay-head-svc 8265:8265

# Step 6: Check 127.0.0.1:8265 for the Dashboard

# Step 7: Log in to Ray head Pod and execute a job.
kubectl exec -it ${RAYCLUSTER_HEAD_POD} -- bash
python -c "import ray; ray.init(); print(ray.cluster_resources())" # (in Ray head Pod)

[jiwq]

LGTM Thanks @kevin85421 for the good work, it helps me to resolve problem.

[kevin85421]

Do not merge this PR. @jiwq found that the raylet processes will not be launched. We need to wait for the Ray community to fix this issue.

ray-project/ray#31736

[kevin85421]

The nightly docker images for arm64 are available. See ray-project/ray#31537 (comment) for more details. I have already updated this PR and PR description.

cc @jiwq @DmitriGekhtman @krfricke @architkulkarni @Jeffwan

[jiwq]

+1 Worked well in my m1, thanks @kevin85421

[architkulkarni]

Tested manually on M1, works for me!

[kevin85421]

Thank @jiwq @architkulkarni for the review! Merge this PR. The CI failure is due to #852.