Skip to content

Latest commit

 

History

History
100 lines (81 loc) · 2.81 KB

ray-installation.md

File metadata and controls

100 lines (81 loc) · 2.81 KB

Installing Ray on Kubernetes

This guide covers the common steps for installing Ray on Kubernetes using KubeRay, regardless of where your cluster is running (local, cloud, or on-premise).

Prerequisites

  • A running Kubernetes cluster
  • kubectl configured with the correct context
  • helm installed

Installation Steps

  1. Add the KubeRay Helm repository:

    helm repo add kuberay https://ray-project.github.io/kuberay-helm/
helm repo update

  2. Install KubeRay Operator:

    helm install kuberay-operator kuberay/kuberay-operator

  3. Create a values file (values.yaml):

    head:
  args: ["sudo apt-get update && sudo apt-get install -y curl; curl -LsSf https://astral.sh/uv/install.sh | sh; export PATH=$HOME/.local/bin:$PATH; uv pip install --system getdaft"]
worker:
  args: ["sudo apt-get update && sudo apt-get install -y curl; curl -LsSf https://astral.sh/uv/install.sh | sh; export PATH=$HOME/.local/bin:$PATH; uv pip install --system getdaft"]

rayCluster:
  headGroupSpec:
    template:
      spec:
        containers:
          - name: ray-head
            image: rayproject/ray:2.40.0-py310  # Use the desired Python version
            command: ["ray", "start", "--head"]
  workerGroupSpecs:
      template:
        spec:
          containers:
            - name: ray-worker
              image: rayproject/ray:2.40.0-py310  # Same image to ensure compatibility

  4. Install Ray Cluster:

    For Apple Silicon (M1, M2, M3, M4) or other ARM64 processors (AWS Graviton, etc.):

    helm install raycluster kuberay/ray-cluster --version 1.2.2 \
  --set 'image.tag=2.40.0-py310-aarch64' \
  -f values.yaml

    For x86/AMD64 processors:

    helm install raycluster kuberay/ray-cluster --version 1.2.2 \
  -f values.yaml

  5. Verify the installation:

    kubectl get pods

Accessing Ray

Port Forwarding

To access the Ray dashboard and submit jobs, set up port forwarding:

kubectl port-forward service/raycluster-kuberay-head-svc 8265:8265

Ray Dashboard

Once port forwarding is set up, access the dashboard at: http://localhost:8265

Submitting Jobs

You can submit Ray jobs using the following command:

ray job submit --address http://localhost:8265 -- python -c "import ray; import daft; ray.init(); print(ray.cluster_resources())"

Troubleshooting

  1. Check pod status:

    kubectl get pods
kubectl describe pod <pod-name>

  2. View pod logs:

    kubectl logs <pod-name>

  3. Common issues:

    • If pods are stuck in Pending state, check resource availability
    • If pods are CrashLoopBackOff, check the logs for errors
    • For ARM64 issues, ensure you're using the correct image tag with -aarch64 suffix