Ray cluster from Ray Cluster Launcher CLI

Ray master branch includes Ray cluster launcher for Staroid. This allows creating a ray cluster using standard ray up <cluster configuration yaml> CLI command.

Install Ray and dependency libraries

First, install ray (1.1.0 or newer) and python dependency packages.

$ pip install ray staroid kubernetes

Configure Staroid access token

Then, let’s configure staroid access token. Get access token and set STAROID_ACCESS_TOKEN environment variable.

$ export STAROID_ACCESS_TOKEN=[your access token]

Cluster configuration file

We can get example Ray cluster launcher configuration files for Staroid from Ray source tree.

$ git clone https://github.com/ray-project/ray.git
$ ls ray/python/ray/autoscaler/staroid/example-*.yaml

Open example configurations and modify them as you need.

Start a Ray cluster

Now, you can create a Ray cluster using ray up command.

$ ray up ray/python/ray/autoscaler/staroid/example-full.yaml

Once cluster is up and running, you can attach your shell to the Ray head node.

$ ray attach ray/python/ray/autoscaler/staroid/example-full.yaml

Ray instance management menu

Check Instance management menu. You’ll see your Ray cluster instances.

You’ll find link to Ray dashbord and Jupyter notebook.

Shutdown Ray cluster

To shutdown cluster,

$ ray down ray/python/ray/autoscaler/staroid/example-full.yaml