ray-exploration

Exploration of the use of ray for data processing pipelines in k8s:

Ray is an open-source unified framework for scaling AI and Python applications. It provides the compute layer for parallel processing so that you don’t need to be a distributed systems expert.

Configuring Ray on Kubernetes

The Ray on Kubernetes docs provides a good starting point. Follow the RayCluster Quickstart to learn how to deploy the KubeRay operator and RayCluster custom resource using Helm.

Warning

If using a local Rancher instance of k8s, you may need to increase CPU and memory resource limits. We reccomend giving Rancher 50-75% of available cores, and 50% of memory

Submitting jobs

First, create a local python env w/ ray installed, and activate the env.

mamba env create -f environment.yml
mamba activate ray-exploration

Then, use the ray submit command to submit jobs:

Note

These examples assume you have port-forwarded the kuberay head service as described in the RayCluster Quickstart

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

You can submit a python script that uses ray like this:

ray job submit --working-dir ./ -- python ray_task_example.py

Additional resources

• RayCluster Configuration
• Quickstart using the Ray Jobs CLI
• Key concepts