Pros and Cons of Ray on a Single Machine

Most of our deployments will be on a single server. Some will scale into clusters but most will remain on a single-server.

A small proportion of the deployments will be Ray clusters from the get-go.

What are the pros and cons of deploying a Ray Core (with Ray Serve) application on a single server relative to not using it at all?

On a single server, Ray’s benefit is more efficient multi-processing. If you have use case that will requires multiple Python processes running flexible workload, Ray should be suitable. If all your workload fit in a single process, or your workload can be easily replicated by just launching multiple instances of it, Ray will not help here.