Shared Ray Cluster

Ray Tune
1

Hi, I’d like to run several Ray Tune jobs that are based on a different models.
Each such model comes from a different code base. I.e., one model will be cleated from the /develop branch while another will be created from the /qa branch.

AFAIK, it is impossible to share the Ray cluster for different jobs.
I.e., the cluster is ephemeral and ad-hoc, suitable for a single Tune job.
I’ve gathered this from Share the Ray cluster - #3 by Alex

Could you please confirm this or state otherwise?
Can jobs be launched on Ray cluster outside the Ray driver process, issuing ray.init()?

2

Hey, you should definitely be able to run multiple jobs in a single cluster, Ray just doesn’t provide strong security isolation between the jobs.

For most organizations/teams this should be fine.

You shouldn’t code on the same cluster as a malicious job though.