"Serverless" Ray

today the process of creating a ray cluster is quite separate from submitting work to it.

Has there been consideration to make the lifecycle management of a cluster more an inherent part of submitting work, and remove that responsibility as much as possible from the developer?