Can a ray-cluster be user local?

How severe does this issue affect your experience of using Ray?

  • None: Just asking a question out of curiosity


I made a few observations:

  1. On a server, when user A runs ray start then user B is able to see that cluster (with ray status) and hence user Bs new ray jobs will start on that cluster.
  2. Just running some ray code (e.g. a torch-training) will implicitly start a cluster (for user A, which by 1. user B will be able to see).

Now I ran into a few (minor) annoyances with this and was wondering if it is possible to have my ray-cluster only visible for my own user.

If not, what would be best practices to run ray code on a server with different users? Just have a special ray-user run a permanent cluster which all users connect to? Just have every user separately run ray start but assign different IP-addresses each time?


I don’t believe Ray has any built-in feature to restrict access to a particular user.

It’s reasonable to have a permanent cluster which all users submit jobs to, but you’d need to enforce your own resource limits per user so that one user doesn’t hog the entire cluster.