Possible to deploy ray operator on K8s cluster without AVX hardware?

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

  • High: It blocks me to complete my task.

Hello … I have a K8s homelab that I was trying to deploy ray onto (using the helm chart). The ray operator pod terminates right away with error 132. I checked the docker logs and they seem to be empty. My hardware is an old AMD server class processor without AVX instructions, and I think that is the cause. Is there a version of the docker image where AVX became mandatory (without it being listed as a requirement)? If so, I can try out an older version of the image. I’d love to experiment with Ray Serve on my hardware for some upcoming projects, so am looking forward to a response.

What Ray version are you using? AFAIK this isn’t a direct dependency of Ray, but might have been pulled in by one of our dependencies.

Operator pod terminating immediately is a bit concerning. If possible, you can try running docker directly on your hardware, and then trying to docker run to get a shell into the docker image, and then run the pod’s command directly (should be able to find this info with kubectl get pod/<ray operator pod> -o yaml). This might provide some more useful output and let you poke around.