How severe does this issue affect your experience of using Ray?
- Medium: It contributes to significant difficulty to complete my task, but I can work around it.
I’m looking at setting up an autoscaling ray cluster in Azure. We depend heavily on a docker and would like to be able to autoscale our cluster using Azure’s container instances (more info). The current azure autoscaler seem to only support plain VM’s.
Has anyone else looked into using Azure’s container instances or would the community be interested in a PR adding said functionality? If so - should this be added as separate node_provider?
Can you either
User Ray’s docker field? (example here: ray/example-gpu-docker.yaml at master · ray-project/ray · GitHub)
Use KubeRay with AKS?
or if not, could you describe more about your use case and why you want to use container instances specifically?
The docker field is our current approach. It’s fine but we find building a new full VM to be a little overkill. The container images would be a faster (in terms of spinning up/down instances) and cheaper alternative.
Exploring k8s is a possibility but that requires a bit more exploration on our end for figuring out how it would play with our current docker images.
For a little bit of context this is in a research environment and we have a variety of docker images that contain some fairly intricate custom library / dependency setup.