Is Serverless Ray Possible (with Azure Container Apps, Fargate etc...)?

1

Hi there,

I’m interested in adding Ray to my ML infrastructure, but I am using Azure Container Apps rather than Azure Container Instances or Azure Kubernetes Service.

In addition to the coordination mechanisms that make it simple to deploy a k8s cluster with ACA, it also allows for autoscaling containers to zero, which isn’t possible with AKS.

I was planning to create a sort of containerized microservice architecture with ACA, such as this from one of your blog posts

image
image1109×499 57.6 KB

But I’d very much prefer to make use of Ray’s capabilities to make more efficient use of resources, etc… Is it possible to use Ray with ACA or would it conflict with Ray’s ?

Or, conversely, is Ray sufficiently simple that I could deploy on AKS with minimal devops skill/attention?

Thanks!