Serverless For Ml Inference On Kubernetes: Panacea Or Folly