Whenever you're ready, walk me through how you'd approach this.
What are the primary considerations when designing for high availability in a cloud-native environment?
Discuss the question in the context of an end-to-end ML serving platform, without assuming a specific product, traffic volume, or cloud provider. Explain how you would reason about redundancy, failure isolation, deployment safety, stateful dependencies, model serving, observability, disaster recovery, and graceful degradation.