Whenever you're ready, walk me through how you'd approach this.
How do you manage and maintain high availability in a distributed system?
Explain how you would design and operate a distributed ML system that remains available during instance, zone, dependency, deployment, and model-serving failures. Cover architecture, redundancy, failover, graceful degradation, observability, testing, and recovery objectives.