Your question is Service Failures and Recoverability. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How do you plan for service failures and ensure system recoverability?
Discuss this as a practical end-to-end ML system design problem. Explain how you would identify failure domains, design graceful degradation and recovery paths, protect against bad models or features, and validate recovery through testing. Address online inference, batch pipelines, dependencies, data consistency, observability, rollback, disaster recovery, and operational runbooks.