Whenever you're ready, walk me through how you'd approach this.
What is your approach to ensuring observability and monitoring in a distributed system?
Discuss how you would instrument services, collect and correlate logs, metrics, and traces, define service and model health indicators, and detect incidents across distributed ML serving components. Explain how you would handle alerting, debugging, feature drift, training-serving skew, and monitoring failures without creating excessive cost or alert fatigue.