How should Kubernetes observability and troubleshooting be implemented for production?

Collect structured logs and metrics centrally, correlate traces, monitor nodes and workloads, retain events, define alerts around service objectives, and use kubectl diagnostics systematically.