Diagnose overloaded Kubernetes cluster

A Kubernetes-based microservices system is experiencing overload (e.g., high tail latency, request timeouts, and autoscaling thrash). How would you debug and mitigate it end-to-end? Specify the metrics, logs, and traces you would examine; the tools and commands you would use; the hypotheses you would test; and your immediate mitigations versus longer-term fixes.