The Kubernetes API Server is the heart of your cluster, but at scale, it often becomes a hidden bottleneck, throttling critical controllers, freezing deployments, or crashing under "thundering herd" list-watch storms. This talk is a deep-dive operational clinic for SREs running high-throughput clusters (10k+ pods) who need to move beyond default configurations.

Attendees will learn how to dissect API latency using Audit Logs and Prometheus metrics (apiserver_request_duration_seconds) to identify "noisy neighbor" controllers that starve critical system components. We will dissect the Priority and Fairness (APF) flow control system, replacing legacy-- max-requests-inflight to guarantee that critical system calls (like node heartbeats) never get dropped, even during massive scale-up events. The session includes a live "autopsy" of a real-world API outage caused by unoptimized LIST calls and demonstrates how to fix it using API Streaming (WatchList), proper client-side caching, and more.