Affinity, Anti-Affinity, and Topology Spread — Telling the Scheduler What 'Spread Out' Means

Two webapp replicas on the same node is one node failure away from an outage — and the scheduler will happily do that to you, because nothing told it not to. Day 21 covers the three tools that tell it: nodeAffinity (which nodes a Pod prefers or requires), podAntiAffinity (keep replicas apart), and topologySpreadConstraints (the modern '±1 Pod per node' contract). Each one demoed live, including the two failure modes that bite hardest: hard anti-affinity strangling the HPA, and topology spread silently counting the tainted control-plane as a domain.