KubeCon + CloudNativeCon North America 2024

Pod resource requests, limits and priority are some of the most fundamental concepts of Kubernetes clusters, and they're easy to understand: if nodes have the resources you need, you get scheduled, and if not, you don't... right? Joe will walk you through some of the surprising behaviors you may encounter with the seemingly basic rules that Kubernetes follows when scheduling and running pods -- and how those rules themselves may not be what you think! We'll dig into eviction and preemption (and why the difference matters) and why priority sometimes doesn't solve the problems you think it will. We'll finish with recent changes to pod resource management that are upending long-standing basics of pod scheduling, particularly the in-place pod resizing feature alpha-released in the last few versions of Kubernetes. You'll leave with a deeper understanding of the (not-so-) simple mechanics, as well as how to debug them when things get messy.