VPA can rightsize pods and avoid last minute resource capacity incremental—but unexpected evictions make teams nervous. HPA scales replicas based on demand—but without proper resource requests, it can create inefficient scaling decisions. This talk focuses on extracting value from VPA in recommendation mode: configuring the recommender, interpreting CPU/memory suggestions, applying them safely via pipelines, and validating against SLOs. Also, we’ll see, how to set appropriate scaling thresholds and prevent aggressive scaling with HPA. Additionally, we’ll talk about VPA’s update mode trade-offs, HPA’s scaling metrics, Pod Disruption Budgets (PDBs), and how VPA and HPA complement each other without conflicts. You’ll learn patterns to reduce OOMs and throttling while avoiding flapping and mid-traffic evictions.

Target audience - intermediate to advanced
Target system - Container orchestrator (k8s)