Karpenter's consolidation feature continuously repacks Kubernetes workloads onto fewer nodes to reduce cloud costs. The post covers three consolidation policies (WhenEmpty, Balanced, WhenEmptyOrUnderutilized) with different risk profiles, how bin-packing simulation works, the consolidateAfter timer behavior (15 minutes recommended for most production clusters), and safety mechanisms including disruption budgets, the do-not-disrupt annotation, and live migration for stateful workloads. A recommended rollout sequence starts with WhenEmpty in non-production, progresses to Balanced, and only moves to WhenEmptyOrUnderutilized with mature PDB coverage and observability in place. Key Prometheus metrics for monitoring consolidation activity are also provided, along with common blockers like tight PDBs and topology spread constraints.
