Karpenter consolidation: 6 settings worth tuning in 2026

Short summary

Karpenter's node consolidation defaults cause production incidents when mistuned. Six settings control the cost vs. disruption tradeoff: consolidationPolicy (WhenEmpty vs. WhenEmptyOrUnderutilized), consolidateAfter (5-15m for most workloads), disruption.budgets (safety net with cron-scheduled windows), expireAfter (30-90 days), terminationGracePeriodSeconds (60-300s), and NodePool instance pinning. Tune against workload type to prevent pod churn during deploys.

• •consolidationPolicy: WhenEmpty is safe for stateful workloads; WhenEmptyOrUnderutilized saves more but risks disruption.
• •consolidateAfter timer prevents thrashing: use 2m for stateless web, 5m for mixed, 10m for stateful, 15m for batch jobs.
• •disruption.budgets is critical: set explicit node counts (not %), schedule 0% during deploys, and increase during low-traffic windows.

Generated with AI, which can make mistakes.