View a markdown version of this page

Disruption preference policy - AWS Prescriptive Guidance
Actions

Disruption preference policy

Disruption Preference Policy tells Karpenter which nodes to disrupt first when consolidating — prioritize removing the most expensive or least utilized nodes to maximize savings per disruption event.

The configuration covers:

  • NodePool-level budget reasons (Underutilized, Empty, Drifted) to target on-demand nodes first for Spot replacement

  • Pod-level karpenter.sh/do-not-disrupt annotations to control which workloads move first during consolidation

  • Topology spread constraints to prevent consolidation from concentrating pods in a single zone

For the NodePool and Deployment configuration, see the disruption-preference-policy.yaml and verify-disruption-preference.sh for verification commands.

Actions

Combine budgets with preference policies to create a disruption strategy that removes the most expensive nodes first, during the safest windows, at a rate that doesn't impact availability.

Key takeaway: Not all disruptions save the same amount. Prioritizing on-demand node removal over Spot, and empty nodes over underutilized ones, maximizes the dollar value of each disruption event.