AWS re:Invent 2025 - Supercharge your Karpenter: Tactics for smarter K8s optimization (COP208)

Supercharging Kubernetes Optimization with Karpenter

Balancing Cost and Performance

• Achieving the right balance between cost and performance is a key challenge for DevOps and platform engineers running complex production workloads on Kubernetes.
• Simply throwing more resources at the problem can provide stability and performance, but at a high cost.
• Scaling down to the bare minimum can reduce costs, but may lead to resource constraints and availability issues.
• The goal is to find the sweet spot - providing just the right infrastructure to meet the demands of the workloads without over or under-provisioning.

Karpenter's Consolidation Policies

• Karpenter's "consider consolidation" policies automatically monitor nodes and terminate underutilized ones to save costs.
• This feature requires careful configuration and guardrails:
  • Using Pod Disruption Budgets (PDBs) to control how many replicas can be interrupted at once.
  • Setting consolidation delays to prevent Karpenter from reacting to short-term usage dips.
• Proper configuration allows consolidation during off-peak periods without impacting production traffic.

Choosing the Right Instance Mix

• Providing Karpenter with a diverse set of instance types and sizes gives it more flexibility to balance cost and availability.
• Avoid including instances that are too small (high network overhead) or too large (poor consolidation efficiency).
• Monitor the cluster behavior to determine the right instance size thresholds for your specific use case.
• Consider the overhead of DaemonSet containers running on each node, as this can impact costs, especially for licensed software.

Leveraging Spot Instances

• Karpenter makes it easy to manage spot instances, automatically identifying the best spot prices and availability.
• Spot instances can provide significant cost savings for workloads that can tolerate interruptions, such as:
  • Web servers behind a load balancer
  • Stateless microservices
  • Batch jobs
  • Development and staging environments
• Configure Karpenter to use a mix of spot and on-demand instances, with spot as the primary and on-demand as a fallback.
• Consider using Graviton-based instances, which can offer similar or better performance at a lower cost.

Optimizing for Cost Efficiency

• Limiting node pools to a single Availability Zone can reduce cross-AZ networking costs for non-production environments, but should not be used in production.
• Connecting Kubernetes metrics to actual costs is challenging, especially in large, multi-cluster environments.
• Karpenter makes instant decisions, but the underlying infrastructure may not be able to keep up, leading to operational complexity.
• Continuous automation and optimization platforms, like Zesti, can help by:
  • Providing deep visibility into usage and utilization
  • Using predictive models to make more efficient scaling decisions
  • Automatically resizing workloads and distributing them across spot and on-demand instances
  • Maintaining a fleet of pre-cached, hibernated nodes for instant recovery from spot interruptions

Key Takeaways

• Karpenter's consolidation policies, instance selection, and spot instance management can significantly improve cost efficiency in Kubernetes.
• Careful configuration and guardrails are necessary to balance cost savings and production stability.
• Combining Karpenter with a comprehensive optimization platform can further enhance cost savings and operational efficiency.
• Leveraging Graviton-based instances and limiting cross-AZ traffic can provide additional cost benefits, but require careful consideration.