Download to read offline
Uploaded byAWS Chicago
Julia Furst Morgado The Lazy Guide to Kubernetes with EKS Auto Mode + Karpenter
AWS Community Day Midwest 2025 Julia Furst Morgado The Lazy Guide to Kubernetes with EKS Auto Mode + Karpenter Good engineers automate what’s necessary, but great engineers automate what others think is impossible. In this talk, we’ll see how EKS Auto Mode and Karpenter let you stop worrying about node groups, scaling policies, and capacity planning. Learn when these tools perform at their best (and where they might not fit), and leave with actionable insights on making your Kubernetes clusters self-managing.
More Related Content
Similar to Julia Furst Morgado The Lazy Guide to Kubernetes with EKS Auto Mode + Karpenter
![Support, Monitoring, Continuous Improvement & Scaling Agentic Automation [3/3]](https://cdn.slidesharecdn.com/ss_thumbnails/agenticcommunityseries-day3-cfd-251120170304-ddef8112-thumbnail.jpg?width=640&height=640&fit=bounds)
Julia Furst Morgado The Lazy Guide to Kubernetes with EKS Auto Mode + Karpenter
- 1. THE LAZY GUIDETO KUBERNETES with EKS Auto Mode & Karpenter THE LAZY GUIDE TO KUBERNETES with EKS Auto Mode & Karpenter
- 2. Hey there! I'm JuliaFurst Morgado Global Technologist at Veeam AWS Container Hero CNCF Ambassador Organizer: AWS Community Day NY KCD NY CNCF Meetup NY @juliafmorgado
- 3. Control Plane Managed byAWS EKS account Customer account Data Plane Getting started with Amazon EKS today Managed Node Group Managed by Customer
- 4. Control Plane Managed byAWS EKS account Customer account Data Plane Scaling and Optimizing with CA Managed Node Group Managed by Customer
- 5. Cluster Autoscaler CHALLENGES SLOW SCALING WASTED RESOURCES ClusterAutoscaler reacts slowly and only works with ASGs. Overprovisioning + idle nodes = high cost, low efficiency
- 6.
- 7. Pod auto scaling Pending pods NodePoolNodeClass Cluster Autoscaler Auto Scaling Group VM Fleet (instant)
- 8. Control Plane Managed byAWS EKS account Customer account Data Plane Scaling and Optimizing with Karpenter Managed Node Group Managed by Customer
- 9. Compute Flexibility Instance typeflexibility Attribute-based requirements → sizes, families, generations, CPU architecture No list → picks from all instance types in EC2 universe, excluding metal Limits how many EC2 instances this NodePool can provision
- 10. Compute Flexibility AZ flexibility Provisionin any AZ Provision in specified AZs
- 11. Compute Flexibility CPU architectureflexibility: X86-64 Arm64
- 12. Compute Flexibility Purchase optionsflexibility: On-demand, if nothing specified Prioritizes Spot if flexible to both capacities types
- 13. Spot Instance Handling 2-minSpot instance interruption notice via Amazon EventBridge events Set as environment variables in Karpenter controller Deployment object NodePools can be configured for a mix of On-Demand and Spot Karpenter has built-in Spot interruption handler Not required to use Node Termination Handler
- 14. Workload Consolidation After Bin-packing: Betterutilization of worker nodes = reduced costs
- 15. Workload Consolidation After Bin-packing: Betterselection of worker nodes = reduced costs
- 16. Other CHALLENGES SLOW SCALING WASTED RESOURCES ClusterAutoscaler reacts slowly and only works with ASGs. Overprovisioning + idle nodes = high cost, low efficiency "I just want to run my application on EKS and not worry about all these management overhead" MANUAL INFRASTRUCTURE Node groups, instance types, scaling policies, networking, RBAC, multi-AZ setup → too much infra work OPS OVERHEAD Infra maintenance, upgrades, cost tuning, learning curve, continuous monitoring = burnout FRAGILITY Small misconfigs cause outages. Stateful apps still feel risky.
- 17. Control Plane Managed byAWS EKS account Customer account Data Plane Managed Node Group Managed by Customer Manage Worker Nodes (select and provision ideal compute)
- 18. Control Plane Managed byAWS EKS account Customer account Data Plane Managed Node Group Managed by Customer Upgrade Control Plane (Test and upgrade addons, upgrade AMIs)
- 19. I just wantto run my applications on EKS and not worry about all these management overhead
- 21. Control Plane Capabilities Managedby AWS Managed by Customer EKS account Customer account Data Plane Compute Networking Storage Managed EC2 instances
- 22. EKS with Self ManagedKarpenter Manage Karpenter and other addons Manage worker nodes Can use any EKS Optimized or custom AMIs Run undefinitely (not recommended) or choose custom hours/days Automatically rightsizes and optimize cost Automatically upgrade data plane or control when, test and update addons No additional cost, OSS Karpenter is free AWS manages Karpenter and core addons (EBS CSI, LB Controller) AWS manages worker nodes Bottlerocket AMIs only No SSH into nodes Worker nodes will recycle after max 21 days (default 14), you can't delay it Automatically rightsizes and optimize cost AWS automatically upgrade data plane update addons that works Additional 12% Ondemand EC2 cost Security patches auto-applied Disruption budgets respected EKS Auto Mode
- 23. EKS with Self ManagedKarpenter Have in-house Karpenter knowledge Have a platform team to manage cluster and addons You have requirements to use custom AMIs, or non expriring worker nodes need to use custom AMIs or install specific agents or software on the nodes need advanced networking options like security groups per pod or custom CNI need full control over the upgrade process (cluster, nodes, controllers) Want to go to market as fast as possible with Kubernetes Want to reduce operational overhead and simplify cluster operations? If you've been loiking to use EKS Fargate but couldn't due to limitations You don't have a platform team and want to focus on delivering business value You don't have nuanced requirements to use custom AMIs, non expiring worker nodes EKS Auto Mode no infrastructure, simplicity fast start, low control fine-grained tuning, flexibility, enterprise- grade
- 24.
- 25.