Scaling K8s Nodes Without Breaking the Bank or Your Sanity
summary
Explains using EC2 Spot to cut node costs up to 90% discounts while handling the two-minute interruption notice via rebalance recommendations and the AWS Node Termination Handler; recommends not setting a max price and requesting many flexible instance types with a capacity-optimized allocation. Compares Cluster Autoscaler with a new groupless node autoscaler called Carpenter that uses a Provisioner CRD, EC2 Fleet instant mode, respects PodDisruptionBudgets, supports spot defaulting and TTL-based scale-downs; a live demo was attempted but failed.
AI-written summary · about
chapters (14)
- 0:00Introduction to Kubernetes
- 0:28Understanding EC2 Spot
- 1:21Spot pricing and best practices
- 2:20Instance types and allocation
- 4:13Handling interruptions
- 5:27Kubernetes node management
- 6:55Auto scaling nodes
- 8:26Introduction to Carpenter
- 10:26Carpenter provisioner specs
- 13:06Carpenter flexibility features
- 16:04Scaling down with Carpenter
- 17:44Rebalance with Carpenter
- 18:31Conclusion and Q&A
- 20:18Technical Q&A session
related talks
27mScaling K8s Clusters Without Losing Your Mind or Money
21mWhy should you bother about cluster autoscaling- Karpenter
22mUse Karpenter to scale, optimize & upgrade K8s
01h 01mJust-in-time Nodes for Any AWS EKS Cluster: Karpenter
34mReduce Costs and Optimize on Your Cloud with Karpenter
36mCost-effective Compute Clusters with Spot and Pre-emptible Instances
19mJust in time Worker Nodes with Karpenter
27mHarness the power of Karpenter to scale, optimize & upgrade K8s