Scaling K8s Clusters Without Losing Your Mind or Money
summary
Carpenter replaces node groups and Cluster Autoscaler by provisioning EC2 capacity directly via EC2 Fleet requests, making scale decisions in seconds and bin‑packing pods across many instance types. It implements spot best practices, manages spot lifecycle and uses a price‑capacity‑optimized allocation to lower interruption risk — customers report up to 90% discount on spot capacity — and supports Graviton multi‑arch images to gain up to 40% better price‑performance, all without predefining node groups.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:17What is Carpenter
- 2:21Auto scaling requirements
- 6:40Cluster autoscaler basics
- 9:15Cluster autoscaler challenges
- 12:25Carpenter introduction
- 13:48Carpenter node provisioning
- 15:32Kubernetes-native Carpenter
- 17:00Configuring Carpenter provisioners
- 19:08Instance type flexibility
- 20:35Spot instances for cost
- 22:00Carpenter spot best practices
- 23:04Graviton processor benefits
- 24:07Mixed Graviton/x86 clusters
- 25:31Efficiency and cost summary
- 26:33Carpenter development
related talks
26mScaling K8s Nodes Without Breaking the Bank or Your Sanity
21mWhy should you bother about cluster autoscaling- Karpenter
27mHarness the power of Karpenter to scale, optimize & upgrade K8s
22mUse Karpenter to scale, optimize & upgrade K8s
36mGroupless Autoscaling with Karpenter
22mKarpenter Efficient scaling of Kubernetes clusters
47mBuilding sustainable architectures with Kubernetes
34mReduce Costs and Optimize on Your Cloud with Karpenter