Saving Money Exposed Our Architecture - Escaping ClusterAPI with Karpenter
summary
Migrates 19 clusters (≈680 nodes, ~8,000 pods) off ClusterAPI into a Terraform-managed Amazon EKS layout and replaces controller-heavy node management with Karpenter plus some Fargate for control plane safety. Node spin-up drops to ~30s; CI spot adoption reaches ~80% and some clusters see costs cut to about one‑third. Work succeeds but uncovers fragile apps: many services run single replicas, ~1min cold starts, missing retries/backoffs—so platform adds topology‑spread, PDBs, graceful shutdowns. Terraform and GitOps ownership make the migration low‑risk.
AI-written summary · about
chapters (10)
- 0:00Introduction and background
- 6:51Cluster API challenges
- 12:06Escaping Cluster API
- 13:56Introducing Karpenter
- 18:35Architectural flaws revealed
- 21:47Resilient application design
- 23:32Spot instances and disruptions
- 26:24Platform improvements
- 28:32Cost savings and results
- 30:18Key takeaways
related talks
33mNavigating and Optimizing Karpenter - Avoiding Common Pitfalls
22mHarnessing Karpenter: Transforming Kubernetes Clusters with Argo Workflows
34mReduce Costs and Optimize on Your Cloud with Karpenter
47mPostmortem: A Modern Approach to Managing Kubernetes Clusters
22mKarpenter Efficient scaling of Kubernetes clusters
22mUse Karpenter to scale, optimize & upgrade K8s
27mHarness the power of Karpenter to scale, optimize & upgrade K8s
47mBuilding sustainable architectures with Kubernetes