Navigating and Optimizing Karpenter - Avoiding Common Pitfalls
summary
Karpenter replaces ASGs with Kubernetes-native scaling using nodeclass/nodepool/nodeclaim CRDs, runs in production for ~1.5 years, and saved cost by preferring reserved instances via weight settings while reaching ~74% spot nodes in one dashboard. Operational wins and failures: use a separate static nodepool for bootstrap, enlarge root volumes, always reserve for kube/system, enable the Interruption queue (SQS + spot interruption handling), set PDBs and resource requests, start with one-empty consolidation and ramp underutilized consolidation slowly to avoid mass evictions.
AI-written summary · about
chapters (12)
- 0:00Karpenter introduction
- 3:39Karpenter vs other autoscalers
- 7:39Karpenter CRDs and node pools
- 10:13Node pool configuration
- 14:54Monitoring Karpenter
- 17:20Karpenter observability demo
- 19:25Karpenter pitfalls and best practices
- 23:40Handling spot interruptions
- 25:29Instance type selection
- 27:35Advanced node pool strategies
- 29:44Optimizing workloads
- 31:28Q&A
related talks
27mHarness the power of Karpenter to scale, optimize & upgrade K8s
22mUse Karpenter to scale, optimize & upgrade K8s
18mKarpenter : Your trusted Companion for Kubernetes Scaling
22mKarpenter Efficient scaling of Kubernetes clusters
34mReduce Costs and Optimize on Your Cloud with Karpenter
33mKarpenter: Best Practices and Cost Optimization Techniques
36mGroupless Autoscaling with Karpenter
33mOptimizando Clústeres de Kubernetes Gestión Eficiente de Cargas de Trabajo Karpenter