Groupless Autoscaling with Karpenter
summary
Karpenter replaces static node groups by requesting instances directly from the cloud, using EC2 Fleet to pick cheapest compatible types and zones, and creating node objects that immediately bind pending pods. It runs a fast allocator for latency-sensitive scale-up and a slow reallocator for cost-driven defragmentation, uses first-fit-descending bin packing, and reduces ASG overhead (~30s). Demo: single pod runs in ~49–56s; 100 pods launch across two large instances (12x + 24x) in about two minutes.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 5:20Groupless Autoscaling
- 8:34Architecture
- 10:05Launching Capacity
- 12:30Scheduling
- 15:09API Conformance
- 16:26WellKnown Labels
- 22:00Defragmentation
- 23:07Latency
- 24:35Kubernetes Custom Resources
- 25:03Kubernetes Roadmap
- 25:58Karpenter Demo
related talks
22mKarpenter Efficient scaling of Kubernetes clusters
19mJust in time Worker Nodes with Karpenter
22mUse Karpenter to scale, optimize & upgrade K8s
01h 01mJust-in-time Nodes for Any AWS EKS Cluster: Karpenter
27mHarness the power of Karpenter to scale, optimize & upgrade K8s
21mWhy should you bother about cluster autoscaling- Karpenter
18mKarpenter : Your trusted Companion for Kubernetes Scaling
31mKubernetes workloads with Karpenter & Amazon EKS