"Karpenter on Kubernetes: Cost-Aware Autoscaling for Real Workloads"
Karpenter has changed what "autoscaling" means on Kubernetes: not just adding nodes, but continuously choosing better capacity under real scheduling constraints and real cloud market volatility. This book is for experienced platform engineers, SREs, and cloud architects who run production clusters and want cost efficiency without turning reliability into a gamble. If you've outgrown node groups, hand-tuned instance lists, and reactive scaling policies, you'll find a pragmatic, systems-level approach here.
You'll build a correctness-first mental model of autoscaling—starting with the Kubernetes scheduling inputs that determine feasibility (requests/limits, affinity, taints, DaemonSet overhead, PDBs), then moving into how Karpenter's control loops translate unschedulable pods into concrete capacity decisions. From there, the book treats Karpenter's CRDs (NodePool, NodeClass, NodeClaim) as a configuration language for intent, guardrails, and provider integration. You'll learn to engineer flexible instance selection, design multi-AZ and multi-capacity strategies, and harvest savings safely through consolidation, disruption budgets, and drift management—especially under stateful and latency-sensitive workloads.
The final arc is production operations: governance for ephemeral nodes, deep observability to explain "why" Karpenter acted, and lifecycle practices for installation, hardening, and upgrades across major API milestones. Throughout, the emphasis is on decision criteria, failure modes, and repeatable pl
