Kubernetes Autoscaling and Cost

This is for you if

• Your HPAs flap, sit at max, or never scale down and you're not sure why
• You own a cluster bill and need to explain or cut it with data
• You're choosing between Cluster Autoscaler, Karpenter, VPA and KEDA
• You're preparing for senior platform, SRE or DevOps interviews

What's inside

• Why requests drive every autoscaler and cost report, with throttling and OOM kills reproduced and measured
• HPA v2 in depth: the algorithm, resource, container, custom and external metrics, behavior and the tolerance field now stable in 1.37
• VPA modes including InPlaceOrRecreate, and in-place pod resize (stable since Kubernetes 1.35) done by hand
• KEDA scalers, activation vs scaling, and scale to zero with a Redis queue lab
• Cluster Autoscaler vs Karpenter: NodePools, consolidation policies, disruption budgets and spot, plus an optional EKS lab with full teardown
• A right-sizing workflow with PromQL, bin-packing, pod priority, preemption and placeholder headroom
• Cost allocation with OpenCost and FinOps practices that work for platform teams
• Autoscaling pitfalls and a symptom-to-fix troubleshooting playbook with four worked cases
• 24 senior-level interview Q&As, dense cheat sheets and a 30-day practice plan

After this guide you can

• Set requests and limits from data and explain how they drive scheduling, scaling and cost
• Tune HPA behavior so workloads scale fast without thrashing
• Right-size pods with VPA and resize them in place without restarts
• Scale event-driven workloads to zero with KEDA
• Configure Karpenter consolidation, budgets and spot safely
• Find and fix what blocks scale-down, and report cost per team

Contents

1. The mental modelp. 4
2. Requests and limits: why they drive everythingp. 9
3. HPA v2: scaling replicas properlyp. 14
4. VPA and in-place pod resizep. 22
5. KEDA: event-driven scaling and scale to zerop. 30
6. Node autoscaling: Cluster Autoscaler vs Karpenterp. 36
7. Right-sizing, bin-packing and pod priorityp. 46
8. Cost visibility and FinOps for platform teamsp. 52
9. Autoscaling pitfalls and the troubleshooting playbookp. 56
10. Production checklist and design decisionsp. 62
11. Interview questionsp. 65
12. Cheat sheetp. 74
13. 30-day practice planp. 77