Right-Size Your Kubernetes Deployments
summary
Shows two practical ways to reduce wasted Kubernetes cluster capacity: a CLI (Krr from Robusta) that queries Prometheus and emits percentile-based resource suggestions, and an operator (Goldilocks) that auto-creates Vertical Pod Autoscaler recommendations and exposes a Grafana-ready history. Recommends preferring CPU requests (avoid hard CPU limits), align RAM request with limit, and notes Kubernetes 1.27 adds alpha in-place resizing; cites observed waste (≈70% requested CPU unused, ≈18% of RAM requested).
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:22Requests and limits
- 6:48Dimensioning methods
- 8:13Demo with KRR
- 14:32Demo with Goldilox
- 22:48Grafana integration
- 25:38Best practices and tools
- 27:21Kubernetes future evolutions
- 28:03Resources and conclusion
related talks
36mDon't Be Greedy: Rightsize Your Kubernetes Cluster With Prometheus
31mFailing to Autoscale?
47mOptimizing Kubernetes Resource Requests/Limits for Cost-Efficiency and Latency
35mAutoscaling Kubernetes Deployments: A (Mostly) Practical Guide
28mScaling in Kubernetes Safely on On-Prem KaaS Across 1,300+ Clusters and 40,000+ Nodes
33mThe Big Kubernetes Mistake Paying More Than Necessary
34mDeep Dive: SIG Autoscaling
36mOptimizing Resource Usage in Kubernetes