Autoscaling Kubernetes Deployments: A (Mostly) Practical Guide
summary
Covers cluster autoscaler, vertical and horizontal pod autoscalers with concrete knobs: min/max replicas, stabilization windows, and step caps; shows HPA scaling from 1 to 4 replicas on CPU in a recorded demo using Pixie demos repo and the hey load generator, and an error-rate case that adds nine replicas to drop errors; highlights pitfalls like 8% CPU with 5s latency and mismatched sidecar CPU, discusses Prometheus/Linkerd metrics via the custom metrics API, and even implements a Turing-complete autoscaler.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:36Auto-scaling fundamentals
- 3:33Types of Kubernetes scaling
- 8:06Horizontal Pod Autoscaler
- 12:28Scaling knobs and controls
- 16:13Selecting metrics
- 21:47Turing-complete autoscaler
- 26:42Q&A and conclusion
related talks
34mIntroduction to Autoscaling
30mAutoscaling in Kubernetes
51mHorizontal Autoscaling with Kubernetes
29mKubernetes Autoscaling: Beyond One Size Fits All
35mIntroduction to Autoscaling
17mScaling Made Simple: Implementing Autoscaling for Optimal K8s Performance
34mPredictive autoscaling patterns in Kubernetes
38m50 Shades of K8s Autoscaling - Mario Leander Reimer
