SIG Autoscaling Intro
summary
Overview of Kubernetes autoscaling: HPA adjusts replica counts using CPU or custom metrics via metric-server or adapters (Prometheus, cloud providers); ensure the chosen metric actually falls when pods increase or HPA will hit min/max. VPA runs in recommendation / initial / auto modes, updates requests by killing pods today, and needs better kernel OOM reporting to avoid surprises. Cluster Autoscaler provisions nodes on demand and removes unused ones, proven at 1000 nodes, 30,000 pods scale. Avoid naive HPA+VPA combos; use careful metrics and limits.
AI-written summary · about
chapters (13)
- 0:00Intro
- 0:25Question of Life
- 4:00Goals of Autoscaling
- 5:03How to maintain decent load
- 6:21Example
- 8:03What is the load
- 12:33Scaling with Custom Metrics
- 13:48Vertical Pot Autoscaler
- 17:15When to use it
- 19:47Right Capacity
- 21:18Cluster Autoscaling
- 24:05Status Report
- 29:26Questions
related talks
24mSIG-Autoscaling: Introduction + HPA/VPA Updates
34mIntroduction to Autoscaling
35mIntro: SIG Autoscaling
36mIntro: Autoscaling SIG
29mKubernetes Autoscaling: Beyond One Size Fits All
34mPredictive autoscaling patterns in Kubernetes
51mHorizontal Autoscaling with Kubernetes
17mScaling Made Simple: Implementing Autoscaling for Optimal K8s Performance