Horizontal Autoscaling with Kubernetes
summary
Explains exactly how to go from manual kubectl scaling to an HPA using the metric server, showing CPU/memory defaults and why CPU can miss I/O or queue-driven load; demonstrates installing Prometheus plus the Prometheus adapter to scale on custom metrics and using cada to call an HTTP metric API so pods scale to values (demo goes 1→6→10 and respects a max of 10). Covers stabilization windows, min/max replicas, step limits, and the trick and tradeoffs of scale to 0.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:44Horizontal Autoscaling
- 7:24Why isnt it automatic
- 16:43Testing
- 22:10Metrics
- 24:19Prometheus
- 29:37cada
- 32:40Metric API
- 35:36Scaling Deployment
- 40:11Questions
related talks
30mHorizontal Pod Autoscaler Reloaded - Scale on Custom Metrics
35mSIG Autoscaling Intro
29mKubernetes Autoscaling: Beyond One Size Fits All
35mAutoscaling Kubernetes Deployments: A (Mostly) Practical Guide
34mAutoscale your Kubernetes Workload with Prometheus - Frederic Branczyk, CoreOS
34mIntroduction to Autoscaling
14mA Tale of Autoscale: Increasing Resiliency while Reducing On-Call Toil
08mGetting Creative with HPA Scaling in Kubernetes