Auto-scaling party:VPA, HPA, KEDA, nodes, how do they dance?-Nic Vermande & Oluebube Princess Egbuna
summary
Explains how VPA, HPA, KEDA and Cluster Autoscaler actually behave together and what to install to make them work: metric-server and Prometheus for metrics, Cluster API for programmatic node provisioning, and KEDA scalers for event sources like Kafka using a lag threshold trigger. Notes VPA is disruptive because VPA evicts pods when changing requests, HPA uses averages and cannot scale to zero while KEDA can scale to zero, and end-to-end autoscaling often shows 2–5 minute reaction delays.
AI-written summary · about
chapters (10)
- 0:00Introduction and speaker intro
- 1:29Scaling basics and challenges
- 7:28Manual scaling in Kubernetes
- 10:21Cluster Autoscaler overview
- 13:52Cluster Autoscaler with Cluster API
- 17:41Vertical Pod Autoscaler explained
- 23:29Horizontal Pod Autoscaler details
- 29:10KEDA event-driven autoscaling
- 32:42Key takeaways and best practices
- 34:07Q&A session
related talks
35mSIG Autoscaling Intro
29mKubernetes Autoscaling: Beyond One Size Fits All
36mIntro: Autoscaling SIG
34mAutoscaling Made Easy with Kubernetes Event-Driven Autoscaling (KEDA)
24mSIG-Autoscaling: Introduction + HPA/VPA Updates
37mKubernetes Event-driven Autoscaling with KEDA
47mKEDA, HPA and VPA? An Introduction to Scaling
34mIntroduction to Autoscaling