50 Shades of K8s Autoscaling - Mario Leander Reimer
summary
Shows concrete autoscaling setups: uses HPA (resource and custom metrics via metric-server and Prometheus adapter) to scale a PHP app from 1 to 6 replicas, uses VPA in recommendation mode and Goldilocks to tune CPU/memory, demonstrates event-driven scaling with kada using RabbitMQ queue length, and compares cluster autoscalers (default cloud autoscaler on GKE vs Carpenter on AWS using spot instances). Includes numbers: 5-node cluster spins new nodes, nodes removed after ~10s of idle.
AI-written summary · about
chapters (11)
- 0:00Introduction and speaker intro
- 1:27From Cloud Native to Hyperscale
- 3:53Workload and cluster scaling basics
- 10:07Demo setup and tools overview
- 12:18Horizontal Pod Autoscaler demo
- 17:51Vertical Pod Autoscaler explained
- 21:22Event-driven autoscaling with Keda
- 25:41Cluster Autoscaler in GCP
- 30:20Cluster Autoscaler in AWS with Carpenter
- 34:18Summary and closing remarks
- 35:21Q&A session
related talks
35mAutoscaling Kubernetes Deployments: A (Mostly) Practical Guide
29mKubernetes Autoscaling: Beyond One Size Fits All
17mScaling Made Simple: Implementing Autoscaling for Optimal K8s Performance
35mSIG Autoscaling Intro
51mHorizontal Autoscaling with Kubernetes
22mIntelligent Autoscaling in Enterprise...
38mAutoscaling Containers... with Math
21mMusic Matched to Different Ways to Scale in K8s, What Could Go Wrong?