Autoscaling in Kubernetes
summary
Explains exactly how Kubernetes scales pods and nodes: horizontal pod autoscaler uses pod CPU requests and target utilization (example: sum usage / target → desired replicas), metrics come from kubelets via Heapster; cluster autoscaler watches pods/nodes, simulates expanding node groups (instance groups/ASGs) and adds or removes nodes after 10‑minute safety checks and disruption budget checks; upcoming vertical pod autoscaler adjusts container requests from historical usage. Details include failure modes, config locations, and when autoscaling can fail or over/under provision.
AI-written summary · about
chapters (20)
- 0:00Intro
- 1:28Avg. utilization
- 1:4015% utilization
- 1:50Why to overprovision?
- 2:40Autoscaling in Kubernetes
- 3:10Autoscaling replica count
- 3:52Maintaining the decent load
- 5:59What is usage?
- 8:25HPA Architecture
- 9:13HPA Best Practices
- 12:56Philosophy of Node Count
- 14:18Basic Idea of Automation
- 16:06Nodes in Cluster Autoscaler
- 16:34Main Loop Checks
- 19:10Unneeded nodes
- 20:35When to kill a node?
- 21:13Node killing process
- 22:42CA Best Practices
- 25:52What is missing to reach GA?
- 28:43Vertical Pod autoscaler
related talks
35mIntroduction to Autoscaling
34mIntroduction to Autoscaling
35mIntro: SIG Autoscaling
35mAutoscaling Kubernetes Deployments: A (Mostly) Practical Guide
17mScaling Made Simple: Implementing Autoscaling for Optimal K8s Performance
24mSIG-Autoscaling: Introduction + HPA/VPA Updates
34mPredictive autoscaling patterns in Kubernetes
34mDeep Dive: SIG Autoscaling