Autoscale Your on-Premises Bare Metal Kubernetes Clusters
summary
Combines Cluster API, Metal3 (metal3-io) and Cluster Autoscaler to provision and autoscale on‑prem bare‑metal Kubernetes by treating machines as CRDs and driving provisioning via Ironic/BM operator over BMC. Demo builds a management cluster, creates one‑master/one‑worker workload cluster on two servers, then triggers autoscale to add a third node when a pod requests extra CPUs; provisioning takes ~15–20 minutes and succeeds. Requires clusterctl, capi providers, machineSet annotations and a kubeconfig secret in the management cluster.
AI-written summary · about
chapters (9)
- 0:00Start
- 0:43Background
- 1:50Solution
- 3:28How does Kubernetes work
- 3:53How to extend Kubernetes API
- 13:30Provisioning
- 19:16Deploy Cluster Autoscaler
- 20:57Demo
- 32:16The community
related talks
30mIntroduction To Metal³ — Kubernetes-Native Bare Metal Provisioning
36mIntroducing Metal³: Kubernetes Native Bare Metal Host Management
24mImage-Based Bare-Metal Provisioning With Metal3.io, Cluster API and…
39mCluster API & Cluster Autoscaler
36mBare metal Kubernetes on mixed x64 and ARM64
34mmetal-stack - kubernetes on bare metal
32mDon't Start With 500 Clusters: Patterns in Scaling Multi-Cluster Kubernetes…
20mMetalkube: Deploy Kubernetes on Bare Metal