Building a Cluster Management API using Kubicorn
summary
Demo shows a working declarative cluster API implemented with kubicorn and a controller that exposes machines CRD objects; creating a cluster takes ~7.5 minutes, the controller provisions VMs on GCP, machines get a spec/status link to nodes, and kubectl gains a machines verb. The demo scales one node to five, then upgrades control plane from 1.7.4 to 1.8.3 and recreates nodes in parallel; tooling reconciles desired state and the operations succeed in the demo.
AI-written summary · about
chapters (34)
- 0:00Introduction
- 1:30Robert Bailey
- 3:41Infrastructure should be boring
- 4:20Cluster lifecycle should be boring
- 4:50Reinventing the wheel
- 6:29Moving to another cloud
- 7:18Kubicorn Admin
- 8:41Infrastructure Layer
- 9:33Kubicorn
- 12:04Problem Space
- 12:28Cluster Management API
- 12:44Cluster Creation
- 15:15Machines
- 18:06In practice
- 18:36Generic tools
- 19:27Demo
- 21:00Machine
- 21:24Node
- 22:10Create New Node
- 22:30Create New Machine
- 23:02Node Groups
- 24:02ClientSide Tool
- 24:18Label Matching
- 24:38Idempotent Requests
- 27:09Upgrading the Control Plane
- 28:08Running the new version
- 28:29Upgrading the nodes
- 29:00Not ready state
- 29:24New VMs provisioned
- 29:39Scaling
- 30:40Repairing nodes
- 31:38Roles
- 32:03Controller Model
- 32:32Bootstrapping
related talks
21mCluster API Deep Dive
46mSIG Cluster Lifecycle (Cluster API)
33mCluster API Deep Dive
17mDedicated Cluster management using Cluster API and Kube-Bind
32mCluster API as Code
39mCluster API & Cluster Autoscaler
30mBuild Your Own Cluster API Provider the Easy Way
07mDemo with Meaghan Kjelland - Software Engineer