High Availability Kubernetes on Bare Metal [A]
summary
On bare metal, controllers, etcd and front-end load balancers get small HA clusters that present a single virtual IP so kubelets and proxies point to one address; HA is implemented with Pacemaker+Corosync managing the floating IPs. Two-node controller and two-node LB clusters show immediate failover during node kill tests with only brief connection loss, provided at least three independent network links to avoid split‑brain; Corosync requires multicast/broadcast, so this approach won’t work on most clouds.
AI-written summary · about
chapters (5)
- 0:00Start
- 6:33Bare-Metal Big questions
- 12:33Controllers as mini HA-cluster
- 13:57Load Balancets as mini HA-cluster!
- 22:09Corosync Pace Maker
related talks
32mLoad Balanced Services on Bare Metal with MetalLB
37mManaging and Running Multiple Kubernetes Clusters in Hybrid Setups
36mCan on-prem be also a cloud Networking made easy
40mOperating a Self-Healing Bare-Metal Kubernetes Platform at Global Scale
20mMetalkube: Deploy Kubernetes on Bare Metal
34mBridging Baremetal and Virtualized Clouds: Operating Telco-Grade Kubernetes at Scale
32mAutoscale Your on-Premises Bare Metal Kubernetes Clusters
36mManaging und Running Multiple Kubernetes Clusters