Can on-prem be also a cloud Networking made easy
summary
OpenShift Bare Metal Networking shows a concrete on‑prem recipe: API and Ingress VIPs implemented with keepalived + HAProxy to prevent TCP session churn, a bootstrap-in-place installer that hands VIP priority to the bootstrap node during install, and nodeIP/runtime-config hacks that set kubelet bind addresses. Demonstrates local CoreDNS per node to avoid preexisting DNS, explains why kubelet dual‑stack with cloud providers is unsupported today, and warns about the annoying IPv6 mapped IPv4 socket bug.
AI-written summary · about
chapters (17)
- 0:00Container networking on bare metal
- 3:55Containers on bare metal vs cloud
- 6:43API access with virtual IPs
- 10:38Preventing IP flapping
- 13:27HAProxy for API stability
- 15:36Bootstrapping Kubernetes clusters without extra nodes
- 17:02Managing API VIP during install
- 19:09VIP scaling limits
- 20:59Isolated worker nodes
- 21:21Simplifying DNS for clusters
- 23:07DNS for container workloads
- 25:17Binding components to IPs
- 27:24Override kubelet IP binding
- 29:53Dual-stack node IP challenges
- 31:39Dual-stack cloud limits
- 32:41IPv6-mapped IPv4 pitfalls
- 34:54Bare metal vs on-premise
related talks
41mKubernetes on-premises, why you don't want to do it?
36mPractical experiences with OpenStack on OpenShift
22mOpenShift, or Kubernetes on steroids
01h 06mOpenShift - the power of Kubernetes for engineers
35mHybrid Cloud Powered by Kubernetes [I]
34mBack to Basics, Getting Traffic Into your Kubernetes Cluster
35mHigh Availability Kubernetes on Bare Metal [A]
49mThe state of on-prem Kubernetes