Kubernetes on-premises
summary
A report on migrating production to an on-premises Kubernetes platform with three clusters (~70 physical nodes, ~200 namespaces, ~1,300 pods) that now serves 80–90% of traffic. It describes the stack choices (CoreOS, nginx Ingress, a Docker registry on NFS), deployment metrics (environment provisioning ~10 minutes, CI→release ~16 minutes, production handover 4 minutes), and two resolved failures (an etcd rollback issue and nginx ordering cascades) leading to stable, zero‑downtime releases.
AI-written summary · about
chapters (15)
- 0:00Kubernetes on-prem intro
- 3:33Migration motivations
- 6:23Hardware and software choice
- 10:58Traffic management with Ingress and DNS
- 13:06Monitoring and alerting with Grafana and Nagios
- 15:17etcd upgrade problems and fixes
- 16:43Lessons learned and benefits
- 19:16Software architecture and framework
- 21:22Namespace details and app config
- 23:30Pod monitoring, logging and metrics
- 24:56Zero downtime deployment
- 26:20CD and deployment pipeline
- 28:08Nginx challenges and solutions
- 30:12Infrastructure results
- 31:41Q&A
related talks
41mKubernetes on-premises, why you don't want to do it?
19mBuilding an On-Premise Kubernetes...
40mThe Good, the Bad, and the Ugly of Managing Production Systems with Kubernetes
45mLearning How to Pronounce Kubernetes to Production in 3 Months!
41mKubernetes — A Year in Production
22mAnatomy of a production grade Kubernetes cluster
49mThe state of on-prem Kubernetes
40mFrom dev to prod. One Year of Using Kubernetes in Production