Kubernetes on bare metal, on a budget
This section is about running Kubernetes in production on real hardware, not on a managed cloud service. Some of it is hardware I own and operate; a lot of it is gear bought for almost no money — second-hand and low-cost machines that prove you do not need a big budget to run a serious cluster.
Bare-metal Kubernetes means you own the parts a managed cloud usually hides: the container network, load balancing, ingress, and persistent storage. I run a software-defined network for pod connectivity and network policy, distributed storage so that stateful workloads survive a node failure, and an ingress layer that terminates and routes traffic. None of that is exotic, but on bare metal you have to actually understand it, because there is no console to click.
What I write about is the findings from operating it: a live migration between container network plugins, how storage behaves under load, the failure modes that only show up in production, and how I track them down.
Expect posts on cluster networking, storage, ingress and troubleshooting. The aim is the transferable lesson, not a walkthrough to copy blindly.