Kubernetes PXE Provisioning Automation

Problem

Validating servers at rack scale means installing an operating system on every machine, repeatedly, across two architectures. Done by hand it is the same sequence of steps per box, and the cost scales linearly with the size of the rack — which makes it the first thing that breaks when the rack gets bigger.

The supporting artifacts had their own problem: the provisioning images and repository had accumulated to 663 MB, and every rebuild and every fetch paid that cost.

What I did

Built a Kubernetes-based test system — kubeadm, Helm, and shell — that drives PXE-based OS provisioning across ARM and x86 servers, covering both Windows and Linux targets.

Getting PXE to behave across two architectures means working below the convenient abstractions: ARP, DHCP, and TFTP are part of the workflow, not implementation details, and the boot path differs enough between ARM and x86 that it has to be handled explicitly rather than assumed.

Separately, audited what was actually in the images and the repository, stripped the base images down to what provisioning genuinely required, and collapsed redundant layers.

Result

• 663 MB → 2.4 MB image and repository footprint — roughly a 99% reduction
• Rack-scale validation across ARM and x86 driven by the system rather than by hand

This is the earliest piece of the automation thread that later became agent skills for firmware development — the same instinct to remove a repeated manual loop, applied at deployment scale before it was applied to firmware work.

Links

TODO: the work is not public. Add a link if a sanitized writeup is produced.