GitHub's Metal Cloud

This post details the engineering effort behind gPanel, GitHub's physical infrastructure management application. It describes the pipeline for server provisioning using IPMI, iPXE, Facter, and an Ubuntu PXE image. Key components include hardware data collection and modeling in gPanel, state management (unknown, configuring, firmware_upgrade, breakin, memtesting, failed, ready, installing, installed), automated burn-in processes with custom scripts and MemTest86, and OS installation initiated via Hubot. The post also covers the decommissioning workflow.