platform
why
the bare-metal fleet was managed by a wiki page and a very tired person. fleetd turns node lifecycle into declarative state: you describe the pool, the operator negotiates with the machines until reality agrees.
notes
- reconciles power state, firmware, and os image through one crd
- drains are gradual and slo-aware — the operator watches error budgets before it moves
- the wiki page is now three paragraphs, which i consider the real deliverable
running in production across two data centres since 2025.