Learn: Operations

How the platform gets run — stood up, parked overnight for cost, rebuilt from zero, and recovered the morning it breaks. The tool for all of it is platctl, a DAG-aware orchestrator that walks the ~98-unit dependency graph so no human has to remember the order.

This is for platform engineers who operate the stack. Foundations covers what the pieces are, and reading it first helps, but you can follow along here with Terragrunt and a working sense of “control plane vs worker nodes.”

Read in this order

1. Orientation — the platform is a dependency graph you can replay. platctl as a build system for the platform; the four verbs (build → park → rebuild → recover); the git-repo-is-the-platform thesis; operate-by-durable-fix. Metaphors: a build system, a phoenix, a seed vault.
2. Reference — the dense lookup: platctl and the discovered DAG, the bootstrap ordering, teardown, validate, parking, rebuild, the day-2 recovery stories, the status ledger, gotchas.

Go deep

• platctl & the deployment DAG — how the graph is discovered (dependency-block parsing + Kahn’s-algorithm parallel walk), the full bootstrap ordering and why each unit comes before what (the state-backend floor, Cilium-first BYOCNI, bootstrap-then-lockdown), the hooks, teardown and the worktree rule, and two-phase validate.
• Parking & day-2 recovery — the down/up choreography (non-destructive scale-to-zero, the ordering scars), the up recovery sweeps, the three counter-intuitive gotchas, and every day-2 recovery story (the node-imbalance meltdown → descheduler, DB-client ordering, DNS reconnect, and more), under the durable-fix discipline.

Then

• What the pieces are: Foundations. Why parking matters: Cost & FinOps (the biggest non-prod lever). The committed-config floor: Secrets & config. Runbooks: docs/runbooks/platform-rebuild-from-scratch.md, docs/runbooks/cluster-scale-down-up.md.