Skip to content

DO staging topology source audit — refreshed 2026-11-09 ​

This audit distinguishes the current protected norn-fleet main source, an older local checkout, provider inventory, and unverified runtime health. The topology match is not proof of node enrollment, workload placement, or service availability.

Current protected source ​

GitHub reports antiartificial/norn-fleet main at 35e1c893198efbdc9e407d708e5096a32b485f7d (including PR #949's repeated staging-plan readback). Its environments/staging/nyc3/cluster.yaml declares three pools:

  • 3 control Droplets (s-2vcpu-4gb);
  • 2 ingress Droplets (s-1vcpu-2gb);
  • 4 app Droplets (s-2vcpu-4gb).

The current app pool is one pool labeled workload: app; this does not change the future General + Stateful two-pool default. Protected source supports the future pair in validation, generated inventory, and Ansible/Nomad bootstrap, but staging's checked-in cluster configuration remains unchanged. Three focused tests for the two-pool validator, inventory, and bootstrap boundaries passed against main 35e1c893. Managed PostgreSQL and MySQL are separate database services, not compute pools.

The earlier local source audit used norn-fleet checkout cc2be083a00d6afd3964655519dc2abb580150b6, which declared only control and ingress pools. That checkout is behind the protected main source for this topology and must not be used as the current staging contract. Its state informs the origin of the stale five-node documentation, not today's source.

Provider readback ​

Read-only DigitalOcean API queries through explicit work context on 2026-11-09 18:58 UTC found all nine expected NYC3 Droplets active: three control, two ingress, and four app nodes. Both staging-management Droplets and the ingress load balancer were active; management-authority PostgreSQL and application PostgreSQL/MySQL reported online. The current public canary probe at 19:01 UTC returned the expected route proof and source across two allocations on two nodes. The default norn-pilot context is a different account; do not use it for staging inventory.

These reads confirm provider-reported resources and a public canary app route. They do not establish exact protected-state identity, Nomad/Consul health, node enrollment, full workload health, Norn API health, or a successful platform release. The latest protected Genesis readback (42418784411) failed in the source-capability check after private transport. Its pinned state and route inputs are from Oct 21, and the log does not identify which source object is absent. A new readback needs a fresh signed one-use intent bound to reviewed current inputs after that object availability is diagnosed.

Reconciliation and next action ​

The current DO staging Fleet is a 3-control + 2-ingress + 4-app, nine-node testbed. The nine-node statechart has the same pool counts and useful partial create/recovery invariants, but aggregates credentials and inputs and does not exercise the protected provider workflow. It must not be described as a live recovery qualification.

The agreed General + Stateful two-app-pool default is for a future fresh Fleet. Its input contract and separate model do not change current staging. Stable pool IDs are selected as general and stateful. Sizes, min/desired/max, cost limits, placement, stateful recovery policy, and an integrated protected apply/recovery fixture remain open before a fresh provider profile or plan exists.

Before any DO write lane, obtain fresh protected state, route, and host-key receipts; verify private control APIs, etcd quorum, capabilities, workload and route health; and diagnose source-object availability. Provider inventory is not a substitute for those gates.