Skip to content

Operations and evidence ​

Norn records application and platform work as operations with explicit progress and outcomes. An accepted request, a live process, and a completed operation are different states.

WorkExpected evidence
PreflightCandidate source, input validation, build/test result, terminal outcome
Application deploy or rollbackOperation identity, deployment stages, serving allocation and route/version proof
Snapshot, restore, or migrationExact app/database binding, selected snapshot or migration, safety/recovery evidence
Platform upgradeSigned source/artifact and host identity, drain/compatibility checks, observation and terminal receipt
Infrastructure changeReviewed desired source, provider/state identity, plan, authorized executor and provider readback

The store depends on the release strategy ​

The self-hosted strategy uses PostgreSQL control state and local host maintenance integration. The tested Fleet strategy uses etcd control state and an external management/runner path. Application databases are separate from those control stores. PostgreSQL queue implementation details are not a description of every Fleet operation.

Recovery preserves effect identity ​

A controller persists enough intent and progress to inspect the operation after restart. If an external service accepted a request but its response was lost, read back the same effect before deciding whether another action is safe. Unknown effects hold for resolution rather than receiving an automatic repeat.

Read-only work and some explicitly replay-safe transitions can retry. A mutable stage does not become replay-safe because its process exited. Inspect the specific controller's journal and recovery contract.

Observation checks the user-visible result ​

Beacon events, logs, allocation health, and operation receipts answer different questions. Combine them with a request to the intended application's endpoint. For platform changes, observe workload continuity and API ownership throughout the accepted window, with an explicit outage and sample-gap policy.

Operator workflow, controller recovery, and Fleet operations describe the corresponding procedures.