Skip to content

Operability

Starting, degrading and failing.

REQ-0050 — startup loads the policy bundle, seeds, and starts the scheduler, in that order

The lifespan does three things and two of them can stop the process:

  1. init_policy_engine() — raises without a bundle (REQ-0003), so the service does not come up unauthorised.
  2. auto_seed() — reads SEED_DIR through the same loader as everything else, so a missing or malformed active_kinds.yaml raises here (REQ-0069). A service that started without a catalogue would suppress every notification as an unknown kind.
  3. run_scheduler() — a background task. Shutdown sets its stop event and awaits the task, so a poll in flight finishes rather than being cancelled mid-dispatch.

auto_seed is skipped, with a log line, when SEED_DIR is unset or names a directory that does not exist. That is the one way to start without a catalogue, and it is deliberate.

GET /health reports only that the process is up. It checks nothing — not the database, not the seed, not the scheduler — so it is a liveness probe and must not be read as a readiness one: a service whose PostgreSQL has gone answers {"status": "ok"} and fails every request behind it.