Health

Liveness probe — always HTTP 200, but honest about dependencies.

Deliberately still 200 even when degraded: Docker's healthcheck calls this endpoint, so returning non-200 when Postgres is down would mark the container unhealthy and restart-loop a service that is other-wise perfectly able to serve live search traffic.

What changed and why: this used to return a flat {"status": "ok"} with no dependency checks at all. On 2026-08-05 postgres and redis were stopped and never restarted, and because search still streams live results straight from the providers, this endpoint reported "ok" for three days while every search went unpersisted and the booking path's Redis offer cache was gone. A liveness probe that cannot go yellow is indistinguishable from one that is not running.

The verdict is cached for 30s so this stays cheap.