System status
Affected: IndexingDerived intelligence
GET /api/health returns 200 while every component is operational or deliberately paused, and 503 when anything is degraded, down or unknown — so an external monitor can key off the status code alone.
The same payload carries per-component state, the timestamp of the last check, and — where a share is meaningful — the proportion of that component running normally. It also says whether the 48-hour history could be read, so an empty window is never mistaken for a quiet one.
Checks run on separate infrastructure every 15 minutes. Where a component is served by many interchangeable sources, it is only marked down once every one of them has failed — while some are still working we say so, and show how much. Where it is a single path, such as the query endpoint or the database, one failed check is that component being down.
This page is served by Memra itself, so it cannot report an outage that takes the whole app down. The watchdog probes the site from outside, so such an outage is still recorded and still alerts our team — you would see it here once the site is reachable again.