[adamjohnwright]

Spec 006 (specs/006-content-service-to-node). From the review of everything merged since #485.

Unread answers let go. An answer whose body is never read holds its connection until it is collected. Solr's refusals, which anyone can trigger with a query Solr cannot parse, and every report-service answer were never read. Each is now cancelled at once. A cancel that fails, on a body already broken, is ignored: it neither replaces the Solr error nor changes the report's log line, which names the status first.

/data/discover/{id} loads only an event. It used to collect every relationship of the object, both ways, before checking it was an event. For a species or a database that means a heavy query on the shared graph, only to answer a 422. It now asks first for the object's class and whether it has any relationship, answers the 404 or 422 from that as before, and loads only an event. On beta a species id now answers its 422 in about 13 ms.

Smaller items:

• /search/icon/{identifier}: a spec now holds the answer Java's failure gives when Solr counts a match but sends no document. The first commit's change there is undone, because the line already threw (the review found it).
• warm.ts: when the edge reports the wrong release, the message says a purge may be what is needed, not only a restart.

Checks

• Each new spec was shown failing without its fix.
• Tools unit tests: 1,132 passed. Lint, loose, implicit-any and dead code unchanged.
• On beta, the content service rebuilt from this branch:
  • discover for an event (by stId and by dbId), a non-event (422) and two unknown ids, plus an icon and an unknown icon, were all identical before and after;
  • a species id answers its 422 at once.

🤖 Generated with Claude Code