Search Storyflo's audio archive.

When expand=True (default) and local results are thin (< 5), we also race a Hacker News + Brave Search News + GNews fan-out and surface the top external hits as 'suggestions' the listener can forward to story@storyflo.com to ingest. External results are NOT in our DB until the listener forwards — they're discovery nudges, not silently-ingested content.

Hot-query speedup: responses are wrapped by a 60s in-process LRU+TTL cache keyed by (query, vertical, limit, offset). The watchdog smoke-tests ?q=tech every minute and the frontend hero suggests a small fixed set of common terms, so a tiny per- worker cache collapses >90% of that load to one warm response per minute. degraded=True responses are NOT cached by default (don't lock in an empty result on a statement_timeout blip). Operator can bypass per-request with X-Storyflo-Search-Cache: bypass (used by smoke probes that want to measure cold-pass latency) or globally via env SEARCH_CACHE_TTL_SEC=0.

Pool-exhaustion fix (mirrors PR #657 for /v1/search/hybrid): deliberately no db: AsyncSession = Depends(get_db_replica) parameter. With basic-plan PG sized at DB_POOL_SIZE=5 + MAX_OVERFLOW=3 (total 8 conns, see [[storyflo-prod-deploy-discipline-2026-06-19]]), a request- scoped Depends session would be held for the full duration of cache.get_or_compute — including for non-owners merely awaiting another coroutine's Future inside search_cache.py. The owner opens a short-lived replica session inside _compute and releases it before the external fan-out + Future resolution; non-owners hold zero connections while waiting.