Raise catalog cache TTL to 24h and optimize global search

[tiankaima]

Summary

• Raise PUBLIC_DETAIL_RUNTIME_CACHE_TTL_MS and PUBLIC_DETAIL_KV_CACHE_TTL_MS from 60s/1h to 24h (revision-scoped keys still invalidate on static import).
• Optimize /api/search by replacing paginated list summaries (COUNT + heavy includes) with lightweight findMany queries.
• Add a 120s in-memory response cache keyed by locale + query + limit + user.

Motivation

Production metrics showed ~0.06% section KV hit rate and ~956ms avg search latency. The 60s L1 TTL was too short for high-cardinality section traffic; search was paying for COUNT queries and full summary includes on every request.

Test plan

• bunx vitest run tests/unit/global-search-service.test.ts
• Deploy and verify section kv_hit rate improves on repeat traffic (perf(catalog): section detail KV cache hit rate near zero in production #709)
• Verify /api/search p95 drops on production

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Updated (UTC)
✅ Deployment successful! View logs	life-ustc	a49715e	Jul 31 2026, 10:21 AM

[tiankaima]

Closing as superseded by merged work on main.

Already in main via #712 (5bc4e4a5):

• 24h catalog detail L1/KV TTL (PUBLIC_CATALOG_RUNTIME_CACHE_TTL_MS / PUBLIC_CATALOG_KV_CACHE_TTL_MS in catalog-runtime-cache.ts, re-exported from catalog-detail-runtime-cache.ts)
• Global search query optimization (global-search-catalog-queries.ts — lightweight findMany instead of paginated list summaries with COUNT)
• Catalog search caching via cachedCatalogRuntimeData (revision-scoped L1/KV/colo, 300s TTL) — supersedes this PR's simpler 120s in-memory global-search-response-cache.ts Map

Already in main via #713 (d4cde989):

• GlobalSearchDialog lazy-load (client bundle / E2E budget)

This branch (a49715e) predates #712/#713 and now conflicts on global-search-service.ts, global-search-catalog-queries.ts, catalog-detail-runtime-cache.ts, and tests. No unique valuable changes remain to cherry-pick.

Related: #709 — TTL raise landed in #712; post-deploy KV hit-rate measurement still tracked there.