fix(manager): bypass GraphQL N+1 for language cache with custom CMS REST endpoint

[tataihono]

Summary

• Problem: The manager's language cache warm fires 4 parallel GraphQL queries that produce ~20K individual DB queries (Strapi v5 has no DataLoader). This exhausts the 25-connection PostgreSQL pool and blocks all CMS requests — auth checks took 73-167 seconds during a language cache refresh.
• Fix: Created a custom CMS REST endpoint (/api/language-geo) that returns all language, country, continent, and country-language data in a single SQL query using raw knex, then updated the manager to call it instead of GraphQL.
• Pattern: Same approach as the video coverage fix (PR fix(manager): optimize video coverage query (22-47s → <1s) #637).

Changes

CMS (apps/cms/)

• New custom API: src/api/language-geo/ with route, controller, and service
• Single SQL query joins country_languages, languages, countries, continents through Strapi v5 link tables
• Filters published_at IS NOT NULL to exclude draft rows
• JS aggregation in service produces { continents, countries, languages } shape

Manager (apps/manager/)

• Replaced 4 parallel GraphQL queries + in-memory transform with a single fetch() to /api/language-geo
• Removed ~175 lines of GraphQL query definitions and aggregation logic
• SWR cache wrapper, TTL settings, and route handler unchanged

Requirements

• R1. Language/geo data fetch must not exhaust the CMS connection pool (single query vs ~20K)
• R2. Language cache refresh must complete in under 5 seconds
• R3. Auth checks remain responsive during language cache refresh
• R4. Language picker still shows all ~4,560 languages grouped by continent and country with speaker counts

Testing

• Both CMS and manager typecheck clean
• Linting passes (pre-commit hooks)
• Data contract preserved: same { continents, countries, languages } JSON shape

Post-Deploy Monitoring & Validation

• What to monitor/search
  • Logs: language-cache label in manager logs, KnexTimeoutError in CMS logs (should disappear)
  • Metrics: CMS PostgreSQL active connections during manager startup
• Expected healthy behavior
  • Language cache warm completes in <5s (vs 30s+)
  • No KnexTimeoutError during language cache refresh
  • /api/users/me responds in <1s during cache warm
• Failure signal / rollback trigger
  • Manager returns 502 on /api/languages → CMS endpoint returning errors → rollback CMS deploy
  • Language picker shows empty or missing data → data shape mismatch → rollback manager deploy
• Validation window & owner
  • Window: First 30 minutes after deploy
  • Owner: tataihono

---

🤖 Generated with Claude Opus 4.6 (1M context) via Claude Code

Co-Authored-By: Claude Opus 4.6 (1M context) [email protected]

[railway-app]

🚅 Deployed to the forge-pr-646 environment in forge

Service	Status	Web	Updated (UTC)
@forge/cms	✅ Success (View Logs)		Apr 3, 2026 at 7:25 am
@forge/manager	✅ Success (View Logs)		Apr 3, 2026 at 7:20 am

1 service not affected by this PR

• @forge/web