Cambios implementados:
1. Redis cache para _classify_cache (catalog_service.py):
- Reemplaza dict in-memory por Redis compartido entre workers
- TTL 5 minutos para clasificación Nexpart
- classify_cache_clear() y classify_cache_stats() actualizados
- Hit rate pasa de ~6% (15 cachés separados) a ~80%+ (cache unificado)
2. Redis cache para vehicle info en smart_search():
- Verifica Redis antes de ejecutar DISTINCT ON + 4 JOINs sobre 2B filas
- Cache miss: query solo para los part_ids faltantes
- TTL 1 hora por part_id
- Impacto: búsquedas repetidas pasan de 500ms–2s a < 50ms
3. Gunicorn gthread (gunicorn.conf.py):
- worker_class = 'gthread' con 4 threads por worker
- 4 workers × 4 threads = 16 requests concurrentes
- max_requests = 1000 para reciclar workers y prevenir memory leaks
Tests: 73/73 pasando
- Gunicorn production server with auto-scaled workers, run.sh, updated systemd service
- Marketplace B2B: cross-tenant inventory search, ordering, seller management with full UI
- Subscription billing: plan limits enforced on products/employees/branches, billing API + upgrade flow
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
