Kordant/tasks/web-production/06-db-connection-pooling.md

06. Database Connection Pooling & Query Optimization

meta: id: web-production-06 feature: web-production priority: P1 depends_on: [] tags: [performance, database, production]

objective:

• Optimize database connections and queries for production load

deliverables:

• Connection pooling configuration
• Query performance audit
• Index optimization
• Slow query logging

steps:

1. Configure connection pooling:
   • If using PostgreSQL: configure PgBouncer or use @libsql/client pooling
   • Set max connections based on server instances (e.g., 20 per instance)
   • Add connection timeout and idle timeout settings
2. Audit all Drizzle queries for performance:
   • Check web/src/server/db/schema/*.ts for missing indexes
   • Review web/src/server/api/routers/*.ts for N+1 queries
   • Add pagination to all list endpoints (default 50, max 100)
3. Add database indexes:
   • createdAt indexes for time-range queries (alerts, exposures)
   • Composite indexes for common filter combinations
   • userId indexes on all user-scoped tables
4. Implement query result caching:
   • Cache user profile lookups (5 min TTL)
   • Cache subscription status (1 min TTL)
   • Cache dashboard summary (30 sec TTL)
5. Add slow query logging:
   • Log queries taking >500ms
   • Alert on >1s queries
6. Set up database performance monitoring

tests:

• Unit: Test query execution plans for major endpoints
• Load: Run 1000 concurrent dashboard loads, verify <200ms p95
• Integration: Test pagination boundaries

acceptance_criteria:

• Database connection pool configured with max 20 connections
• No N+1 queries in any API endpoint
• All list endpoints paginated with cursor or offset
• Slow query logging active
• Dashboard load query <100ms p95
• Alert endpoint query <50ms p95

validation:

• EXPLAIN ANALYZE on major queries shows index usage
• Load test with k6: 1000 concurrent users, p95 < 200ms
• Database CPU <50% under normal load

notes:

• Current schema has some indexes but may need more for production scale
• Drizzle ORM doesn't automatically handle connection pooling — configure at driver level
• Consider read replicas if dashboard load is heavy