performance-optimization

Replace per-record queries with a single query using joins or eager loading, such as Prisma's include option. Verify the fix by checking the database query log and confirming response times drop below your p95 target.

Measure first with Lighthouse and the web-vitals library to find the real bottleneck. For LCP, optimize hero images with responsive srcset and fetchpriority; for INP, break up long tasks over 50ms using scheduler.yield() or Web Workers.

Run EXPLAIN ANALYZE first to read the query plan before adding any index. Index for the query shape with equality columns first, then range or sort columns, and revert the index if the plan does not change.

No, React.memo only helps expensive components that re-render with identical props. Overusing it adds comparison overhead without benefit, so apply it only where profiling shows wasted renders.

Pool exhaustion happens when connections are held too long by long transactions, missing awaits, or leaked clients, or when instances multiplied by pool size exceed the database's max_connections. Diagnose what holds connections before resizing, and use a proxy like pgbouncer for serverless.

Avoid caching when the underlying call is already fast, when data changes as often as it is read, or when staleness causes correctness bugs like balances or permissions. Caching adds a network hop, invalidation complexity, and stampede risk.