web-performance（网络性能）

Web Performance Skill

Version: 1.0 Stack: Apollo Client/Server + Redis + CloudFront + S3

Performance problems are invisible during development. Local servers are fast. Test data is small. You don't notice the 3MB bundle, the N+1 queries, or the full-table scan until real users on real connections with real data report that the page takes 8 seconds to load. By then, fixing it requires rearchitecting the caching layer, splitting the bundle, and rewriting queries — all at once, under pressure.

Performance patterns applied from the start are cheap. Performance fixes applied after launch are expensive. These rules are the cheap version.

Core Principles

1. Measure First — Profile before optimizing. Don't guess.
2. Cache at Every Layer — CDN → API → Database → In-Memory.
3. Minimize Payload — Send only what's needed.
4. Defer Non-Critical — Load critical path first, everything else later.
5. Perceived Performance — Optimistic UI makes things feel faster.

Core Web Vitals Targets

Apollo Client Caching

Configure InMemoryCache with type policies for pagination merging and computed fields. Use cache-and-network as default fetch policy (stale-while-revalidate). Use optimistic updates for mutations to make UI feel instant.

Fetch Policies

See assets/caching-patterns.md for cache configuration, fetch policy examples, and optimistic update patterns.

Redis Caching

Cache-aside pattern: check cache → miss → fetch DB → store in cache. Invalidate on writes. Use tag-based invalidation for list queries.

TTL Guidelines

See assets/caching-patterns.md for Redis cache service, cache-aside pattern, and invalidation examples.

CloudFront & S3 Optimization

Immutable assets (hashed filenames) get 365-day cache. API routes get no CDN caching. Images get resize-on-demand with WebP format, lazy loading, and 2x srcSet for retina.

See assets/caching-patterns.md for CDK configuration and image optimization component.

Bundle Optimization

Route-based code splitting with React.lazy. Dynamic imports for heavy libraries (load only when needed). Analyze with source-map-explorer.

Bundle Size Guidelines

See assets/optimization-patterns.md for code splitting, dynamic import, and bundle analysis examples.

React Performance

Use useMemo for expensive computations, useCallback for stable callbacks passed to memoized children, and memo for list item components. Don't memoize simple values, always-changing values, or handlers on non-memoized children. Use virtualization for long lists (100+ items).

See assets/optimization-patterns.md for memoization, when-NOT-to-memoize, and virtualization examples.

Database Performance

Select only needed fields. Use include sparingly. Use raw queries for complex aggregations. Add compound indices on filtered/sorted columns.

See assets/optimization-patterns.md for Prisma query optimization and index strategy.

Anti-Patterns

Checklist

Frontend

• Code split by route
• Heavy libraries lazy loaded
• Images lazy loaded
• Images optimized (WebP, sized)
• Bundle < 200KB initial
• useMemo/useCallback where beneficial
• Long lists virtualized

Apollo

• Cache policies configured
• Pagination merges correctly
• Optimistic updates for mutations
• Fetch policy matches data freshness needs

Redis

• Hot data cached
• TTLs appropriate for data type
• Cache invalidation on updates
• Connection pooling configured

Database

• Indices on filtered/sorted columns
• Select only needed fields
• N+1 queries eliminated (DataLoader)
• Expensive queries analyzed

CDN

• Static assets on CloudFront
• Immutable assets cached forever
• Dynamic content not cached at edge
• HTTPS enforced

When to Consider Alternatives

References

• assets/caching-patterns.md — Apollo cache config, Redis, CloudFront, image optimization
• assets/optimization-patterns.md — Bundle splitting, React memoization, virtualization, DB optimization