Kordant/tasks/web-production/07-caching-strategy.md

07. Caching Strategy (Redis + HTTP Cache)

meta: id: web-production-07 feature: web-production priority: P2 depends_on: [] tags: [performance, caching, production]

objective:

• Implement multi-layer caching to reduce database load and improve response times

deliverables:

• Redis caching layer for API responses
• HTTP cache headers for client-side caching
• Cache invalidation strategy
• Stale-while-revalidate pattern

steps:

1. Implement Redis caching for API responses:
   • Create web/src/server/lib/cache.ts with Redis-backed cache
   • Cache user profile: key user:{id}, TTL 5 minutes
   • Cache subscription: key sub:{userId}, TTL 1 minute
   • Cache dashboard summary: key dash:{userId}, TTL 30 seconds
   • Cache blog posts: key blog:{slug}, TTL 1 hour
2. Add cache decorators/procedures:
   • Create cachedProcedure wrapper for tRPC
   • Support cache tags for invalidation
3. Implement HTTP caching headers:
   • Static assets: Cache-Control: public, max-age=31536000, immutable
   • API responses: Cache-Control: private, max-age=30
   • HTML pages: Cache-Control: no-cache (SSR)
4. Add cache invalidation:
   • Invalidate user cache on profile update
   • Invalidate subscription cache on billing event
   • Invalidate blog cache on publish/edit
5. Implement stale-while-revalidate for dashboard data
6. Add cache hit/miss metrics

tests:

• Unit: Test cache set/get/delete operations
• Integration: Test cache invalidation on mutations
• Performance: Compare cached vs uncached response times

acceptance_criteria:

• Redis cache layer active on all read-heavy endpoints
• Cache hit rate >80% for user profile and subscription endpoints
• Cache invalidation working on all mutations
• HTTP cache headers correct per endpoint type
• Stale-while-revalidate pattern on dashboard widgets
• Cache metrics visible in monitoring dashboard

validation:

• Load test: cached endpoint p95 < 20ms
• Verify Redis keys created for cached data
• Update profile → cache invalidated, next request hits DB

notes:

• Redis already used for BullMQ jobs — share connection or use separate DB index
• Be careful caching authenticated data — always include userId in key
• Consider using Vercel KV or Upstash Redis for serverless