Cache hierarchies: CDN → gateway → app → DB

A layered cache turns latency into throughput. Start at the edge (CDN), then the gateway, then your app (in‑memory/Redis), and finally database query/result caches.

Design principles

• Make responses cacheable by default; add Cache-Control and ETag.
• Use read‑through caches for hot paths and background refresh for stability.
• Separate idempotent reads from writes; invalidate precisely with keys/scopes.

Metrics to watch

• TTFB, hit ratios at each layer, p95 latencies, DB QPS.
• Error amplification and thundering herd protection (jitter, request coalescing).