Caching

In-memory (single process)

• ★ lru-cache — battle-tested LRU; the default for in-process Node caches.
• @isaacs/ttlcache — TTL-only sibling library.
• quick-lru — small, fast LRU.
• node-cache — older, simpler.

Multi-tier / multi-driver

• ★ unstorage (UnJS) — universal storage layer; drivers for memory, Redis, Cloudflare KV, S3, R2, Vercel KV, browser, fs. Default for "swap backends without changing app code." Used inside Nuxt / Nitro.
• ★ keyv — universal key-value abstraction; drivers for everything; smaller surface than unstorage.
• cache-manager — multi-tier (e.g. memory + Redis); older but solid.

Distributed (Redis-style)

• ★ Upstash Redis — serverless Redis with HTTP API; works on Workers / Lambda / edge. Generous free tier.
• DragonflyDB — drop-in Redis replacement; faster, multi-threaded.
• Redis itself — boring, durable, ubiquitous.
• KeyDB — multi-threaded Redis fork.
• Cloudflare KV — eventually consistent edge KV; good for rare-write / many-read configs.
• Cloudflare Cache API + Cache Reserve — HTTP cache from inside Workers.
• Vercel KV — wraps Upstash for Vercel.

HTTP caching (the underrated kind)

• ★ Cache-Control + ETag + a CDN — solves 80% of caching needs without a library.
• stale-while-revalidate — Cache-Control: max-age=60, stale-while-revalidate=86400. Standard, wide support.
• Cache-Control: no-store, private, public, max-age=… — get the directives right; tools like http-cache-semantics can help.
• s-maxage — only the CDN caches; browsers ignore.

Framework / data-layer caches

• Next.js unstable_cache / fetch cache — automatic per-request memoization + revalidation.
• TanStack Query — client-side data cache (see Data Fetching).
• SWR — client cache.
• Drizzle / Prisma: nothing built in; layer your own using unstorage / keyv.

CDN-level

• Cloudflare Cache Reserve + Cache API — push cache deep at the edge.
• Vercel Edge Cache + ISR — built into Next.js on Vercel.
• Fastly Compute@Edge — programmable cache.

Patterns to know

• Cache aside — read cache, miss → DB → set cache. Default.
• Write-through — every write updates cache. Avoids staleness, slows writes.
• Stale-while-revalidate — return stale, refresh in background. Best UX for non-critical data.
• Negative caching — cache 404s briefly to avoid hammering origin on missing keys.
• Single-flight / coalescing — when 1000 concurrent requests miss the cache, only one calls origin. Use p-memoize or @isaacs/abort-controller patterns.

Pick this if…

• In-process Node cache: lru-cache.
• Universal driver-swappable cache: unstorage or keyv.
• Distributed cache, edge-friendly: Upstash Redis.
• Distributed cache, traditional Node: Redis or DragonflyDB.
• Just want HTTP cache to work: Cache-Control + ETag + a CDN; don't reach for a library.
• All-Cloudflare: Cache API + KV + Cache Reserve.