nextjs-caching-architecture

Caching Architecture

Priority: P1 (HIGH)

Next.js has 4 distinct caching layers. Understanding them prevents stale data bugs.

The 4 Layers

1. Request Memoization

• Behavior: Calling getUser(1) in layout, page, and component only triggers 1 network call.

• Key: Matches URL and Options exactly.

• Non-Fetch: For DB calls/Prisma, use React cache() manually.

  import { cache } from 'react';
  const getItem = cache(async (id) => db.item.findUnique({ id }));
  

2. Data Cache (The "Server Persistence")

• Default: fetch tracks cache indefinitely (force-cache).

• Scaling:

  • Vercel: Shared globally via KV/Blob.
  • Self-Hosted: Stored in filesystem (Pod ephemeral storage). Critical: Must configure a shared cacheHandler (Redis) for multi-pod Kubernetes setups, otherwise pods have desynchronized caches.

• Granular Control: Use unstable_cache for DB queries caching.

  import { unstable_cache } from 'next/cache';
  const getCachedUser = unstable_cache(
    async (id) => db.user.find(id),
    ['users-key'],
    { tags: ['users'], revalidate: 60 },
  );
  

3. Router Cache (The "Client Stale")

• Problem: User updates a listing, navigates back, and sees old data.
• Cause: Client-side Router Cache holds payload for 30s (dynamic) or 5m (static).
• Fix: Call router.refresh() in a Client Component or revalidatePath() in a Server Action (which automatically invalidates Router Cache).

Strategies

• Opt-Out (Real-Time):
  • Data: fetch(..., { cache: 'no-store' })
  • Route: export const dynamic = 'force-dynamic'
• On-Demand (CMS/Admin):
  • Use revalidateTag('collection') for surgical updates. Better than revalidatePath (which nukes the whole URL).