Scaling a Wikipedia-style reader to 10M+ monthly visitors

Context

The product reimagined Wikipedia as a clean, ad-free reading experience backed by the public Wikipedia REST API. Content was effectively unbounded and read-heavy, and traffic spiked unpredictably around trending topics.

Constraints

• Upstream rate limits on the Wikipedia API made naive per-request fetching a non-starter at scale.
• Read-heavy, spiky traffic demanded consistent latency, not just good averages.
• Small infrastructure budget — the answer had to be caching discipline, not horizontal brute force.

Approach

• Redis as the read layer. Article and search responses were cached with TTLs tuned per content volatility, collapsing repeated upstream calls into single fetches.
• Next.js SSR + incremental revalidation served fully rendered pages fast and kept popular articles warm.
• Request coalescing ensured a cache miss on a hot article triggered exactly one upstream fetch, not a thundering herd.

Architecture

Readers hit Next.js SSR, which reads from Redis; only cache misses fall through to the Wikipedia API, and every miss repopulates the cache for the next reader.

Results

• API responses stayed under 100ms at the cache layer.
• Scaled past 10M monthly visitors without proportional infrastructure growth.
• Upstream API load dropped sharply thanks to coalescing and TTL tuning.

What I'd do differently

I'd add a stale-while-revalidate tier from the start so that even cache-expiry moments served instantly while refreshing in the background.