Design the Facebook News Feed — how would you generate and serve personalized feeds at scale?

The News Feed needs to aggregate posts from friends, pages, and groups, rank them by relevance, and serve them with sub-second latency to billions of users. Architecture: 1. Data Model: Posts table (author, content, media, timestamp), Social Graph (friend relationships, follows, group memberships), Engagement signals (likes, comments, shares, view duration). 2. Feed Generation — Fan-out approaches: - Fan-out on Write: When a user posts, push the post to all followers' feed caches. Pro: fast reads. Con: expensive writes for users with many followers (celebrity problem). - Fan-out on Read: When a user requests their feed, pull posts from all their connections at read time. Pro: cheap writes. Con: slow reads. - Hybrid: Fan-out on write for regular users, fan-out on read for users with 10k+ followers. This is what Facebook does. 3. Ranking: ML ranking model scores each candidate post. Features include: post age, author relationship strength, content type preference, engagement velocity, previous interaction patterns. The model produces a relevance score; posts are sorted by score. 4. Serving Layer: Pre-computed feeds cached in a distributed cache (Memcached/Redis). When a user opens the app, serve from cache and asynchronously refresh. Pagination via cursor-based scrolling. 5. Real-time Updates: New posts from close friends trigger a push notification to the feed service, which injects the post at the top of the cache. WebSocket for real-time feed updates when the app is open.