Design a News Feed

You're designing the news feed for a social network like Facebook or Twitter. Users follow friends/influencers and see posts in their feed, ranked by relevance. The system must handle high-velocity post ingestion, personalized ranking, and real-time updates as users scroll.

Scale assumptions:

• 500 million daily active users (DAU)
• Average user follows 500 accounts
• 1% of followers post per day (5M posts/day, average)
• Ranking must consider: recency, engagement (likes, comments), follower count, past user interactions
• Feed latency target: <100ms to serve 20 posts
• Retention: feeds computed on-demand (no pre-fan-out due to storage cost)

Derived numbers:

• Post ingestion: 5M posts/day ÷ 10^5 sec ≈ 58 writes/sec average, ~200 peak
• Feed reads: 500M DAU × 3 feed refreshes/day ≈ 17.5M reads/day ≈ 200 reads/sec average, ~1K peak (assuming uniform)
• Candidate generation: per-user feed = scan recent posts from followed accounts (expensive; need index)
• Ranking: score candidate posts (~1K posts per user per day) by ML model (latency-critical)
• Hot users: celebrities with 50M+ followers amplify post reach (fanout bottleneck)