News Feed System

Overview

The News Feed system design (like Facebook or Twitter) is a classic interview question that tests your understanding of fan-out, feed ranking, and real-time updates. The core challenge is serving a personalized, ranked feed to millions of users efficiently. Do you push (fan-out on write) or pull (fan-out on read)? How do you rank posts? How do you handle new followers and real-time updates? This design matters in interviews because it's a well-known product and the trade-offs are nuanced—push can lead to write amplification for celebrities, pull can cause read storms. Demonstrating you understand both approaches and when to use hybrid or pre-combined feeds shows systems maturity.

Requirements

Functional

Display posts from followed users in reverse chronological or ranked order
Support post creation, like, comment, share
Real-time updates when new posts arrive
Infinite scroll (paginated feed)
Mute/hide posts or users

Non-Functional

Low latency — feed load <200ms
High throughput — millions of users, thousands of posts/sec
Consistency — new posts appear within seconds

Capacity Estimation

Assume 300M users, 500 avg followees. 10K posts/sec. Push: 500 * 10K = 5M writes/sec to feed tables. Pull: 300M reads/sec on feed load.

Architecture Diagram

Component Deep Dive

Post Service

Creates posts, stores in DB. Publishes event to message queue for fan-out.

Fan-out Service

On new post: writes to feed tables of all followers (push) or does nothing (pull). Hybrid: push for active users, pull for inactive.

Feed Service

Retrieves feed for user. Reads from pre-computed feed (push) or aggregates on read (pull). Applies ranking.

Ranking Service

Scores posts by relevance, recency, engagement. ML model or heuristic. Runs offline or at read time.

Cache

Redis. Caches user's feed (first page). Invalidate on new post (push).

Message Queue

Kafka. Decouples post creation from fan-out. Enables async processing.

Database Design

Posts table: post_id, user_id, content, created_at. Feed table (push): user_id, post_id, created_at, score. Social graph: follower → followee. Cassandra/DynamoDB for feed; MySQL/Postgres for posts.

API Design

Method	Path	Description
`POST`	/api/posts	Create post. Body: {content, media?}. Returns post_id.
`GET`	/api/feed?cursor=&limit=	Get user's feed. Paginated.
`POST`	/api/posts/{id}/like	Like a post.
`GET`	/api/posts/{id}	Get single post with comments.

Scalability & Trade-offs

Push vs pull: Push = fast reads, slow writes (celebrity problem). Pull = slow reads, simple writes. Hybrid balances both.
Ranking: Chronological is simple; ML ranking improves engagement but adds latency and complexity.
Consistency: Eventual consistency is acceptable for feed; strong consistency for likes/comments.

Related System Designs

Social