System Design: Designing Twitter (Timeline and News Feed)

Twitter (now X) is a massive real-time messaging system. The core technical challenge is not storing the tweets, but delivering them to millions of followers' timelines with sub-second latency.

1. Core Requirements

• Tweet Publishing: A user can post a new tweet.
• Timeline (Feed): A user can see tweets from people they follow.
• High Availability: The system must be always available.
• Scalability: Handling millions of users and high-profile "Celebrity" accounts.

2. The Fan-out Challenge

"Fan-out" is the process of delivering a single tweet to all the followers of the author.

Option A: Fan-out on Read (The Pull Model)

When a user opens their timeline, the system searches for all people they follow, fetches their latest tweets, and sorts them by time.

• Pros: Fast writes.
• Cons: Slow reads. Doing a join across thousands of authors for every timeline refresh is extremely expensive for the database.

Option B: Fan-out on Write (The Push Model)

When a user posts a tweet, the system immediately pushes a reference to that tweet into the "Timeline Cache" (usually in Redis) of every follower.

• Pros: Blazing fast reads. The user's timeline is already pre-computed in Redis.
• Cons: Slow writes. If a celebrity with 50 million followers tweets, the system must perform 50 million Redis writes immediately.

3. The Hybrid Solution: Handling Celebrities

Twitter uses a hybrid approach to balance these trade-offs:

1. Regular Users: Use Fan-out on Write (Push). Their tweets are pushed to their followers' caches immediately.
2. Celebrities (High Follower Count): Use Fan-out on Read (Pull). Their tweets are NOT pushed to millions of caches. Instead, when a follower of a celebrity views their timeline, the celebrity's tweets are merged into the timeline on-the-fly.

4. Storage & Caching

• Tweet Store: Cassandra or a similar wide-column store is ideal for storing tweets indexed by user_id and timestamp.
• Timeline Cache: Redis stores the list of tweet IDs for each user's feed.
• Media Store: Amazon S3 for images and videos, served via a CDN (CloudFront).

5. Search and Trends

• Search: Use Elasticsearch or a custom inverted index to handle hashtag and keyword searches.
• Trends: Use Apache Storm or Flink for real-time stream processing to identify "Trending" topics based on tweet frequency.

Summary

The secret to Twitter's scale is Pre-computation. By pre-calculating timelines for 99% of users and only using the "Pull" model for high-follower accounts, Twitter maintains the responsiveness that makes it a real-time platform.