Skip to main content

Design Twitter

Problem​

Design Twitter with timeline, tweets, follow/unfollow.

Requirements​

  • Post tweets (280 chars)
  • Timeline (following users' tweets)
  • Follow/unfollow users
  • Scale: 300M DAU, 100M tweets/day

Key Challenges​

  1. Fanout problem: User posts → fanout to followers' timelines
  2. Celebrity problem: Can't fanout to 100M followers
  3. Hot keys: Trending tweets (cache)

High-Level Design​

Data Model​

CREATE TABLE users (
  user_id BIGINT PRIMARY KEY,
  username VARCHAR(50) UNIQUE
);

CREATE TABLE tweets (
  tweet_id BIGINT PRIMARY KEY,
  user_id BIGINT,
  content VARCHAR(280),
  created_at TIMESTAMP
);

CREATE TABLE follows (
  follower_id BIGINT,
  followee_id BIGINT,
  PRIMARY KEY(follower_id, followee_id)
);

Timeline Generation​

Fanout-on-Write (Push Model)​

User posts tweet → Push to all followers' caches
Timeline read: Read from cache (fast)

Pros: Fast reads
Cons: Slow writes for celebrities

Fanout-on-Read (Pull Model)​

User posts tweet → Store in DB
Timeline read: Query all followed users, merge tweets

Pros: Fast writes
Cons: Slow reads (many queries)

Hybrid (Twitter's Approach)​

  • Regular users: Fanout-on-write
  • Celebrities (>1M followers): Fan out-on-read
  • Merge both in timeline

Scalability​

  • Cache: Redis cluster (timeline = sorted set by timestamp)
  • Sharding: User ID for tweets/follows
  • Message queue: Async fanout (Kafka)

Next: Design Instagram.