System Design

Case Study: Twitter

Requirements

Functional

• Post tweets (280 chars)
• Follow users
• Timeline (home feed)
• Search tweets

Scale

• 500M users
• 200M DAU
• 500M tweets/day
• 100B timeline requests/day

Core Challenge: Timeline

Problem: User’ning timeline’ini qanday generate qilish?

Approach 1: Pull (Read-time)

-- User timeline request
SELECT tweets.* FROM tweets
JOIN follows ON follows.following_id = tweets.user_id
WHERE follows.follower_id = {current_user}
ORDER BY created_at DESC
LIMIT 20;

Slow: Million followers = heavy query

Approach 2: Push (Write-time)

// User posts tweet
async function postTweet(userId, content) {
  const tweet = await db.saveTweet(userId, content);
  const followers = await db.getFollowers(userId);
  
  // Fan out to all followers
  for (const followerId of followers) {
    await redis.lPush(`timeline:${followerId}`, tweet.id);
  }
}

Celebrity problem: 100M followers = 100M writes!

Hybrid Approach (Twitter actual)

Normal users (< 1M followers): Push model
Celebrities (> 1M followers): Pull model

Timeline = Precomputed + Celebrity tweets merged

Architecture

Write Path:
Tweet → API → Fanout Service → Redis (timelines)
                              → Cassandra (tweet storage)

Read Path:
Timeline request → Redis (precomputed)
                 → Merge celebrity tweets
                 → Return

Keyingi dars: WhatsApp loyihalash.