Twitter / X: Real-Time Timeline Generation & Fan-Out Architectures

Twitter (now X) operates under extreme read-heavy velocity, processing over 500 million tweets per day with traffic surges exceeding 150,000 tweets per second during global live events (such as the World Cup).

To architect this, Twitter divides feeds into two fundamentally different constructs:

1. User Timeline (My Tweets): All tweets posted by a single specific user. This is trivial to serve via a sharded index lookup: SELECT tweet_id FROM tweets WHERE user_id = ? ORDER BY timestamp DESC LIMIT 20.
2. Home Timeline (Followed Feed): The real-time composite stream of all tweets posted by everyone a user follows. If a user follows 1,000 accounts, computing this dynamically on every page refresh requires querying 1,000 separate user timelines, merging and deduplicating millions of records in memory, and sorting them chronologically.

Executing this query dynamically for hundreds of millions of active users would instantly crash relational database clusters.