Design Twitter: feed fanout and timeline system design
Pull Model (Fanout-on-Read)
The simplest approach: when @alice loads her feed, the API queries the Database for posts from every account she follows, merges, and sorts.
Request Flow
@bob posts a tweet
The tweet is written once to the Database. No fanout occurs at write time β fast writes, expensive reads.
POST /tweet { text: "Hello world" }
β INSERT INTO tweets (author_id, text, ts)
β 200 OK // one DB write, done@alice loads her timeline
The API issues one query per followed account to the Database, fetches their recent posts, then merges and sorts.
GET /timeline β SELECT * FROM tweets WHERE author_id IN (id1, id2, ... id500) ORDER BY ts DESC LIMIT 20 // 500 index lookups + in-memory merge sort // Latency grows linearly with following count
Why We Need This
Pull is trivially correct β every read is always fresh. But it scales poorly: latency grows with following count, and popular accounts hit the DB from millions of followers' queries simultaneously.
Key Insight
Pull = O(following_count) queries per feed load. Works at low follow counts. Beyond ~50 follows, you need caching or pre-computation β or both.
Step 1 of 6
