Design Twitter: feed fanout and timeline system design
Hybrid: Solving the Celebrity Problem
@elonmusk has 170M followers — pure push would require 170M Redis writes per tweet. The hybrid Feed Service skips fanout for celebrity accounts. On read, it merges the pre-built cache with live celebrity posts.
Request Flow
Celebrity posts — no fanout
The Feed Service checks follower count. Above the threshold, the tweet is written only to the Database. No cache writes, no queue job.
// On POST /tweet:
if (author.followerCount > CELEBRITY_THRESHOLD) {
db.insert(tweet);
return; // no fanout
}
// Regular user: proceed with push fanoutMerge on read
When @alice loads her timeline, the Feed Service fetches her pre-built Cache (regular follows) AND queries the Database for recent celebrity posts, then merges.
GET /timeline:
1. cache = LRANGE timeline:alice 0 99
2. celebPosts = db.recentPosts(
alice.celeb_follows
)
3. return merge_sort(cache, celebPosts).slice(0, 20)Why We Need This
The hybrid approach cut Twitter's write amplification by an estimated 40%. It also reduced Redis write pressure during viral moments from celebrity accounts.
Key Insight
The hybrid model is the state of the art for feed systems. Regular users get push-based O(1) reads. Celebrity content is fetched on demand and merged. The threshold between push and pull is the single most important tuning parameter in the system.
Step 6 of 6
