Course contentsShow
← System Design
Lesson 1640 of 1,91940. Real-World System: Social Media FeedFree lesson

Celebrity Problem in Push Models

Identify how accounts with millions of followers create prohibitive fan-out costs and explore mitigation strategies.

Celebrity Problem in Push Models

What You'll Learn

You'll learn why accounts with millions of followers create crushing fan-out costs in push-based feed systems and what strategies can tame this problem.

The Core Problem

In a pure push model, every time a user posts, the system writes that post to all their followers' feeds. This works beautifully for average users with hundreds or thousands of followers. But when a celebrity with 50 million followers tweets, the system must perform 50 million immediate writes—one to each follower's feed.

This creates three critical issues:

  1. Write amplification: A single post triggers millions of database writes
  2. Latency spike: The celebrity's post takes much longer to complete than a normal user's
  3. Resource exhaustion: Database connection pools, worker queues, and write capacity get overwhelmed

The Math

If writing to one feed takes 10ms and you have 10 million followers:

  • Total write time: 10ms × 10M = ~28 hours of sequential work
  • Even parallelized across 1,000 workers, that's still minutes of processing
  • Meanwhile, the database is hammered with write load

Mitigation Strategies

Hybrid approach: Keep celebrities out of the push model entirely. When a celebrity posts, don't fan-out to followers' feeds. Instead, during read time, pull recent posts from celebrities the user follows and merge them with their pre-computed feed.

Threshold-based switching: Automatically transition users to pull-based delivery once they exceed a follower threshold (e.g., 100K followers).

Selective fan-out: Only push to highly engaged followers; others get pull-based delivery.

Key Takeaway: The celebrity problem forces hybrid architectures—push for normal users, pull for high-follower accounts—balancing write efficiency with read performance at extreme scale.