The Celebrity Problem in Fanout
What You'll Learn
You'll understand why fanout-on-write breaks down when users have millions of followers, and why writing to millions of feeds becomes technically and economically impossible.
The Core Issue
In fanout-on-write (the push model you've learned), every new post gets copied into each follower's pre-built feed. This works beautifully for regular users with hundreds or thousands of followers—but what happens when a celebrity with 100 million followers posts a single tweet?
The system must now perform 100 million write operations for one post. This is called write amplification, and it creates several catastrophic problems:
Write Amplification at Scale
- Time: Writing to 100M feeds might take minutes or hours, not seconds
- Resources: Database connections, queue workers, and write capacity get overwhelmed
- Cost: Cloud providers charge per write operation—one celebrity post could cost hundreds of dollars
- Locks: Attempting millions of concurrent writes can create database contention and deadlocks
Real Numbers
If a celebrity posts once and you have 1ms latency per feed write:
- 100M writes × 1ms = 100,000 seconds = over 27 hours!
Even with parallelization, you'd need thousands of workers just to handle one post in reasonable time.
Why This Matters
The celebrity problem forces hybrid architectures: regular users get fanout-on-write (fast reads), but celebrities use fanout-on-read (their posts aren't pre-pushed). At read time, the system detects you follow celebrities and fetches their recent posts directly.
Key Takeaway: Fanout-on-write fails for high-follower accounts because write amplification makes it impossibly slow and expensive—one post can trigger millions of writes that exhaust system resources and delay delivery by hours.