Home Timeline: Fan-out on Write Diagram

The Naive Way: Polling

Client re-runs the timeline query every 5 seconds
10M online users means 2M timeline queries per second
Each query scans 200 followed accounts: 400M lookups/s

Fan-out on Write

Precompute each follower home timeline as a materialized view
5,800 posts/s x 200 followers = 1.16M timeline writes/s
Reads become a simple cache hit instead of joins and sorts

Extreme Cases

Users following thousands of accounts: dropping some writes is OK
Celebrity posts: store separately and merge at read time
Adding celebrity posts to millions of timelines is too expensive

The Trade-off

A materialized view speeds up reads and costs extra writes
During spikes, deliveries are queued: slightly stale, still fast
Timelines keep loading fast because they come from cache