Sharding

1. The Incident

The replicas worked. Reads spread across three machines, and product pages are fast. The site survived its first primary failover, when a replica took over from a dead primary. Eighteen months pass. The store grows.

Now the primary is failing, in a way no replica can help. Writes have grown from 300 per second to 2,500 per second: orders, cart updates, inventory changes, reviews. Remember the trap from the last lesson: every replica must apply every write to stay current