Primary-Replica

1. The Incident

Your store runs on one PostgreSQL machine. It has for years. Every product page view, every search, every cart update, every order: one box.

Growth catches up on a Tuesday. Traffic analysis shows 8,000 reads per second (product pages, search, order history) and 300 writes per second (orders, cart changes). The box can handle about 5,000 queries per second, total. Product pages take four seconds to load. Checkout, which writes, crawls too, because reads and writes are fighting over the same CPU, memory, and disk