The problem

An order service writes to one database, and every order gets its id from an auto-increment column. Sales grow. The team splits the orders over 4 databases, and each one keeps its own counter.

Two weeks later, support finds two different orders with id 1 000 017, one in each of two databases. A refund for one customer goes to the other.

The fix the team wants: every server makes its own ids, with no shared counter, and two ids never collide.