Load Balancing

1. The Incident

The stateless fleet is live: six identical servers behind that "simple traffic splitter" from Tuesday. The splitter does exactly one thing. Request one goes to server 1, request two to server 2, around the circle forever. It seems fine. It is fine, until 2:14 AM.

Server 4's disk fills, because of a runaway log file. It does not crash. Worse, it degrades: requests start taking 20 seconds, then failing outright. The splitter knows nothing about health, so it keeps doing its one job