The problem

Tuesday 28 February 2017, Northern Virginia (US-EAST-1). You are on the Amazon S3 team. The S3 billing system is running more slowly than expected, and the team is debugging it.

At 9:37 a.m. PST, a team member follows an established playbook and runs a command. It should take a small number of servers out of one S3 subsystem that the billing process uses.

Then S3 in the region stops answering GET, LIST, PUT and DELETE requests. The S3 console fails. So do new EC2 instance launches and AWS Lambda.