When I was at google there were a couple of times that "big services" went down. Every time the service was actually up, some change or another broke the connection from the gateway to the internal system. It's probably that. If it was just a bad push of a single component, usually you can revert it. If you push a bad gateway config, maybe the gateways are crashing so much you can't roll back.