It was a misconfiguration we applied to a router in Atlanta during routine maintenance. That caused bad routes on our private backbone. As a result, traffic from any locations connected to the backbone got routed to Atlanta. It resulted in about 50% of traffic to our network to not resolve for about 20 minutes. Locations not connected to our backbone were not impacted. It was a human error. It was not an attack. It was not a failure or bug in the router. We're adding mitigations to our backbone network as we speak to ensure that a mistake like this can't have broad impacts in the future. We'll have a blog post with a full explanation up in the next hour or so — being written now by our CTO.

You should really read up on BGP. It really is that flimsy.

It's not failure, it's misconfiguration. Totally different issue to be resilient to fat fingers than fires.