Trying to understand the Facebook 2021 outage and the role of BGP

The Facebook 2021 outage was triggered by a routine BGP route withdrawal that became catastrophic due to a simultaneous loss of internal network reachability, making the routes impossible to re-advertise.

Engineers ran a command that disconnected Facebook's backbone, stopping its internal servers from communicating. This broke the control plane that managed its external BGP sessions. Those BGP sessions then terminated, causing Facebook's routers to automatically send BGP withdraw messages to the internet, deleting all routes to its IP prefixes. Global internet routers, receiving these withdrawals, instantly stopped forwarding all traffic to Facebook because the routes were no longer in their routing tables. The internal network failure prevented Facebook from remotely re-establishing its BGP sessions or accessing its own routers to push a fix, as the very infrastructure needed to resolve the issue was unreachable.

The core concept is that BGP is the internet's routing protocol for exchanging reachability information between autonomous systems. A route withdrawal is a standard BGP update message that informs peers a path is no longer valid. In this case, it was a correct advertisement of a loss of reachability, but the cause was a fatal loss of internal network control.