Fun with BFD

Our tech team had a rude reminder of when to use bi-directional failure detection (BFD), and when not to, that led to a rather full excursion into bug hunting. We all know that BFD is best used when you have a multi-point layer two connection with multiple pieces of active equipment in the path, such as a tunnel or series of devices in bridging mode. The logic being that BFD will detect a failure on the end-to-end path when the individual physical links are mostly – but not all – working and save the effort and time of a routing protocol reconvergence. The issue that arose was that one of our switches was retaining a path in hardware, when the software had marked it as deleted, so the hardware would forward, but the software would not. Turns out, there’s a bug which causes point-to-point OSPF sessions to fail. If you see some early morning VLANs being rearranged, rest assured your packets are being passed while we purge the state tables. It also means we’re removing unnecessary protocols on direct point-to-point links but will retain it where is makes sense such as our intercapital links.