A senior network engineer interview question that trips up more candidates than it should: "You've got two fully diverse uplinks to your ISP, BGP peering on both, one goes down. How fast does traffic fail over?" The instinct answer is "immediately, that's the whole point of having two." The interviewer pushes back: "Walk me through what BGP actually does when the link dies quietly, no clean interface-down signal, just an upstream device that stops forwarding while the physical light stays green."
Most candidates go quiet there, because the honest answer isn't flattering. BGP doesn't know anything is wrong until it stops hearing from its peer and the hold timer runs out. The RFC 4271 default is a 90-second hold time with keepalives every 30 seconds. Cisco's defaults are more conservative still: 180 seconds hold, 60-second keepalives. That means a "fully redundant" path can sit black-holing traffic for up to three minutes, not because the redundancy failed, but because the routing protocol is patiently waiting out a timer nobody thought to change.
The fix is the thing that actually shows up in the job, not the interview answer: Bidirectional Forwarding Detection. BFD runs its own lightweight hello packets, often every 300 milliseconds, and after a handful of missed intervals it declares the session down and tells BGP immediately instead of waiting for a keepalive to lapse. That turns a worst case of up to three minutes into a worst case of under a second.
The part worth sitting with isn't the trivia, it's what it says about the job itself. "Redundant BGP path" doesn't mean automatic instant failover. It means failover bounded by whatever timers are actually configured, and whether anyone bothered to wire BFD into the session. That gap between "the design looks redundant on a diagram" and "the design behaves like it's redundant during a real outage" is exactly the kind of judgment that separates a senior network engineer from someone who can draw a topology.
Codelivly's Network Engineer L3 Book is built around that judgment: advanced BGP behavior, spine-leaf data-center design and cloud architecture case studies, pitched at the senior-engineer level this question comes from. The free Network Design and Architecture and WAN, VPN and Cloud Networking paths on codelivly.com are a solid place to build the underlying architecture ground first.
Top comments (0)