On Mon 05 Jan 2026 23:44:25 GMT, Lukasz Jarosz wrote: > I have N edge routers, which all of them announces exactly the same > prefixes to different upstreams. They are interconnected to downstream > segments and between each other with common switching fabric that > backs up „edge router mesh” (in iBGP sense obviously). If edge router > loses connectivity to upstreams it just stops advertising to > downstreams and other neighbors. Case of losing connectivity to either > downstreams or other neighbors only causes suboptimal routing, but > when edge router loses connectivity to switching fabric it would > become kind of zombie - advertising aggregated routes to prefixes that > it does not have access. This is also condition in which router ends > up when is sole survivor of failure, but I don’t think this would be > recoverable remotely. Also I know this might seem far fetched for > some, but jackasses with excavators already ruined many days of my > life ;)
I avoid announcing the routes directly from the edge routers. I originate them from RRs, so when the edge looses the connnection to the backbone, the announcement is naturally withdrawn. -- Alarig
