Hello Ondrej, Tim,
thanks for pointing out that only the the best route is exported. While I understand that logic, I am not sure if it is smart or really within the intention of the protocol. Let me try to shortly summarise it: - server145 receives an eBGP route for the target that is only intended to be used as a backup, thus has a low local pref - server145 also receives various iBGP routes for the target, which are better than the eBGP route according to our local_pref - Thus server145 does *NOT* export its eBGP learned route via iBGP to other nodes in the same AS <--- this part I doubt whether it is correct or smart - Because of that all other nodes in the same AS are unaware of the routes that server145 received, unless the other eBGP routes / upstreams are disappearing So in a nutshell, the "export only best route" behaviour hides potential other paths by default from the rest of the nodes and they only learn about it after the other eBGP upstream(s) are gone. Wouldn't it be much smarter if server145 would export its routes as well, with the lower local_pref attributes, as they are received by eBGP and thus represent another path outside of the AS? Note: I am not talking about which routes end up in the kernel, this is purely about route exchange within the different bgp speakers of one AS. Does it make sense what I am writing or am I totally off? Or in other words: how do you usually handle redundant upstreams that have different priority? Does everyone in the list "just waits" for one upstream to break until they routes are propagated? That seems incorrect to me. Best regards, Nico -- Sustainable and modern Infrastructures by ungleich.ch
