>-------- Оригинално писмо -------- >От: Jeroen Massar jer...@massar.ch
>Относно: Re: Failover BGP servers? >До: Yasen Atanasov <yasen6...@abv.bg> >Изпратено на: 29.03.2024 12:41 > > > > On 29 Mar 2024, at 10:48, Yasen Atanasov <yasen6...@abv.bg> wrote: > > > > Hello, > > Lets imagine I have 2 BGP servers announcing same subnet. BGP sessions with > > upstream are in different VLAN. Both servers have keepalived and all > > clients are using for default route the virtual address. The idea is if I > > need to upgrade/rebuild one of the servers the clients will still have > > connectivity trough the other server. Do I need to do something special in > > the bird config or just announcing the same net is enough? > > Just announce the same net and ensure you do not have 'state': connection > tracking... otherwise you need to do connection tracking state handover and > things become nasty very quick. > Hi, > Also, in keepalived or as a manual trigger have a script that removes the > announce for the prefix the moment it detects it does not have proper > connectivity. Doesn't that imply that there is proper communication between failing server and upstream BGPs? I get the impression that constantly announcing from both of my servers is not a good idea. Is that true? >One choice there is then also if only the primary does announce for incoming >traffic or not, traffic load balancing can be a good thing, but might not work >unless your upstream does ECMP for you. > > As you note VLAN, ensure that that VLAN has 2 different physical underlying > paths, otherwise you will lose connectivity anyway if a single link that > underlies it dies. > > If it is really a single link, one does not really need separate VLANs btw, 1 > remote router, 2 IPs for your side is then enough. A separate VLAN can be > nice to cleanly separate though. > > BFD and low timers also come to mind too if you really want to do 'failover' > and not 'I know it is going down soon, lets stop announces for the prefix > from the router I am going to play with'. > > Greets, > Jeroen
