no, I want routes exactly to carp. the scenario is the following:
1) two servers decide who is MASTER and who is BACKUP on carp (both internal and external networks), so, from any point of view they behave as a single server (which is exactly what carp was developed for. 2) MASTER learns routes from network (eBGP) 3) BACKUP learns routes from MASTER (since it has no direct connection while it is in BACKUP state) 4) as soon as MASTER goes down, BACKUP takes it's role (with already learnt routing table) so, there's no point in iBGP routes via MASTER, as we will need routes when MASTER goes down. 2010/3/6 Claudio Jeker <cje...@diehard.n-r-g.com>: > On Sat, Mar 06, 2010 at 08:45:24PM +0500, ???? ??????? wrote: >> 2010/3/6 Henning Brauer <lists-open...@bsws.de>: >> > of course there are (many) working bgpd + carp setups. >> > >> > * ???? ??????? <chipits...@gmail.com> [2010-03-06 15:14]: >> >> second router learns routes from carp master (since it has no direct >> >> connection while it is BACKUP), but I only see routes using "bgpctl >> >> show rib", not using "netstat -rn". also, there's seems to be loop, >> >> because I see many "icmp dup" while pinging some system. also, when >> >> BAACKUP becomes MASTER, it drops all the rib and learnes routes from >> >> uplink.which is very bad. >> > >> > the preferred setup is to have one bgp session per carp host to your >> > upstream, i. e. you had two, and use carp on the inner interface. >> >> is it possible to have carp on both inner and upstream interfaces ? >> any example configs for that ? >> >> > >> > what you are seeing is kinda expected, the routes are invalid from the >> > backup host's POV since it does not have a valid route to the nexthop >> > (this is half guessed since you didn't provide any details) >> >> >> there are static routes to BGP speakers >> >> # cat /etc/hostname.carp4 >> vhid 2 pass xxxx carpdev vlan4 advskew 200 x.x.x.x/29 >> !/sbin/route add N.N.N.N1/32 x.x.x.x >> !/sbin/route add N.N.N.N2/32 x.x.x.x >> >> well, carp4 is down while the second router is in BACKUP state (and >> thus I see routes in RIB, but not in FIB ?) >> I'd expect those routes move to FIB as second router becomes MASTER >> > > I guess you want to look at "set nexthop self" so that the routes don't > point over the unavailable carp route but instead over the route the ibgp > session runs on. > > -- > :wq Claudio
