Hello,
I recently tried to move from Zebra 0.93b to Quagga 0.96.4x-5 (Debian)
in an attempt to fix a crashing OSPF (which worked), but ran into a
problem with routes becoming desynchronized over our iBGP, causing
routing loops for some destinations.
I have rolled back to Zebra with no configuration changes and verified
that the problem does not occur there.
Further investigation shows that the problem is occurring for overlapping
advertised routes; for example, 24.202.128.0/17 and 24.202.128.0/18. In
our configuration, these two networks received different outgoing paths.
With Quagga, both routers received the two prefixes correctly in bgpd
("show ip bgp" shows them fine), but zebra ignored the 24.202.128.0/18
route on one of the routers. This caused the kernel to fall back to the
less specific route, which happened to be the other router, which had a
route back to the first router, causing a loop.
Has anything been changed in the code that would affect this? I can
attempt to debug this further if nothing is immediately obvious.
Thanks,
Simon-
I recently tried to move from Zebra 0.93b to Quagga 0.96.4x-5 (Debian)
in an attempt to fix a crashing OSPF (which worked), but ran into a
problem with routes becoming desynchronized over our iBGP, causing
routing loops for some destinations.
I have rolled back to Zebra with no configuration changes and verified
that the problem does not occur there.
Further investigation shows that the problem is occurring for overlapping
advertised routes; for example, 24.202.128.0/17 and 24.202.128.0/18. In
our configuration, these two networks received different outgoing paths.
With Quagga, both routers received the two prefixes correctly in bgpd
("show ip bgp" shows them fine), but zebra ignored the 24.202.128.0/18
route on one of the routers. This caused the kernel to fall back to the
less specific route, which happened to be the other router, which had a
route back to the first router, causing a loop.
Has anything been changed in the code that would affect this? I can
attempt to debug this further if nothing is immediately obvious.
Thanks,
Simon-