BGP Networks not advertising after an adjacency failure

Douglas Bradfield · ‎05-31-2012

We have had some BGP adjacency failures and not all network advertisements recover. The neighbor recovers in about 9 seconds but one network does not recover properly. It still shows up in the routing table but we cannot get access to the far end devices until we do a shut, no shut on the internal gig interface. At no time does any of the circuits or links go down. This is a multi-homed environment on two different providers.

router bgp 65000

bgp log-neighbor-changes

network 10.10.0.0 mask 255.255.0.0

network 10.10.0.254 mask 255.255.255.255

timers bgp 5 15

neighbor 10.12.30.80 remote-as 65001

neighbor 10.12.30.80 description local router

neighbor 10.12.30.80 ebgp-multihop 255

neighbor 10.12.30.80 soft-reconfiguration inbound

neighbor 10.12.30.80 route-map RM--CC out

neighbor 10.10.0.253 remote-as 65000

neighbor 10.10.0.253 update-source Loopback0

neighbor 10.10.0.253 next-hop-self

neighbor 10.10.0.253 soft-reconfiguration inbound

May 30 12:15:42.351 EDT: %BGP-5-ADJCHANGE: neighbor 10.12.30.80 Down BGP Notification sent

May 30 12:15:42.351 EDT: %BGP-3-NOTIFICATION: sent to neighbor 10.12.30.80 4/0 (hold time expired) 0 bytes

May 30 12:15:42.354 EDT: %BGP_SESSION-5-ADJCHANGE: neighbor 10.12.30.80 IPv4 Unicast topology base removed from session BGP Notification sent

May 30 12:15:51.423 EDT: %BGP-5-ADJCHANGE: neighbor 10.12.30.80 Up

The 10.10.0.0/16 will not recover but the 10.10.0.254 will.

Giuseppe Larosa · ‎05-31-2012

Hello Douglas,

10.0.0.254/32 is likely the IP address of the loop0 on the local node.

How is 10.0/16 learned by the node?

there is a static route for IP prefix 10.0/16 pointing to the internal interface?

Is learned by an IGP?

Hope to help

Giuseppe

Douglas Bradfield · ‎05-31-2012

The 10.0.0.254 is the loopback

There is a corresponding 10.0/16 static route pointing back to internal Interface.

Also this has been up and running without issue for a year.

But this same exact incident has happened twice in the last two weeks. No config changes have been made.

Have you ever heard of anything like this?

lgijssel · ‎05-31-2012

When checking the bgp advertized routes, is the missing route listed?

If the other side also has soft reconfiguration enabled, you may try a soft reset.

clear ip bgp 10.12.30.80 soft out

Alternatively, there is this link:

http://www.cisco.com/en/US/tech/tk365/technologies_tech_note09186a00800945ff.shtml

Hope this helps.

regards,

Leo

Douglas Bradfield · ‎05-31-2012

The missing route is listed, but it doesn't work. once we do a shut, no shut, the network entry "restarts" and in the routing table it restarts the time of the route addition. see below

B 10.0.0.0/16 [20/0] via 10.12.30.82, 1d01h

B 10.0.0.253/32 [20/0] via 10.12.30.82, 1d02h

B 10.0.0.254/32 [20/0] via 10.12.30.82, 1d02h