haroldgsm
Member
OP
Grumpy Old Sysadmin
- Joined:
- May 2024
- Posts:
- 329
- From:
- Ohio, US
Running a pair of route reflectors for our iBGP mesh at two sites. Clean setup, or so I thought. Site A advertises 203.0.113.0/24 to its RR. Site B advertises 198.51.100.0/24 to its RR. Both RRs peer with each other, standard cluster.
The problem: Site A's RR is reflecting 203.0.113.0/24 to Site B's RR, and vice versa. I have prefix-lists applied on both RRs outbound to the other RR. The lists explicitly permit only the local site's prefix. Yet both prefixes show up everywhere.
ip prefix-list SITE-ONLY seq 5 permit 203.0.113.0/24
!
route-map OUTBOUND permit 10
match ip address prefix-list SITE-ONLY
!
neighbor 10.0.0.2 route-map OUTBOUND out
Show ip bgp on the receiving RR still shows both prefixes. What am I missing? Back when we did this with static full meshes, at least you knew who to blame. Kids these days with their reflectors and their filters that don't filter.
Mark my words, it's something stupid. Twenty years in this business and I still fall for the obvious.
IPv4, IRC, and irssi — fight me
moebig
Member
RAID is not backup
- Joined:
- Jul 2024
- Posts:
- 115
- From:
- Casablanca, Morocco
Wesh harold
The prefix-list is fine but check your direction, incha'allah the route map ain't "out" on the eBGP session by mistake? RRs are iBGP, don't mix them up
I seen this: the "out" filter on iBGP RR don't work like we think, the RR reflects before the filter or some shit like that
RAID 1: because paranoia pays
kenji3
Member
- Joined:
- Jul 2024
- Posts:
- 138
- From:
- Osaka, JP
Route reflectors process outbound route-maps differently. The reflector reflects routes to clients and non-clients before outbound filtering applies to the peering session itself. I encountered this in our Singapore POP.
You may need to filter at ingress from your clients, or use a different approach. What does your topology look like—are both RRs clients of each other, or is one the main reflector?
conbini > datacenter snacks
MARIA3
Member
- Joined:
- Jul 2024
- Posts:
- 126
- From:
- Madrid, ES
Kenji's right, filter at ingress from the client instead.
siesta first, deploy later