Posted on October 8, 2018 7:31 pm
 |  Asked by Raphael Mazelier
 |  92 views
Tags:
0
0
Print Friendly, PDF & Email

Here my scenario :


sw1 sw2
! !
ebgp ebgp
! !
servers x1 servers x2

Basically I have some servers which peer locally on each arista switchs.
All the servers advertise some /32 (via bird), which I want to be equally distributed on each switch.

router bgp 49477
router-id 193.169.66.2
graceful-restart restart-time 300
bgp always-compare-med
maximum-paths 64
bgp listen range 193.169.66.96/27 peer-group prdinfcdn remote-as 65042
no bgp bestpath ecmp-fast
neighbor ipv4-ibgp peer-group
neighbor ipv4-ibgp remote-as 49477
neighbor ipv4-ibgp update-source Loopback0
neighbor ipv4-ibgp additional-paths receive
neighbor ipv4-ibgp additional-paths send any
neighbor ipv4-ibgp maximum-routes 12000
neighbor prdinfcdn peer-group
neighbor prdinfcdn remote-as 65042
neighbor prdinfcdn maximum-routes 200
neighbor 193.169.66.1 peer-group ipv4-ibgp
neighbor 193.169.66.1 next-hop-self
redistribute connected route-map connected-to-bgp
!
address-family ipv4
neighbor ipv4-ibgp activate
neighbor ipv4-ibgp graceful-restart
neighbor prdinfcdn activate
neighbor prdinfcdn route-map ipv4-prdinfcdn-in in
neighbor prdinfcdn route-map ipv4-prdinfcdn-out out

I want one path by server on both switch. So In my example 3 paths by switchs.
But on my example I have only :

1 path on sw1 (from the direct peer attached)
2 paths on sw2 (from the 2 direct peers).

If I shut the peer on sw1, I got the two pathq from ibgp (thanks to additionnals paths); but it was up it was preferred.


prdnetswr501(config)#sh ip bgp 193.169.66.132/32 detail
BGP routing table information for VRF default
Router identifier 193.169.66.1, local AS number 49477
BGP routing table entry for 193.169.66.132/32
Paths: 3 available
65042
193.169.66.66 from 193.169.66.66 (193.169.66.66)
Origin IGP, metric -, localpref 100, weight 0, valid, external, best
65042
193.169.66.2 from 193.169.66.2 (193.169.66.2)
Origin IGP, metric -, localpref 100, weight 0, valid, internal, ECMP head, ECMP contributor
Not best: Peer Type
Rx path id: 0x1
65042
193.169.66.2 from 193.169.66.2 (193.169.66.2)
Origin IGP, metric -, localpref 100, weight 0, valid, internal, ECMP, ECMP contributor
Not best: Peer Type
Rx path id: 0x3
Advertised to 3 peers:
10.247.161.9 193.169.66.35
peer-group ipv4-ibgp:
193.169.66.2 (1 add-path)

My question is : is there an option to have both path (I think the problem is have a mix of ibgp/ebgp). ?

0
Posted by ad
Answered on October 8, 2018 9:17 pm

Hi,
The ASCII diagram does not seem to present very well in the resulting formatting.
How are the SW1 and SW2 connected, just iBGP? Is there no dual-homing from the servers?

What routing mode do you use, Multi-agent or standard, RibD?

The statement seem normal, but if SW1 and SW2 are iBGP peers then they should exchange each other prefixes.
1 path on sw1 – from the direct peer attached
2 paths on sw2 – from the 2 direct peers.

Note that you may want to enforce the BGP administrative distance with the BGP distance command to influence the eBGP vs iBGP selection preference away from the default : 200 for both iBGP and eBGP! ref: http://www.arista.com/en/um-eos/eos-section-31-4-bgp-commands#ww1116652

For example:
!
router bgp 123
distance bgp 20 200 200
!

Are the subnets to the servers advertised by an IGP? If not then maybe you need to configure next-hop-self to the IBGP peers. But from your output you don’t seem to have reachability issues. I don’t see the content of your route-maps such as the one named connected-to-bgp.

I am not clear what you are trying to achieve. If you want all the prefixes to be learned by eBGP rather than a mix of iBGP+eBGP then maybe SW1 and SW2 should b eBGP peers rather than iBGP. You could correct the AS-Path.

Post your Answer

You must be logged in to post an answer.