Posted on September 9, 2019 5:10 pm
 |  Asked by Jason Van Patten
Print Friendly, PDF & Email

Hey folks –

I’m playing around on a KVM server with vEOS 4.22.0F, trying my hand at the EVPN/ESI multihoming. The basic network is 2 x Spines connected to 2 x Leafs with a single Linux (Debian) VM connected to both Leafs. VXLAN with EVPN is running between the 4 vEOS devices and the EVPN multihoming seems to be working well. The server thinks it has an 802.3ad bundle, and packets can flow through either link. All good.

The challenge I’m running into is detecting and/or simulating link failure properly. If, for instance, I “shut” the Ethernet interface on the Leaf, the server still thinks it’s up and continues trying to respond through it. Packets get dropped. Vice versa, if I “ifconfig eth0 down” on the server, the same thing happens in reverse: the Leaf still thinks the interface is up and tries cramming frames down it. It’s only when I shut both sides down that the link is properly failed, and packets restart (or continue) flowing.

LACP is the solution to this, I think. The problem is, I don’t think the Linux bridge interfaces that I have on the hypervisor between the Leafs and server are passing the LACP packets. Neither end can see the other via LACP, so I can’t use that and keep the Port-Channel on the Leaf side up.

Is that expected? If I have a Linux bridge between two VMs (one being vEOS) should I expect the LACP packets to be eaten? Or is there something up, perhaps, with the config on the Linux VM? Any ideas or guesses?


Posted by RIOPEL
Answered on September 18, 2019 4:59 am

Personally I have not been able to get lacp working on vEOS, so instead running “channel-g….mode on”. Cisco 9Kv’s have a hack to get lacp working, but haven’t seen anything similar in vEOS. Please post solution if you get it working.

Post your Answer

You must be logged in to post an answer.