Description
This is typical VXLAN/EVPN config that used to work with one-deploy:
vn:
evpn0:
managed: true
template:
VN_MAD: vxlan
VXLAN_MODE: evpn
IP_LINK_CONF: nolearning=
PHYDEV: eth1
AUTOMATIC_VLAN_ID: "YES"
GUEST_MTU: 1450
AR:
TYPE: IP4
IP: 172.17.2.200
SIZE: 48
NETWORK_ADDRESS: 172.17.2.0
NETWORK_MASK: 255.255.255.0
GATEWAY: 172.17.2.1
DNS: 1.1.1.1router:
hosts:
u1a1: { ansible_host: 10.2.80.10 }
node:
hosts:
u1b1: { ansible_host: 10.2.80.20 }
u1b2: { ansible_host: 10.2.80.21 }Unfortunately this no longer works with newer versions of FRR like 10.6.0. Older versions like 10.2.1 in SLES 15 work just fine.
u1b1# show bgp l2vpn evpn vni 2
VNI not found
When you force VNI like this:
address-family l2vpn evpn
neighbor fabric activate
advertise-all-vni
vni 2
exit-vni
exit-address-family
You can see:
u1b1# show bgp l2vpn evpn vni 2
VNI: 2
Type: L2
Tenant-Vrf: default
RD: 10.2.80.20:2
Originator IP: 10.2.80.20
Mcast group: 0.0.0.0
MAC-VRF Site-of-Origin:
Advertise-gw-macip : Disabled
Advertise-svi-macip : Disabled
SVI interface : unknown
Import Route Target:
65000:2
Export Route Target:
65000:2
From the intial investigation it follows that zebra is unable to associate VNI with the bridge SVI interface : unknown and the only way to make EVPN work is:
ip link set eth1.2 type vxlan local 10.2.80.20
where 10.2.80.20 in this particular environment is the main IP of the HV node.
It seems to me that VXLAN/EVPN solution needs to be re-evaluated.
To Reproduce
You can use one-deploy with config suggested above in Ubuntu 24.04 for example, or deploy manually as described in OpenNebula documentation, then deploy 2 VMs and try to ping between them. In SLES 15 similar config will just work without problems.
Expected behavior
VNI should be detected correctly, BGP should propagate routing data.
Details
- Affected Component: Networking
- Hypervisor: KVM
- Version: 7.2 (but it's more related to FRR version)
Additional context
It seems downgrading to 10.5 helps, so can be considered as a temporary workaround.
Progress Status
Description
This is typical VXLAN/EVPN config that used to work with one-deploy:
Unfortunately this no longer works with newer versions of FRR like 10.6.0. Older versions like 10.2.1 in SLES 15 work just fine.
When you force VNI like this:
You can see:
From the intial investigation it follows that zebra is unable to associate VNI with the bridge
SVI interface : unknownand the only way to make EVPN work is:where 10.2.80.20 in this particular environment is the main IP of the HV node.
It seems to me that VXLAN/EVPN solution needs to be re-evaluated.
To Reproduce
You can use one-deploy with config suggested above in Ubuntu 24.04 for example, or deploy manually as described in OpenNebula documentation, then deploy 2 VMs and try to ping between them. In SLES 15 similar config will just work without problems.
Expected behavior
VNI should be detected correctly, BGP should propagate routing data.
Details
Additional context
It seems downgrading to 10.5 helps, so can be considered as a temporary workaround.
Progress Status