If have tested this bug on several VyOS versions, including 1.2.3, 1.2.6, 1.2.8 and 1.3.0-epa1. With a simple BGP setup that uses an IPv6 peer-group, it is possible to trigger a few problems. When you delete an IPv6 peer with a peer-group configured, you get the following error message:
[ protocols bgp 65535 ] % Specify remote-as or peer-group commands first Error configuring routing subsystem. See log for more detailed information [edit]
If you didn't try to do any other changes while doing the commit, this isn't even a problem. Somehow the IPv6 peer is still being removed from FRR and the VyOS config.
If you did try to do other changes, e.g. deleting an IPv4 BGP peer or changing some prefix limit, then only the IPv6 BGP peer will be deleted from FRR. The other changes won't happen in FRR and at this point the FRR configuration is out of synch with the VyOS config, where those changes did "happen". The only viable solution if found after something like that happened is to either reboot the system or do the changes in FRR manually via vtysh.
This is an example configuration with which the problem can be replicated:
protocols {
bgp 65535 {
address-family {
ipv4-unicast {
network 192.168.2.0/24 { }
}
}
neighbor 192.168.0.11 {
address-family {
ipv4-unicast {
nexthop-self { }
soft-reconfiguration {
inbound { }
}
}
}
description "test peer 1"
remote-as "65535"
}
neighbor 192.168.0.10 {
address-family {
ipv4-unicast {
maximum-prefix "200"
nexthop-self { }
soft-reconfiguration {
inbound { }
}
}
}
description "test peer 2"
remote-as "65500"
}
neighbor fe80::1 {
address-family {
ipv6-unicast {
maximum-prefix "200"
}
}
description "test peer 3"
remote-as "65500"
}
neighbor fe80::2 {
address-family {
ipv6-unicast {
maximum-prefix "100"
peer-group "PEERSv6"
}
}
description "test peer 4 - remove to trigger problems"
remote-as "65501"
}
peer-group PEERSv6 {
address-family {
ipv6-unicast {
remove-private-as { }
soft-reconfiguration {
inbound { }
}
}
}
description "IPv6 Peer-Group"
}
}
}If you need any further information, please let me know.