Rancher, Calico: v4 works while v6 does not

Moritz Wörmann

unread,

Mar 25, 2023, 7:11:36 AM3/25/23

to metallb-users

Hi all,

i have deployed metallb on a Rancher Bare Metal cluster with two nodes. As i require dual stack, my only option for CNI with rancher is calico.

My goal was that metallb assigns IPv4 out of a private range and v6 out of my own public prefix. The two nodes sit behind a VyOS Edgerouter which has BGP sessions with each node.

The assignment works, the services get addresses out of the correct pools.

But: While v4 works flawlessly, v6 does not. Not even from the nodes themselves:

Bildschirmfoto vom 2023-03-25 12-08-29.png

The Addresspool is "announced" both via L2 and BGP.

I have attached the result of 'calicoctl node diags' and all metallb logs, as well as the addresspool definition.

Has anyone else ever experienced something like this? I have tried to use different prefixes (even LL ones), but no luck.

Thanks in advance!

Kind Regars

mowoe

metallb-logs.txt

adress-pool.yaml

diags-20230325_103812.tar.gz

Carsten Laun-De Lellis

unread,

Oct 11, 2023, 7:56:25 AM10/11/23

to metallb-users

I had similiar problems, but got them fixed with providing clusteri-ip-cidr as well as service-ip-cidr during k3s setup.

Also in service you have to set the

type: LoadBalancer
ipFamilyPolicy: RequireDualStack
ipFamilies:
- IPv4
- IPv6

according to your needs.

Best regards,

Carsten

Carsten Laun-De Lellis

unread,

Oct 11, 2023, 7:57:59 AM10/11/23

to metallb-users

Sorry. Forgot to mention that I have returned back to flannel, because live with dual stack seems to be a lot easier.

Best regards,

Carsten

On Saturday, March 25, 2023 at 12:11:36 PM UTC+1 lukas.rump...@gmail.com wrote:

Reply all

Reply to author

Forward