no joy with the accept_ra fiddling, the image are all ubuntu or debian cloud-init ready and are working on all in one node but not dedicated compute node, it's as if the dhcpv6 packet (before address is allocated to interface) are not passing between the control node and the dedicated compute node.

Le 2022-09-20 10:53, Michael Hazan a écrit :

Thanks for the suggestion, i will look into the accept_ra setting. However, the very same image that boot on the « old » all in one node are getting ipv6 and ipv4 without issue. The network interfaces are configured similarly on all nodes.

 

Michael

 

 

De : Neil Jerram
Envoyé le :mardi, 20 septembre 2022 10:47
À : Michael Hazan
Cc : Lajos Katona; openstack-discuss@lists.openstack.org
Objet :Re: RE : [neutron][nova][ipv6]instance on compute node not getting dhcpv6

 

They may be outdated now, but I wonder if any of the steps described at https://projectcalico.docs.tigera.io/networking/openstack/ipv6 would help you?  In particular, given that you have no route, perhaps you need the accept_ra config.

 

Best wishes,

    Neil

 

 

On Tue, Sep 20, 2022 at 9:12 AM Michael Hazan <mhazan@mandint.org> wrote:

Hello,

The 3 nodes are running Ubuntu 20.04 and using libvirt. The instance are Ubuntu or debian based. The openstack virtual network is configured with both a v4 and a v6 dhcp pool. The instance should be getting both v4 and v6 (on the same interface) but the instance that are hosted on the compute node only get v4. If i run the dhclient -6 -r the instance actually get it's ipv6 but it's wrongly configured with a /128 (instead of 64) and no route.

 

Michael

 

De : Lajos Katona
Envoyé le :mardi, 20 septembre 2022 09:39
À : mhazan@mandint.org
Cc : openstack-discuss@lists.openstack.org
Objet :Re: [neutron][nova][ipv6]instance on compute node not getting dhcpv6

 

Hi,

Do you have some more details of your environment, which backend you use? OVS?

Your VMs have both IPv4 and v6 addresses and you mean that they got IPv4 as expected?

 

Lajos

 

<mhazan@mandint.org> ezt írta (időpont: 2022. szept. 19., H, 16:19):

Hello all,

i manage a small openstack cluster installed via Ubuntu package, it has
been upgraded since the day of Newton and currently running Victoria. it
started as a single node with the basic services installed, later a
dedicated storage node and compute node were added. after upgrading to
victoria, any instance that boot on the dedicated compute node start up
but fails to get it's global ipv6 via stateful dhcp6, only
linklocal(fe80) ip is shown by the instance. The ipv6 is created in
openstack, the dhcpv4 works and there are no issue if the instance boot
on the "old" all in one node. i can manually allocate (by editing the
instance network config file) the ipv6 config and it works. any pointer
as to what could cause this issue ?

BR, Michael