[PVE-User] OSPF when migrating from v8 to v9

[Stefan Hanreich]

On 10/10/25 7:55 AM, Julien OHAYON wrote:
> Yes, indeed, during the upgrade nothing changes. Unfortunately, we will have to migrate over several days, and above all, there is still a huge unknown that we would have liked to be able to control: how the switch to the new configuration will happen.
> 
> 
> 
> What we would have liked (and I don’t think I’m the only one in this case) is to be able to apply the configuration on the new nodes running v9, without it affecting those still in v8. Because if I have to migrate all my nodes and then press the button afterwards thinking everything will work fine… No, that’s not ideal.
> 
> 
> 
> Unfortunately, with network interactions — and even more so interactions with routing protocols deeply embedded in the infrastructure — you want to have control over how the switchover will occur.
> 
> 
> 
> For now, I don’t see how to migrate to v9 without risking major downtime (a lab will not show the exact behavior we can expect in this type of SDN infrastructure).


Understandable, I took a closer look with a colleague today and we found
a way that should hopefully work for you.


What should work as a workaround is setting 'ospfd=yes' in
/etc/default/frr - which provides a way to override the /etc/frr/daemons
file. While that file (/etc/default/frr) is deprecated, it is still read
and respected by FRR and therefore provides a way of overriding the
value independent of our tooling. This prevents the OSPF daemon from
getting disabled on applying the SDN configuration, since our tooling
doesn't touch it.

After you are finished upgrading everything to 9, you can then test
migrating to the SDN fabrics by freeing up one node in your cluster
(migrate everything away, rename the frr.conf.local, ...) and define a
new fabric that only contains the node you want to use for testing.
Applying the SDN configuration then leaves the OSPF configuration for
all other nodes intact, while configuring OSPF using the Fabrics on the
one node that is part of the fabric. Alternatively, a virtualized
Proxmox VE instance could be used for testing the fabrics specifically.


If that test is successful, you can start migrating the other nodes over
one by one.

I tested this procedure on a cluster here locally and it worked
perfectly here. It would still be advisable to test this procedure on a
test cluster (can be virtualized) before you proceed on your production
cluster.


Kind Regards
Stefan