During a proposed refactor of a cloudceph host it was noticed that the cloudceph hosts manage a bunch of ip addresses directly in the puppet repo. It seems the in theory we should be able to also manage theses ip addresses in netbox and have them synced to puppet with the netbox->hiera pipeline.
Are there any issues with this proposal? What would be needed to start exploring this?
Longer timeframe but T325531: ceph: test and decide 1 network interface setup could potentially simplify the setup even more.
If I understand correctly this is not about driving the server's IP config from Netbox, but defining which IPs/interfaces to use by the service (Ceph) itself.
In that case that does make sens to me and indeed goes toward the current effort of removing hard-coded IPs from Puppet.
One question is how to distinguish which NIC/IP is of which "role" (public/cluster) but I guess matching on the IP prefix would be enough for now.
The last point I can think of is about workflows. For example when adding a new host to the cluster, all nodes would pick it up even if the service itself (Ceph) is not ready. Same when a node is removed. If that can be an issue maybe a Hiera flag "ready for traffic" could work out.