fix(systemd/network): exclude node* and kube* from dhcp by default

[maxime1907]

[systemd/network: exclude node* and kube* from dhcp by default]

Currently, when using kubeadm join, it says that the command works but sometimes it fails because the container nodelocaldns creates a dummy interface called nodelocaldns by default and systemd-networkd manages it with the default file zz-default.network that adds DHCP and IPV6RA support.

So when nodelocaldns tries to do an equivalent of ip addr add, the command works but it will silently fail because the interface is managed by two different programs (systemd-networkd and nodelocaldns)

How to use

Check that the interfaces created by kube-proxy and nodelocaldns are not managed by systemd-networkd with the command networkctl.

Check that kubeadm join properly works so the node can join the cluster by setting its default nodelocaldns interface to the IPs defined with -localip in the node-cache container.

Testing done

Checked that everything in How to use works

[maxime1907]

cc @pothos @krnowak

[pothos]

Thanks

[pothos]

I actually had excluded dummy devices from being managed by default. Have to check why this is not working: https://github.com/flatcar/init/blob/flatcar-master/systemd/network/zz-default.network#L8

[pothos]

Ah, Type vs Driver!

[pothos]

Fix prepared in #90

We can still keep your changes, we also have this for other similar cases and these are probably the ones we would backport. I still would love to see Kubernetes to provide the unit itself in /run/systemd/network/ instead of forcing distros and users to take care of this.