Clean up scripting, fix awk, add ip routes for IBX datacenters

[cprivitere]

I cleaned up the yaml for the pre and post kubeadm commands.
Removed the if block logic around when kube-vip is enabled since it works fine when installed before kubeadm is run even on a join.
Removed some packages that don't need to be explicitly installed via apt (they'll get pulled in as dependencies as necessaryt).
Updated the awk logic for the kubernetes version detection to use apt-cache madison instead of apt-cache policy due to the latter having an issue returning a result for the current version of kubernetes.
Changed the kube-vip arguments to use the metadata server for all relevant info about bgp configuration.
Added a section that adds the ip routes necessary for configuring bgp. The newer datacenters all require these routes for kube-vip to be able to work.

[davidspek]

Thanks for the nice cleanup. However, I am seeing that the second control plane node isn't able to join the cluster. When running the kubeadm join command manually on the node it fails with [ERROR DirAvailable--etc-kubernetes-manifests]: /etc/kubernetes/manifests is not empty, which is what prompted me to generate the kube-vip manifest after the join command.
Do you have any idea why you aren't seeing this issue?

Also, it seems like the provisioning is slightly slower, but this might be due to the added apt packages.

[cprivitere]

I'm going to rerun my test. I think I know what flag is missing that's causing the second control plane to not come up, but I'm curious now why I didn't see it as well.

One thing I'm confused about, I'm actually specifying fewer apt packages than before. So if this is taking longer, it's for some other reason.

[cprivitere]

Yeah I just tested it again and all the control plane nodes came up fine. What facility are you testing in? Maybe that's our difference here.

[davidspek]

I'm running in AMS1, but I doubt that would make a difference since I'm pretty sure it is related to the Kubeadm process since only /etc/kubernetes/manifests/vip.yaml is created. So the manifests created by Kubeadm to launch the Kubernetes components are never even created.

[davidspek]

This is related to the conversation over at detiber#65 (comment).

[cprivitere]

kubernetes/kubernetes#104942 is the cause of our disparity. I was installing 1.23.4 which has a fix to skip checking if that folder exists. Fortunately there's a flag we can pass to get the same result in the meantime while we wait for prior versions to fall out of support. I've pushed a commit with that change now.

[davidspek]

@cprivite Thanks for pushing that commit. I can confirm it works as expected now.