What are you doing with your initial nodes?

[jhoelzel]

Hey guys,

to begin with i am in love with rke2 and its my favorite distro so far and i really appreciate all the work that is going into it =)

How are you treating the initial server of a HA environment? Since it has to be started without the server url to initialize the cluster, i have been having some thoughts about what to do with it.

I use terraform for provisioning since it easily integrates with everything else i have and henceforth i always end up with a different machine for the intial server than for all other master. Now i can call that machine a master all day long too but the fact remains that it will always have a different config than all other master because of the server URL.

I was wondering what other people are doing with this at the point where the cluster is running.
do i:

• leave everything as is forever and mark the node in terraform as "unchangeable"?
• update the config on the initial server once the cluster is running to then include the server url too?
• provision 4 maters in total and just delete the initial one?

I am using rolling release operating systems with rke2 so im not worried about updates, but more about what is to be done once i have a hardware failure on the initial node.

Currently, when i delete it, it will provision a new cluster for itself when i "terraform over it".

Thanks in advance for your ideas

[brandond]

do i:

1. leave everything as is forever and mark the node in terraform as "unchangeable"?
2. update the config on the initial server once the cluster is running to then include the server url too?
3. provision 4 maters in total and just delete the initial one?

I would probably do 1 or 2, depending on whatever's easiest for you.

[ghost]

I use HAProxy to LB kube-apiserver and rke2-api in tcp mode. I set up HAP first, then 1st server node, then others. Finally put server statement on server #1. So all servers are 'equal'. If you are fully automated, option number 3 would not require to set up this special process for node number 1 and would be simpler.