search

luxas / kubeadm-installer

A kubeadm installer container for CoreOS, Ubuntu, Debian, CentOS and Fedora
MIT License
21 stars 26 forks source link

Pre-existing ETCD cluster needed? #11

Open skloeckner opened 7 years ago

skloeckner commented 7 years ago

Do I have to create an etcd cluster for the bare metal machines I plan on using first for CoreOS? Or does Kubernetes(By that I mean kubeadm) configure that for me because all the minions end up pointing to the master, anyway?

Also, I cannot post issues on the new maintainer's repo because it's still a fork of this project. Not sure how to solve that on Github but if the new maintainer wants to solve issues, he needs his own repo.

@xakraz

xakraz commented 7 years ago

Hi @skloeckner ,

Currently kubeadm creates by default 1 etcd server pod as part of the Control Plane on the master node.

You can provide an etcd endpoint if you have an already running etcd-server cluster.

If you need HA, maybe you can do it in 2/3 steps:

  1. Use kubeadm to bootstrap k8s on 1 node.
  2. Provision other nodes and refer them as "masters"
  3. Update the pods, deployment, daemonssets .... so that the Control Plane resources are spread on the other master nodes

That is what I plan to try very shortly.