What are the risk of using kubeadm in production?

10/26/2016

Currently playing with kubernetes, I need to deploy a cluster by myself on my own hardware or cloud provider (I would love to use GCE but it is not possible in a near future).

I saw kubeadm allow a quick and easy cluster bootstrapping except it does only provide one kubernetes master.

As I'm looking for a solution I can use in production:

  • What would happen if the master reboot for an unknown reason ?
  • Using kubeadm, is it possible to enable cloud provider features such as LB or persistent volumes plugins ?
-- Alex Grs
kubernetes

1 Answer

10/29/2016

I am also trying to bring up some experimental setups using ubu 16.04 and kubeadm, with the following experience:

the master reboot situation is the most critical point with kubeadm, as the cluster is not booting up properly after a reboot. Another SO user reported the issue here, where I shared my scripts to relaunch the cluster if the weave net is stuck in CrashLoopBackOff.

I also let it alone for a while, and after a lot of restarts it started to work...but this means a quite long down time to your cluster.

When wiping the cluster, you loose all your configuration. The only way to prevent this is to restore etcd somehow...however I didn't find any acceptable solution, yet.

About cloud providers, there are experimental ones since 1.4.3 AFAIK. I didn't try it, but it is a way to go. However, if you are planning more generic, bare metal for example, you should take a look at part 3 of this article series about deploying kube on scaleway with kubeadm, which covers installing glusterfs as PV. In Part 2 he also describes using traefik as ingress controller.

As you see, operating a production cluster with kubeadm is not a easy task, but as they say, it's still alpha. I am watching this project excited, hoping it gets production ready, soon.

-- David Steiman
Source: StackOverflow