Kubeadm: fresh kubeadm 1.5 install fails due to existing /var/lib/kubelet

[ozdanborne]

If you follow the kubeadm 1.5 install instructions to the T your master won't succesfully init because /var/lib/kubelet already exists. seems /var/lib/kubelet is created when installing kubelet in this line

sudo apt-get --y install kubectl=1.5.3-00 kubelet=1.5.3-00 kubernetes-cni=0.3.0.1-07a8a2-00

So users will need to kubeadm reset before their first kubeadm init.

[kartoch]

Problem still present in 1.6.3.

[luxas]

@dgoodwin Did you ever find out why this race condition exists?

@timothysc Have you seen/looked into this?

[dgoodwin]

I did not but there is some analysis in https://github.com/kubernetes/release/issues/171#issuecomment-270334974 and an open PR in https://github.com/kubernetes/release/pull/256

[ianvernon]

I'm hitting this in the 1.6.4 version of kubeadm as well.

[madorn]

same here.

[luxas]

Can any of you reproduce this with kubeadm v1.7?

[jstomphorst]

I repoduced it with a centos 7 and kubeadm v1.8 but kubeadm reset helps...

[jamiehannaford]

https://github.com/kubernetes/kubernetes/pull/53317 should also help with this (since the kubelet was saving its PKI assets in /var/run/kubelet). Hopefully when we have that in kubeadm, we can close this issue after a suitable waiting period (unless anybody runs into related issues).

[seslav]

I'm having the same issue and cannot start kubeadm

kubectl-1.8.0-0.x86_64
kubelet-1.8.0-0.x86_64
kubeadm-1.8.0-1.x86_64
kubernetes-cni-0.5.1-0.x86_64

I'm on CentOS7
Any workarounds?

[seslav]

I found a solution for me. The problem was the swap
So I executed swapoff -a then kubeadm reset and finally kubeadm init
Hope it helps

[luxas]

v1.5 isn't supported anymore. v1.8.0 had a problem with similar symptoms, but it was fixed in v1.8.1