Website: Issue with /docs/getting-started-guides/kubeadm/

[bostone]

This is a...

• [ ] Feature Request
• [x] Bug Report

Problem:
When attempting to run "kubeadm init" step preflight fails with this message:

Some fatal errors occurred:
        /proc/sys/net/bridge/bridge-nf-call-iptables contents are not set to 1

Proposed Solution:
Add "sysctl" call reference to the manual (if that is indeed the way to fix this)

Page to Update:
https://kubernetes.io/docs/getting-started-guides/kubeadm/

1.6.0

[bostone]

Actually this is really bad. Even setting bridge-nf-call-iptables=1 doesn't solve the problem. The "kubeadm init" runs to this line and then hangs indefinitely

[apiclient] Created API client, waiting for the control plane to become ready

[bostone]

I think this is problem with 1.6.0 release. The previous version worked reliably but not anymore

[bostone]

Yes indeed, kubeadm is currently broken https://github.com/kubernetes/kubernetes/issues/43815

[zacharysarah]

@bostone Thanks for opening this issue. It looks like the underlying issue was resolved, so I'm going to close this issue. Please feel free to re-open it if you notice similar behavior. 👋

[anothersrewhocancode]

still getting an error with kubeadm 1.7.4

[naseemkullah]

for redhat family nodes: if you've disabled selinux and not yet rebooted the node (not mentioned in k8s docs to do so) then please do it.

after that, that error was gone, but as mentioned by someone above, now dealing with:

"[apiclient] Created API client, waiting for the control plane to become ready"

[zacharysarah]

@heckj 👋

[naseemkullah]

Hey in the end the error was that kubelet was not in an "ok" state due to a mismatch with docker in terms of cgroup driver.

By editing /etc/systemd/system/kubelet.service.d/10-kubeadm.conf with the cgroup driver that matches docker, running a systemctl daemon-reload, and finally restarting the kubelet service, it is in an "ok" state.

[heckj]

thanks @naseemkullah - was there an error in the getting started, or something we simply just didn't specify, that would have helped make sure the relevant versions were aligned?

[naseemkullah]

np @heckj , well the rebooting after selinux is disabled was not mentioned in the docs unless i missed it. As for the kubelet not being ok due to mismatch of docker and kubelet cgroup drivers, this may be caused due to the fact that i used latest versions of both components, even though it was recommended to use an older version of docker. but once i set kubelet's cgroup driver to be the same as docker's, kubeadm init worked like a charm. Hope that helps, if not please let me know.

[naseemkullah]

@heckj if this helps at all, here are the ansible tasks used to deploy latest kubeadm on centos7, (along with latest version of docker) successfully, please take note of the "reboot node" handler called when selinux is disabled.
https://github.com/naseemkullah/mp-roles/blob/master/install-kubeadm/tasks/main.yml

[heckj]

Thanks @naseemkullah

I'm traveling at the moment, but will keep this open and assigned to me to update the CentOS related installation docs to make a mention of the need for a reboot after disabling SELinux.

[heckj]

PR up, the documentation I think you were originally referencing is now deprecated, but I went ahead and made a note about the need to reboot for SElinux disabling on CentOS. I also found some broken links and resolved those in the same PR.