Update kube-proxy daemonset to set overrideHostname

[detiber]

Fixes: #173

Signed-off-by: Jason DeTiberus detiber@gmail.com

[timothysc]

So if I get this change, there is a local setting override in a global config map for the kube-proxy that needs to be set so you update that setting with an init container before the proxy is run?

If so, that seems more like a general problem with the defaults for the proxy...

@detiber

[seh]

This is another manifestation of assuming that the hostname of the machine matches the node name, which doesn't work out in AWS, where the node name has to match the host's DNS name, not its (short, unqualified) hostname.

[detiber]

So if I get this change, there is a local setting override in a global config map for the kube-proxy that needs to be set so you update that setting with an init container before the proxy is run?

Correct, the other option would be to update the command arguments in the daemonset using the downward api, but #sig-network has marked that that option as deprecated. Considering the time constraints, I will file an upstream issue against kubeadm and add a comment linking to it.

If so, that seems more like a general problem with the defaults for the proxy...

I agree and it has been brought up in a few different issues in the past, I've added a comment here: kubernetes/kubernetes#57518

This really stems from the larger issue of multiple components assuming that node name == hostname.

[detiber]

@timothysc updated to include a comment and link to the kubeadm issue.

[timothysc]

/lgtm

I'm ok with as a temporary workaround, but let's work upstream to fix this.