ARM APIServer timeout is too short for upgrade.

[brendandburns]

An APIServer on ARM takes ~1.5 minutes to come up. However, the health-check kicks in earlier than that, and the kubelet kills the APIServer.

This prevents successful kubeadm based upgrades on ARM.

We should make this timeout be configurable.

It's even worse on self-hosted clusters, because you can not manually edit the timeout when the apiserver is down. I had to use systemctl stop kubelet to prevent the kubelet from killing the apiserver repeatedly.

[luxas]

In what way would you like this to be configurable? Via flags or the Config file?

[0xmichalis]

Dupe of #413?

[brendandburns]

Yeah, this is a dupe of #413.

I think a flag is sufficient.