Improve kubeadm reset for 1.13

[neolit123]

1.12 planning doc:
https://docs.google.com/document/d/1YP8sCtgo58yqQ9sR4nsn9iuVFmglCCBMivCTwjd1ERE/edit#

we need better context for the set of tasks for this one.

[rosti]

There are some kubeadm reset related issues:

• Stale iptables rules seen after "kubeadm reset" #689
• Pods are not getting cleaned in CRI-O after running kubeadm reset. #932

[timothysc]

@rosti The general theme is kubeadm reset is not consistent on the cleanup.

1 - issue WRT iptables rules imo is outside the scope of kubeadm and falls on the proxy. But that doesn't mean that the proxy shouldn't support a cleanup option that kubeadm can call.

the other CRI cleanup issues definitely is in our wheelhouse.

[timothysc]

/assign @rosti
/unassign @timothysc

[fabriziopandini]

Orthogonal to container cleanup we can improve kubeadm reset us by reading the crisocket from the node (instead of forcing the user to the user to re-pass the --criSocket flag)

[rosti]

As this is moved to 1.13 we can improve a lot. There is some Linux specific stuff, some hard coded paths, etc. which could not be addressed for 1.12 since they were not bugs (leftovers from reset).

[timothysc]

closing due to other issues for tracking and other PRs are already in for 1.13