@cheburakshu you can just clean up the docker instances eg docker rm $(docker ps -a -q)

But ideally before you reset, you would delete all deployments. But even then, probably still some traces will linger around in mount points on the host (eg /etc/, /var/lib/ depending on the specific deployment).

A bug related to this was opened kubernetes/kubernetes#44911

@coeki I raised a bug on weavenet for which he opened it on kubernetes. Please see - weaveworks/weave#2911 referenced in #44911.

@cheburakshu My bad, didn't read the issue 44911 well.

But after reading a bit through that issue, and related issues, the question arises if this is a kubeadm reset related issue. Cause it seems that it's more a general problem of kubernetes, in handling a graceful shutdown/delete/error of pods and/or nodes/deployments, and what they are doing to a node/host.

As @bboreham pointed out, there's no good way to clean up stuff done to the node/host after deletion of a deployment/daemonset, as pointed out in kubernetes/kubernetes#35183.

I'm not sure if kubeadm reset would/should be responsible for all components added to a kubernetes cluster created with kubeadm functionality, as it is meant IIUC, to clean up kubeadm core components (oops on core ;) ), and as mentioned clearing out the link to what plugin was used, but not cleaning up what that plugin did.

Maybe kubeadm reset in the docs should explain this better. Or functionality added to kubeadm to delete any deployments first, before deleting the cluster. But deployments added after setting up the cluster, would in it's turn depend on the behavior of kubernetes/plugins itself to handle that better, so kubeadm reset can use that.

It seems like a loop, but depends on what the responsibilities of what kubeadm reset should be.

cc @jbeda @luxas

/cc @klizhentas

xref: https://github.com/kubernetes/community/pull/483/files seems like the proper way the system should behave and it's not a kubeadm issue, but we should definitely document the cleanup process.

FWIW, I believe kubernetes/community#541 is a better match for 98% of the problem wrt Weave Net: we can clean up in preStop if we know it's forever.

The other 2% of the problem is things like "the pod already died when you're trying to delete it", and I'm not sure there is any complete solution in that scenario.

So I'm going to close this one as it's not really kubeadm so much as the kubelet and weave.