18:34 <knobby> `juju upgrade-charm etcd` `juju upgrade-charm ...` `juju config kubernetes-master dns-provider=core-dns`
18:35 <knobby> I then saw that I was on 1.14 for my workers and 1.13 for my master, I don't recall ever setting the channel on the worker, but it's possible it's set to ""
18:35 <knobby> I noticed that kube-dns was gone and no new dns was coming up
18:35 <knobby> I then thought it might be the cdk-addons snap, so I went and sudo snap refresh cdk-addons on the masters
18:36 <knobby> waited for hooks and no interesting things happened
18:36 <knobby> then I `juju config kubernetes-master channel="1.14/stable"` and then ran the upgrade action on each node
18:36 <knobby> at which point I got core-dns, but all my pods were exploding because they couldn't resolve names
18:37 <knobby> cat /etc/resolv.conf pointed at an old service ip
18:37 <knobby> kubectl delete po/<pod> results in a working pod
I guess we need to either:
1. Backport CoreDNS to older cdk-addons, or
2. Make it a no-op if you pick CoreDNS but you're on 1.13 or older snaps
I'm okay with either. The trouble with #2 is making it obvious to the user why nothing is happening.