Closed jakolehm closed 6 years ago
It seems that this does not happen every time (tested upgrade with another cluster).
DNS is still broken even if service endpoints are there.
Upgrade seems to work ok with cri-o. Maybe this is docker related issue?
Root cause seems to be missing cni config for kubelet (on upgrade).
Basically it cannot find matching pods -> endpoints are empty.
Saw this in controller-manager logs:
I managed to fix it by editing service and saving it (
kubectl edit svc ...
).