Comment # 1 on bug 1161289 from Olivier Nicolas

Not related to multi-master, I have a simple test cluster: 1 master and 1
worker.
After the autoupdate, the cluster was in bad shape. 

I have successfully followed your procedure:
 * rollback to 1.16
 * selectively update kubernetes-kubeadm package
 * follow the offical guide
 * then update all others packages.