I've successfully deployed a Kubernetes 1.3.5 cluster on 1 master + 6 nodes (all running CentOS) including the DNS and Kubernetes Dashboard addons. Everything seemed to be working OK at first. However, when I tried to run
kubectl proxy --address=<master-external-ip> --port=9090 --disable-filterand access http://<master-external-ip>:9090/ui I got the following output
Error: 'dial tcp 172.16.38.2:9090: i/o timeout'
Trying to reach: 'http://172.16.38.2:9090/'
However, if I start flanneld on the master everything works and I can actually reach the Dashboard. Now, I've used kube-up.sh to install the cluster and it didn't install any configs or systemd service for Flannel, which leaves me confused—should Flannel also run on the master?
Yes, it should, otherwise packets going through the API server proxy can not be routed to their final destination: the dashboard pod running on some other machine.