Manual reboot required

[bensonz]

Not quite sure what is going on here but every machine seems to be needing some reboot after a while. Here's what it's doing:

Each node today has encountered this issue, master, edge and worker. However all three of them seems to be fine after a manual reboot.

[bensonz]

This problem occurs because the system tries to pull flannel image and always timeout.
The flanneld services seem to be able to magically recover after some time, but our kubelet does not re-try after the failure.
Potential solution includes:

• add restart policy to kubelet
• let system know NOT to always pull new flannel