Description of problem:

On a host with no default IP or route, MicroShift does not start after a reboot.

Version-Release number of selected component (if applicable):

4.14

How reproducible:

only tried one time

Steps to Reproduce:

1. Build a VM in the normal way.
2. Install MicroShift.
3. Stop MicroShift and clean up OVN data.
   1. systemctl stop microshift
   2. systemctl stop kubepods.slice
   3. /usr/bin/microshift-cleanup-data --ovn
4. Stop external network access by bringing the NIC down. Your device name may be different, if so replace the "enp1s0" part of this command with the device that has the default gateway.
   1. nmcli connection down enp1s0
   2. nmcli connection modify enp1s0 connection.autoconnect no
5. Add a fake IP on the loopback interface.
   1. nmcli con add type loopback con-name bogus-microshift ifname lo ip4 10.44.0.1/32
6. Add that same IP to the MicroShift config file as the node.nodeIP setting.
7. Add an entry for the node's hostname in /etc/hosts using the same IP.
8. Configure DNS to look at the local name server (note this is not the same IP, it's on a different subnet).
   1. echo “nameserver 10.44.1.1” >> /etc/resolv.conf
9. Restart microshift
   1. systemctl start microshift
10. Verify it works
    1. oc get pods -A
11. Reboot the host
12. Verify it works (watch for a while, because the initial pod information is cached from before the reboot and is not accurate)
    1. oc get pods -A

Actual results:

The openshift-dns/dns-default-* pod will enter a crash loop. The logs show:

plugin/forward: no nameservers found

Expected results:

Everything works.

Additional info: