I'm going to use this story as a place to document the various bits of the process that Emilia will need to test if it works.

This all depends on the baremetal day one networking features, as described here: http://blog.nemebean.com/content/day-one-networking-openshift I'll try to link relevant sections of the docs as we go.

The high level overview of the steps is as follows:

1. Configure dev-scripts with BOND_PRIMARY_INTERFACE set to true.
2. Configure NETWORK_CONFIG_FOLDER with a config that enables DHCP on enp1s0 and enp2s0, but disables addressing on enp3s0 completely.
3. Also configure ASSETS_EXTRA_FOLDER with a similar configuration, except with br-ex added to enp2s0. enp3s0 must still be disabled.
4. Deploy the cluster.
5. Modify the NMState br-ex machine-configs to bond enp2s0 and enp3s0. Note that this step alone is not expected to affect the cluster at all since the configs are only applied at deploy-time.
6. On each node, delete the /etc/nmstate/openshift/applied file.
7. Also delete the /etc/nmstate/cluster.yml.applied file (I may have the filename slightly wrong, but there should be a file that corresponds to cluster.yml to indicate that it was already applied).
   1. I'm not sure whether NMState will recognize that the contents of cluster.yml changed, so for the initial test we should do this step. As a possible optimization later we could try without deleting this file and see if NMState will do the right thing on its own. If not, that might be a good RFE against NMState.
8. Create a dummy machine-config to force a reboot. Changes to files in /etc/nmstate/openshift don't trigger a reboot on their own so this has to be a separate step after all of the above actions have been taken.
9. After reboot, br-ex should be on bond0, which is made up of enp2s0 and enp3s0.

I'll leave some sample configs for these steps in the comments here to avoid bloating the description too much.