This is a clone of issue OCPBUGS-62167. The following is the description of the original issue:
—
This is a clone of issue OCPBUGS-54682. The following is the description of the original issue:
—
Description of problem:

Under the following circumstances, restarting NetworkManager fully breaks br-ex:

• Node main interface is a bond
• Node uses a customized br-ex
• All the above is defined as a machineconfig (i.e. as per this https://docs.redhat.com/en/documentation/openshift_container_platform/4.18/html/installing_on_bare_metal/installer-provisioned-infrastructure#configuring-networking_ipi-install-installation-workflow)

If all the above is met, br-ex becomes unusable if NetworkManager is restarted (either on intention or because it crashed and auto-restarted).

Issue can only be solved by fully rebooting the node.

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

4.17.17

How reproducible:

Always

Steps to Reproduce:

1. Install nodes with customized br-ex in day one: https://docs.redhat.com/en/documentation/openshift_container_platform/4.18/html/installing_on_bare_metal/installer-provisioned-infrastructure#configuring-networking_ipi-install-installation-workflow
2. Restart NetworkManager (or wait for it to crash)

Actual results:

br-ex does not work, to the point that it cannot even get an IP via DHCP.

Expected results:

br-ex to work

Additional info:

Quickly checking, it seems that bond interface may be removed and re-added to br-ex in such a way that it gets a different OVS port number, but the flows are not updated accordingly. However, this may not be the only corruption possible.

More details in comments