After updating the cluster to 4.12.42 (from 4.12.15), the customer noticed some issues for the scheduled PODs to start on the node.
The initial thought was a multus issue, and then we realised that the script /usr/local/bin/configure-ovs.sh was modified and reverting the modification fixed the issue.
Should be reproducible by setting inactive nmcli connections with the same names as the active once
Not tested, but this should be something like
1. create inactive same nmcli connections
2. run the script
Script should manage the connection using the UUID instead of using the Name.
Or maybe it's an underline issue how nmcli is managing the relationship between objects.
The issue may be related to the way that nmcli is working, as it should use the UUID to match the `vlan.parent` as it does with the `connection.master`