There's a race between the UDN/NAD creation (or processing that creation) in the nodes vs the workloads being created: if, for whatever reason, the UDN gets created first, only the pod default network will be plumbed into the workloads.

We need to work-around this until the SDN team handles it - by ensuring we have created the logical switch for the primary UDN in each node before attempting to create workloads attached to it.