nodeip-configuration.service is not configured to run after nmstate service. So, is to be configured via nmstate service (e.g. as per https://docs.redhat.com/en/documentation/openshift_container_platform/4.18/html/installing_on_bare_metal/installer-provisioned-infrastructure#ipi-install-establishing-communication-between-subnets_ipi-install-installation-workflow).

This can be problematic in bond scenarios, where the following can happen:
- Node gets some networking configuration via networkmanager before nmstate can setup the connections properly. Because of that, some interface temporarily has a wrong IP, namely IP1.
- nodeip-configuration service runs and it wrongly deems IP1 to be the right node IP
- Then nmstate runs and setups the right networking. This means that, for example, we have a bond whose IP is IP2. IP2 is the right IP for the node, not IP1.
- However, as nodeip-configuration service ran before nmstate and stored the wrong IP1 in /run/nodeip-configuration/primary-ip , wait-for-primary-ip service blocks forever trying to use the wrong IP1 instead of the right IP2.

Possible workarounds:
- Reboot the node
- Write the right IP in /run/nodeip-configuration/primary-ip and manually restart wait-for-primary-ip service

Proposed solution: Make nodeip-configuration.service to run after nmstate service to ensure that nmstate config is applied.

4.18.7

Often (race condition) as long as steps to reproduce are met.

1. Start with a cluster that has a bond interface defined via host-level nmstate in machine-config, like as per https://docs.redhat.com/en/documentation/openshift_container_platform/4.18/html/installing_on_bare_metal/installer-provisioned-infrastructure#ipi-install-establishing-communication-between-subnets_ipi-install-installation-workflow . The environment uses DHCP and the DHCP server might assign IPs to the individual interfaces that are part of the bond before the bond is configured via nmstate.
2. Add a new node 
3. 

Quite often, nodes end up stuck forever in wait-for-primary-ip service because nodeip-configuration ran too early, so it wrote one of the wrong IPs that the node interfaces temporarily had.

nodeip-configuration to run after nmstate service, so that we are 100% sure that it runs with the right definitive network configuration.

I found this while doing an internal reproducer, but I wouldn't be surprised if some production user eventually finds this too.