This is issue tracks implementing a fix for the OCPBUGS-31515 behavior. While that Jira card handles the scenarios for MAPI, the solution for CAPI will require a different approach given the changes upstream has made.

The upstream issue is https://github.com/kubernetes-sigs/cluster-api-provider-azure/issues/5515.

https://learn.microsoft.com/en-us/azure/networking/troubleshoot-failed-state#provisioning-states is helpful in understanding how Azure's API reacts.

In summary:

• A customer has observed that NICs can be attached to a VM and functional, but the Azure API marks metadata for the NIC as "ProvisioningFailed", which then caused cascading problems because dependent resources will not proceed until the "ProvisioningFailed" status is cleared.
• MAPI had a fix with OCPBUGS-31515
• Cluster API Provider for Azure has changed their code around this handling significantly, opting to use the Azure Service Operator (ASO) for Azure resources. The NIC code resides in https://github.com/kubernetes-sigs/cluster-api-provider-azure/tree/main/azure/services/networkinterfaces