Over the past few months we've had requests from both the VSphere and OpenStack teams to support a deployment model that I am describing as hybrid IPI/UPI. It involves nodes being deployed via the IPI process, but without any of the normal on-prem networking infrastructure (DNS and loadbalancer, primarily). Those components will be provided by the customer, as they would in a UPI deployment. This enables more flexible deployment architectures than traditional IPI, at the cost of more complexity in the deployment process.

While this doesn't strictly concern on-prem networking since our components are the ones being removed in such an architecture, it does raise some questions of how we support such a deployment and since it muddies the waters of the IPI and UPI terminology, do we need a new way to describe this to avoid confusion when discussing it? I feel like calling it either IPI or UPI is incorrect.