At present, we expect CAPI created Machines to be reflected into MAPI if and only if the CAPI created Machine has an owner, and that owner is also reflected in to MAPI.

For reading clients, this could cause issues as they will not be able to determine the whole set of Machines in the cluster.

We need to explore whether all Machines should be reflected, and in particular, answer the following questions:

• Do we expect users to be operating on the Machine level?
• What extensions are we aware of, that require Machine level knowledge?
• Do MHC/Autoscaler need a complete picture?
• What failure scenarios might happen if an extension built on top of Machines does not have the complete picture?
• How will we handle scenarios where users are leveraging CAPI only features (eg MachineDeployments)
• Do we need to continue to backport CAPI features to the MAPI API and conversion layer?

CC deads@redhat.com who raised this topic for discussion at the F2F in July '24