Question on capabilities: is it safe to disable baremetal+MachineAPI capabilities in UPI install mode?

[titou10titou10]

For our OKD clusters, we do not enable capabilities that we don't need

Currently we use the following in"install-config.yaml"

capabilities:
  baselineCapabilitySet: None
  additionalEnabledCapabilities: 
  - baremetal
  - MachineAPI                 # Mandatory with "baremetal"
  - marketplace
  - OperatorLifecycleManager
  - OperatorLifecycleManagerV1 # 4.18
  #- openshift-samples
  - Console
  #- Insights                  # OKD is disconnected from RH...disable Insight missing connection Alert
  #- Storage
  - CSISnapshot
  - NodeTuning
  #- ImageRegistry
  #- Build
  #- DeploymentConfig
  #- CloudCredential
  #- CloudControllerManager
  - Ingress                    # 4.16

It works well.
In lab, disabling the"baremetal"and"MachineAPI"seem to work too as those are used for IPI install it seems. The cluster works well.

The question is, are those capabilities required for automatic upgrades?
What will happen if those are not installed and later we want to upgrade the cluster? I know those capabilities can be enabled later on

TheMachineAPIcapability enables theMachineSet,Machine andMachineHealthCheckCRDs which are only used for IPI install IMHO

Ref:

• Cluster capabilities
• Bare-metal capability
• Machine API capability
• Machine API Operator
• Overview of machine management

[titou10titou10]

A few more info.
I'm now running OKD 4.18, one 3+3 nodes, one SNO, both installed in UPI / ABI mode on Proxmox with thebaremetalandMachineAPIcapabilities disabled

capabilities:
  baselineCapabilitySet: None
  additionalEnabledCapabilities: 
  - marketplace
  - OperatorLifecycleManager
  - OperatorLifecycleManagerV1 # 4.18
  - Console
  - CSISnapshot
  - NodeTuning
  - Ingress                    # 4.16
  #- baremetal
  #- MachineAPI                # Mandatory with "baremetal"
  #- openshift-samples
  #- Insights                  # OKD is disconnected from RH...disable Insight missing connection Alert
  #- Storage
  #- ImageRegistry
  #- Build
  #- DeploymentConfig
  #- CloudCredential
  #- CloudControllerManager

I installed v4.18.0-okd-scos.0 and the clusters were running well
Then upgraded both clusters to v4.18.0-okd-scos.1 and everything is still working. No problem at all. Not running tenths of apps/pods not needed

So thebaremetalandMachineAPIdoes not seem to be needed in my case for upgrade, for OKD itself or upgrading SCOS.
I will just have to confirm maybe on next upgrade if SCOS version change if it is still the case has for v4.18.0-okd-scos.0 and v4.18.0-okd-scos.1 SCOS version for the nodes stayed the same: CentOS Stream CoreOS 418.9.202502120632-0 but I'm confident it will work