Background

By the time this feature is written. power monitoring for Red Hat OpenShift documents that: 

 "The estimator feature is experimental, not supported, and should not be relied upon."

This features aims to remove such limitation, allowing our users to make use of the estimator power source under certain conditions.

Requirement

Power monitoring shall provide support for metrics coming from the "estimator" source.
As models are applied based on the underlying platform, upon the completion of this feature, power monitoring for Red Hat OpenShift shall:

• Provide a list of platforms and conditions in which the estimators are supported (the amount of platforms in the list is left to engineering to decide/validate).
• Identify the underlying CPU/HW
  • This has some challenges. CPU id might not be an option
• If the underlying CPU/HW is supported, apply the corresponding estimator 
• If the estimator is not supported, don't provide any metric and raise awareness to the user (behavior to be defined in POWERMON-211 )
• Provide errors and accuracy of the results
• Document the limitations to users (check the most important limitations in this blog )

In general, the following shall be taken into account:

• Are those models specific to specific hardware? Which one?
• Do these models apply to Virtual Machines?
• Are these models supported in the public cloud? Unser which presumptions?

To clarify:

• How should power monitoring provide the models?
• Any downstream work needed? Both in terms of testing but also building artifacts, CRs, images...?
• Versioning needed in the models
• Backwards compatibility