The infra components of RHOSAK and all components of RHOC get installed on default worker nodes along with any customer workloads.

Since we can not control or measure the load set on these nodes by the customer workloads, it would be hard to meet any SLAs that we are set, or not easily possible to prove when we miss a SLO is due to these additional processes.

Typically CPU and memory resources in Kube can be controlled through limits/requests, but IO can not be controlled and varies by the node type and also cluster provider type. (in our testing we have seen GCP give half the throughput of AWS on a comparative node)

 The ask for BU is what are the SLAs we are offering to the customers?

The follow on is for engineering as to how to meet those SLAs. One proposal that was mentioned before was to use another dedicated pool(s) for the infra or RH deployed components.