Exclude specific nodes from ClusterAutoscaler resource limits via annotation
Issue
In the current implementation of the Cluster Autoscaler, the cores.max and memory.max limits are global, cluster-wide settings. These limits include the resources consumed by the control plane nodes.
This creates challenges for customers (especially on managed services like ARO), as:
- The SRE team may resize control plane instances for maintenance or performance reasons.
- Any increase in control plane capacity consumes a portion of the customer's defined
cores.maxormemory.max. - This can inadvertently block the autoscaler from scaling up worker nodes even if the workload requires more resources, because the global "ceiling" has been reached.
Environment
- Red Hat OpenShift Container Platform (OCP)
- Azure Red Hat OpenShift (ARO)
- Cluster Autoscaler
Subscriber exclusive content
A Red Hat subscription provides unlimited access to our knowledgebase, tools, and much more.