It is important to note that when the OpenShift scheduler is placing pods, it first reviews the Pod CPU request and schedules it if there is a node that meets the requirements. It will then impose the CPU "Limits" to ensure the Pod doesn’t consume more than the intended allocation. The limit can never be lower than the request.
- NUMA Configuration
-
OpenShift provides a topology manager which leverages the CPU manager and Device manager to help associate processes to CPUs. Topology manager handles NUMA affinity. This feature is available as of OpenShift 4.6. For some examples on how to leverage the topology manager and creating workloads that work in real time, see Scheduling NUMA-aware workloads and Low latency tuning.