OpenShift Container Platform

The ClusterPodPlacementConfig object

After installing the Multiarch Tuning Operator, you must create a ClusterPodPlacementConfig object. The object instructs the Operator to deploy its operand, which enables architecture-aware workload scheduling across your cluster.

The ClusterPodPlacementConfig object supports two optional plugins:

The node affinity scoring plugin patches pods to set soft preferences, using weighted affinities, for the architectures specified by the user. Pods are more likely to be scheduled on nodes running architectures with higher weights.
The exec format error monitor plugin detects ENOEXEC errors, which occur when a pod attempts to execute a binary incompatible with the architecture of the node. When enabled, this plugin generates events in the affected event stream of the pod. The plugin triggers an ExecFormatErrorsDetected Prometheus alert if one or more ENOEXEC errors are detected within the last six hours. These errors can result from incorrect architecture node selectors, invalid image metadata that affects architecture-aware workload scheduling, an incorrect binary in an image, or an incompatible binary injected at runtime.

You can create only one instance of the ClusterPodPlacementConfig object.

Example ClusterPodPlacementConfig object configuration

apiVersion: multiarch.openshift.io/v1beta1
kind: ClusterPodPlacementConfig
metadata:
  name: cluster
spec:
  logVerbosity: Normal
  namespaceSelector:
    matchExpressions:
      - key: multiarch.openshift.io/exclude-pod-placement
        operator: DoesNotExist
  plugins:
    nodeAffinityScoring:
      enabled: true
      platforms:
        - architecture: amd64
          weight: 100
        - architecture: arm64
          weight: 50
    execFormatErrorMonitor:
      enabled: true
  fallbackArchitecture: amd64

where:

metadata.name: Specifies the name of the object. You must set this parameter to cluster.
spec.logVerbosity: Optional parameter. Specifies the log verbosity level. You can set the field value to Normal, Debug, Trace, or TraceAll. The value is set to Normal by default.
spec.namespaceSelector: Optional parameter. You can configure the namespaceSelector to select the namespaces in which the Multiarch Tuning Operator’s pod placement operand must process the nodeAffinity of the pods. All namespaces are considered by default.
spec.plugins.nodeAffinityScoring.enabled: Optional parameter. You can enable the node affinity scoring plugin to set architecture preferences for pod placement. When enabled, the scheduler first filters out nodes that do not meet the pod’s requirements. Then, it prioritizes the remaining nodes based on the architecture scores defined in the nodeAffinityScoring.platforms field. The default value is false.
spec.plugins.nodeAffinityScoring.platforms: Optional parameter. Defines a list of architectures and their corresponding scores. The scheduler prioritizes nodes for pod placement based on the architecture scores that you set and the scheduling requirements defined in the pod specification.
spec.plugins.nodeAffinityScoring.platforms.architecture: Specifies the architecture for the node affinity scoring plugin. Accepted values are arm64, amd64, ppc64le, or s390x.
spec.plugins.nodeAffinityScoring.platforms.weight: Specifies the weight for the architecture you specified in the spec.plugins.nodeAffinityScoring.platforms.architecture parameter. The value must be configured in the range of 1 (lowest priority) to 100 (highest priority). The scheduler uses this score to prioritize nodes for pod placement, favoring nodes with architectures that have higher scores.
spec.plugins.execFormatErrorMonitor.enabled: Optional parameter. Set this field to true to enable the execFormatErrorMonitor plugin. When enabled, the plugin detects ENOEXEC errors, caused when a pod executes a binary incompatible with the node’s architecture. The plugin generates events in the affected pods, and triggers the ExecFormatErrorsDetected Prometheus alert if one or more errors are found in the last six hours.
spec.fallbackArchitecture: Optional parameter. Specifies an architecture where pods are scheduled if the image inspector cannot determine the architecture of the image. Valid values are "", arm64, amd64, ppc64le, or s390x. The value is set to "" by default.

In this example, the operator field value is set to DoesNotExist. Therefore, if the key field value (multiarch.openshift.io/exclude-pod-placement) is set as a label in a namespace, the operand does not process the nodeAffinity of the pods in that namespace. Instead, the operand processes the nodeAffinity of the pods in namespaces that do not contain the label.

If you want the operand to process the nodeAffinity of the pods only in specific namespaces, you can configure the namespaceSelector as follows:

namespaceSelector:
  matchExpressions:
    - key: multiarch.openshift.io/include-pod-placement
      operator: Exists

In this example, the operator field value is set to Exists. Therefore, the operand processes the nodeAffinity of the pods only in namespaces that contain the multiarch.openshift.io/include-pod-placement label.

This Operator excludes pods in namespaces starting with kube-. The Operator also excludes pods that are expected to be scheduled on control plane nodes.