This is a cache of https://docs.openshift.com/container-platform/4.17/virt/virtual_machines/advanced_vm_management/virt-understanding-aaq-operator.html. It is a snapshot of the page at 2024-11-25T08:43:01.353+0000.
Configuring the Application-Aware Quota Operator - Virtual machines | Virtualization | OpenShift Container Platform 4.17
    OpenShift docs are moving and will soon only be available at docs.redhat.com, the home of all Red Hat product documentation. Explore the new docs experience today.
  1. Documentation
    2. history bug_report picture_as_pdf
×

Configuring the Application-Aware Quota (AAQ) Operator

You can use the Application-Aware Quota (AAQ) Operator to customize and manage resource quotas for individual components in an OpenShift Container Platform cluster.

About the AAQ Operator

The Application-Aware Quota (AAQ) Operator provides more flexible and extensible quota management compared to the native ResourceQuota object in the OpenShift Container Platform platform.

In a multi-tenant cluster environment, where multiple workloads operate on shared infrastructure and resources, using the Kubernetes native ResourceQuota object to limit aggregate CPU and memory consumption presents infrastructure overhead and live migration challenges for OpenShift Virtualization workloads.

OpenShift Virtualization requires significant compute resource allocation to handle virtual machine (VM) live migrations and manage VM infrastructure overhead. When upgrading OpenShift Virtualization, you must migrate VMs to upgrade the virt-launcher pod. However, migrating a VM in the presence of a resource quota can cause the migration, and subsequently the upgrade, to fail.

With AAQ, you can allocate resources for VMs without interfering with cluster-level activities such as upgrades and node maintenance. The AAQ Operator also supports non-compute resources which eliminates the need to manage both the native resource quota and AAQ API objects separately.

AAQ Operator controller and custom resources

The AAQ Operator introduces two new API objects defined as custom resource definitions (CRDs) for managing alternative quota implementations across multiple namespaces:

  • ApplicationAwareResourceQuota: Sets aggregate quota restrictions enforced per namespace. The ApplicationAwareResourceQuota API is compatible with the native ResourceQuota object and shares the same specification and status definitions.

    Example manifest
    apiVersion: aaq.kubevirt.io/v1alpha1
kind: ApplicationAwareResourceQuota
metadata:
  name: example-resource-quota
spec:
  hard:
    requests.memory: 1Gi
    limits.memory: 1Gi
    requests.cpu/vmi: "1" (1)
    requests.memory/vmi: 1Gi (2)
# ...
    1 The maximum amount of CPU that is allowed for VM workloads in the default namespace.
    2 The maximum amount of RAM that is allowed for VM workloads in the default namespace.

  • ApplicationAwareClusterResourceQuota: Mirrors the ApplicationAwareResourceQuota object at a cluster scope. It is compatible with the native ClusterResourceQuota API object and shares the same specification and status definitions. When creating an AAQ cluster quota, you can select multiple namespaces based on annotation selection, label selection, or both by editing the spec.selector.labels or spec.selector.annotations fields.

    Example manifest
    apiVersion: aaq.kubevirt.io/v1alpha1
kind: ApplicationAwareClusterResourceQuota (1)
metadata:
  name: example-resource-quota
spec:
  quota:
    hard:
    requests.memory: 1Gi
    limits.memory: 1Gi
    requests.cpu/vmi: "1"
    requests.memory/vmi: 1Gi
  selector:
    annotations: null
    labels:
      matchLabels:
        kubernetes.io/metadata.name: default
# ...
    1 You can only create an ApplicationAwareClusterResourceQuota object if the spec.allowApplicationAwareClusterResourceQuota field in the HyperConverged custom resource (CR) is set to true.

    If both spec.selector.labels and spec.selector.annotations fields are set, only namespaces that match both are selected.

The AAQ controller uses a scheduling gate mechanism to evaluate whether there is enough of a resource available to run a workload. If so, the scheduling gate is removed from the pod and it is considered ready for scheduling. The quota usage status is updated to indicate how much of the quota is used.

If the CPU and memory requests and limits for the workload exceed the enforced quota usage limit, the pod remains in SchedulingGated status until there is enough quota available. The AAQ controller creates an event of type Warning with details on why the quota was exceeded. You can view the event details by using the oc get events command.

Pods that have the spec.nodeName field set to a specific node cannot use namespaces that match the spec.namespaceSelector labels defined in the HyperConverged CR.

Enabling the AAQ Operator

To deploy the AAQ Operator, set the enableApplicationAwareQuota feature gate to true in the HyperConverged custom resource (CR).

Prerequisites

  • You have access to the cluster as a user with cluster-admin privileges.

  • You have installed the OpenShift CLI (oc).

Procedure

  • Set the enableApplicationAwareQuota feature gate to true in the HyperConverged CR by running the following command:

    $ oc patch hco kubevirt-hyperconverged -n openshift-cnv \
 --type json -p '[{"op": "add", "path": "/spec/featureGates/enableApplicationAwareQuota", "value": true}]'

Configuring the AAQ Operator by using the CLI

You can configure the AAQ Operator by specifying the fields of the spec.applicationAwareConfig object in the HyperConverged custom resource (CR).

Prerequisites

  • You have access to the cluster as a user with cluster-admin privileges.

  • You have installed the OpenShift CLI (oc).

Procedure

  • Update the HyperConverged CR by running the following command:

    $ oc patch hco kubevirt-hyperconverged -n openshift-cnv --type merge -p '{
  "spec": {
    "applicationAwareConfig": {
      "vmiCalcConfigName": "DedicatedVirtualResources",
      "namespaceSelector": {
        "matchLabels": {
          "app": "my-app"
        }
      },
      "allowApplicationAwareClusterResourceQuota": true
    }
  }
}'

    where:

    vmiCalcConfigName

    Specifies how resource counting is managed for pods that run virtual machine (VM) workloads. Possible values are:

    • VmiPodUsage: Counts compute resources for pods associated with VMs in the same way as native resource quotas and excludes migration-related resources.

    • VirtualResources: Counts compute resources based on the VM specifications, using the VM RAM size for memory and virtual CPUs for processing.

    • DedicatedVirtualResources (default): Similar to VirtualResources, but separates resource tracking for pods associated with VMs by adding a /vmi suffix to CPU and memory resource names. For example, requests.cpu/vmi and requests.memory/vmi.

    namespaceSelector

    Determines the namespaces for which an AAQ scheduling gate is added to pods when they are created. If a namespace selector is not defined, the AAQ Operator targets namespaces with the application-aware-quota/enable-gating label as default.

    allowApplicationAwareClusterResourceQuota

    If set to true, you can create and manage the ApplicationAwareClusterResourceQuota object. Setting this attribute to true can increase scheduling time.

Additional resources