1. Documentation
×

Run automated postinstallation configuration

The virt-platform-autopilot controller replaces manual postinstallation configuration by automatically applying opinionated defaults to the virtualization platform and installed optional components. You can customize or opt out of any managed configuration.

How automated postinstallation configuration works

After you install OKD Virtualization, the virt-platform-autopilot controller continuously reconciles managed resources to keep the expected state and correct configuration drift. It also configures optional components such as the Kube Descheduler Operator and the Cluster Observability Operator if they are installed.

The virt-platform-autopilot controller manages the following configurations:

Load-aware descheduler

Configures the Kube Descheduler Operator to balance virtual machine (VM) workloads across nodes based on CPU and memory use. Requires the Kube Descheduler Operator to be installed.

Observability enhancements

Enhances observability with additional Prometheus alerting rules and advanced Perses dashboards in the OKD web console.

Swap enablement

Enables OKD worker nodes to safely use swap for virtualization workloads. Swap requires pre-provisioned, dedicated storage to be available.

Soft dependency model

The virt-platform-autopilot controller does not install optional components. It configures components that are already installed on the cluster. If an optional component is not installed, the virt-platform-autopilot controller skips that configuration and continues without error.

When you install an optional component later, the virt-platform-autopilot controller detects it on the next reconciliation cycle and applies the relevant configuration automatically.

Reconciliation and anti-thrashing behavior

The virt-platform-autopilot controller continuously reconciles managed resources to keep the expected state. If an external process changes a managed resource, the virt-platform-autopilot controller detects the drift and restores the expected state.

To prevent thrashing, the virt-platform-autopilot controller monitors the rate of changes to managed resources. If the virt-platform-autopilot controller detects rapid or repeated changes, it pauses reconciliation for the affected resources. This prevents a loop where the virt-platform-autopilot controller and another controller repeatedly overwrite each other’s changes. You can monitor paused resources by using the kubevirt_autopilot_paused_resources metric.

Customize or opt-out of managed configurations

The virt-platform-autopilot controller applies opinionated defaults, but you can override managed configurations to suit your environment.

z

Stop reconciliation of a resource

Mark a resource as unmanaged so that the controller does not reconcile it after it is created. You take full ownership of that resource’s configuration.

You can combine these approaches across different resources in the same cluster.

Exclude a resource from being managed by the virt-platform-autopilot controller

Disabled resources are not created or reconciled automatically by the virt-platform-autopilot controller. Prevent the controller from creating or reconciling a resource by disabling it in the HyperConverged custom resource (CR).

Prerequisites

  • You have installed the OpenShift CLI (oc).

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

  • The virt-platform-autopilot controller is enabled on the cluster.

Procedure

  1. Edit the HyperConverged CR and add resources to the platform.kubevirt.io/disabled-resources annotation:

    Example HyperConverged CR:

    apiVersion: hco.kubevirt.io/v1beta1
kind: HyperConverged
metadata:
  annotations:
    platform.kubevirt.io/disabled-resources: |
      - kind: KubeDescheduler
        name: cluster
      - kind: MachineConfig
        name: 90-worker-swap-online

  2. Apply your changes to the HyperConverged CR by running the following command:

    $ oc apply -f <filename>.yaml

Customize specific values in a managed resource

Customize specific values in a managed resource by applying a JSON patch on top of the reconciled configuration to retain your custom settings.

Deleting an unmanaged resource causes any annotations on that resource to be lost. The virt-platform-autopilot controller then creates a new, opinionated version of the resource.

To prevent the virt-platform-autopilot controller from recreating deleted resources, you must disable the resource. You can disable resources by adding them to the platform.kubevirt.io/disabled-resources annotation in the HyperConverged custom resource (CR).

For more information, see "Exclude a resource from being managed by the virt-platform-autopilot controller".
Prerequisites

  • You have installed the OpenShift CLI (oc).

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

  • The virt-platform-autopilot controller is enabled on the cluster.

  • The platform.kubevirt.io/mode=unmanaged and platform.kubevirt.io/ignore-fields annotations must not be present on the resource.

Procedure

  1. Add the platform.kubevirt.io/patch annotation to a resource with a valid RFC 6902 JSON Patch operations array.

    For example, to prevent the virt-platform-autopilot controller from overriding a custom descheduler profile configuration, run the following command:

    $ oc annotate kubedescheduler cluster \
  'platform.kubevirt.io/patch=[{"op":"replace","path":"/spec/profileCustomizations","value":{"devKnobs":{}}}]' \
  -n openshift-kube-descheduler-operator

    The virt-platform-autopilot controller continues to manage all other fields on the resource but skips the specified paths.

    JSON patches are blocked on sensitive resources, including MachineConfig and KubeletConfig resources.

  2. Optional: To restore virt-platform-autopilot controller management for all fields, remove the platform.kubevirt.io/patch annotation.

    For example, to restore virt-platform-autopilot controller management for all fields of the descheduler configuration, run the following command:

    $ oc annotate kubedescheduler cluster \
  platform.kubevirt.io/patch- \
  -n openshift-kube-descheduler-operator

Exclude specific fields from reconciliation in a managed resource

Exclude specific fields from being managed by the virt-platform-autopilot controller so that it ignores them during reconciliation.

Deleting an unmanaged resource causes any annotations on that resource to be lost. The virt-platform-autopilot controller then creates a new, opinionated version of the resource.

To prevent the virt-platform-autopilot controller from recreating deleted resources, you must disable the resource. You can disable resources by adding them to the platform.kubevirt.io/disabled-resources annotation in the HyperConverged custom resource (CR).

For more information, see "Exclude a resource from being managed by the virt-platform-autopilot controller".
Prerequisites

  • You have installed the OpenShift CLI (oc).

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

  • The virt-platform-autopilot controller is enabled on the cluster.

  • The platform.kubevirt.io/mode=unmanaged and platform.kubevirt.io/patch annotations must not be present on the resource.

Procedure

  1. Add the platform.kubevirt.io/ignore-fields annotation with a comma-separated list of RFC 6901 JSON Pointer paths.

    For example, to prevent the virt-platform-autopilot controller from overriding a systemd unit in a managed MachineConfig resource, run the following command:

    $ oc annotate machineconfig 90-worker-swap-online \
  platform.kubevirt.io/ignore-fields='/spec/config/systemd/units/0/contents'

  2. Optional: To restore virt-platform-autopilot controller management for all fields, remove the platform.kubevirt.io/ignore-fields annotation.

    For example, to restore virt-platform-autopilot controller management for all fields of the descheduler configuration, run the following command:

    $ oc annotate kubedescheduler cluster \
  platform.kubevirt.io/ignore-fields- \
  -n openshift-kube-descheduler-operator

Stop reconciliation of a managed resource

Mark a resource as unmanaged so that the virt-platform-autopilot controller does not reconcile it after it is created. You take full ownership of that resource’s configuration.

Deleting an unmanaged resource causes any annotations on that resource to be lost. The virt-platform-autopilot controller then creates a new, opinionated version of the resource.

To prevent the virt-platform-autopilot controller from recreating deleted resources, you must disable the resource. You can disable resources by adding them to the platform.kubevirt.io/disabled-resources annotation in the HyperConverged custom resource (CR).

For more information, see "Exclude a resource from being managed by the virt-platform-autopilot controller".
Prerequisites

  • You have installed the OpenShift CLI (oc).

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

  • The virt-platform-autopilot controller is enabled on the cluster.

  • The platform.kubevirt.io/ignore-fields and platform.kubevirt.io/patch annotations must not be present on the resource.

Procedure

  1. Set the platform.kubevirt.io/mode annotation for the resource to unmanaged.

    For example, to stop the virt-platform-autopilot controller from managing the descheduler configuration, run the following command:

    $ oc annotate kubedescheduler cluster \
  platform.kubevirt.io/mode=unmanaged \
  -n openshift-kube-descheduler-operator

  2. Optional: To restore virt-platform-autopilot controller management for a resource, remove the platform.kubevirt.io/mode annotation.

    For example, to restore virt-platform-autopilot controller management for the descheduler configuration, run the following command:

    $ oc annotate kubedescheduler cluster \
  platform.kubevirt.io/mode- \
  -n openshift-kube-descheduler-operator