Skip to content

Instance types and preferences

FEATURE STATE:

  • instancetype.kubevirt.io/v1alpha1 (Experimental) as of the v0.56.0 KubeVirt release
  • instancetype.kubevirt.io/v1alpha2 (Experimental) as of the v0.58.0 KubeVirt release
  • instancetype.kubevirt.io/v1beta1 as of the v1.0.0 KubeVirt release

See the Version History section for more details.

Introduction

KubeVirt's VirtualMachine API contains many advanced options for tuning the performance of a VM that goes beyond what typical users need to be aware of. Users have previously been unable to simply define the storage/network they want assigned to their VM and then declare in broad terms what quality of resources and kind of performance characteristics they need for their VM.

Instance types and preferences provide a way to define a set of resource, performance and other runtime characteristics, allowing users to reuse these definitions across multiple VirtualMachines.

VirtualMachineInstancetype

---
apiVersion: instancetype.kubevirt.io/v1beta1
kind: VirtualMachineInstancetype
metadata:
  name: example-instancetype
spec:
  cpu:
    guest: 1
  memory:
    guest: 128Mi

KubeVirt provides two CRDs for instance types, a cluster wide VirtualMachineClusterInstancetype and a namespaced VirtualMachineInstancetype. These CRDs encapsulate the following resource related characteristics of a VirtualMachine through a shared VirtualMachineInstancetypeSpec:

  • CPU : Required number of vCPUs presented to the guest
  • Memory : Required amount of memory presented to the guest
  • GPUs : Optional list of vGPUs to passthrough
  • HostDevices : Optional list of HostDevices to passthrough
  • IOThreadsPolicy : Optional IOThreadsPolicy to be used
  • LaunchSecurity: Optional LaunchSecurity to be used

Anything provided within an instance type cannot be overridden within the VirtualMachine. For example, as CPU and Memory are both required attributes of an instance type, if a user makes any requests for CPU or Memory resources within the underlying VirtualMachine, the instance type will conflict and the request will be rejected during creation.

VirtualMachinePreference

---
apiVersion: instancetype.kubevirt.io/v1beta1
kind: VirtualMachinePreference
metadata:
  name: example-preference
spec:
  devices:
    preferredDiskBus: virtio
    preferredInterfaceModel: virtio

KubeVirt also provides two further preference based CRDs, again a cluster wide VirtualMachineClusterPreference and namespaced VirtualMachinePreference. These CRDsencapsulate the preferred value of any remaining attributes of a VirtualMachine required to run a given workload, again this is through a shared VirtualMachinePreferenceSpec.

Unlike instance types, preferences only represent the preferred values and as such, they can be overridden by values in the VirtualMachine provided by the user.

In the example shown below, a user has provided a VirtualMachine with a disk bus already defined within a DiskTarget and has also selected a set of preferences with DevicePreference and preferredDiskBus , so the user's original choice within the VirtualMachine and DiskTarget are used:

$ kubectl apply -f - << EOF
---
apiVersion: instancetype.kubevirt.io/v1beta1
kind: VirtualMachinePreference
metadata:
  name: example-preference-disk-virtio
spec:
  devices:
    preferredDiskBus: virtio
---
apiVersion: kubevirt.io/v1
kind: VirtualMachine
metadata:
  name: example-preference-user-override
spec:
  preference:
    kind: VirtualMachinePreference
    name: example-preference-disk-virtio
  running: false
  template:
    spec:
      domain:
        memory:
          guest: 128Mi
        devices:
          disks:
          - disk:
              bus: sata
            name: containerdisk
          - disk: {}
            name: cloudinitdisk
        resources: {}
      terminationGracePeriodSeconds: 0
      volumes:
      - containerDisk:
          image: registry:5000/kubevirt/cirros-container-disk-demo:devel
        name: containerdisk
      - cloudInitNoCloud:
          userData: |
            #!/bin/sh

            echo 'printed from cloud-init userdata'
        name: cloudinitdisk
EOF
virtualmachinepreference.instancetype.kubevirt.io/example-preference-disk-virtio created
virtualmachine.kubevirt.io/example-preference-user-override configured


$ virtctl start example-preference-user-override
VM example-preference-user-override was scheduled to start

# We can see the original request from the user within the VirtualMachine lists `containerdisk` with a `SATA` bus
$ kubectl get vms/example-preference-user-override -o json | jq .spec.template.spec.domain.devices.disks
[
  {
    "disk": {
      "bus": "sata"
    },
    "name": "containerdisk"
  },
  {
    "disk": {},
    "name": "cloudinitdisk"
  }
]

# This is still the case in the VirtualMachineInstance with the remaining disk using the `preferredDiskBus` from the preference of `virtio`
$ kubectl get vmis/example-preference-user-override -o json | jq .spec.domain.devices.disks
[
  {
    "disk": {
      "bus": "sata"
    },
    "name": "containerdisk"
  },
  {
    "disk": {
      "bus": "virtio"
    },
    "name": "cloudinitdisk"
  }
]

VirtualMachine

---
apiVersion: kubevirt.io/v1
kind: VirtualMachine
metadata:
  name: example-vm
spec:
  instancetype:
    kind: VirtualMachineInstancetype
    name: example-instancetype
  preference:
    kind: VirtualMachinePreference
    name: example-preference

The previous instance type and preference CRDs are matched to a given VirtualMachine through the use of a matcher. Each matcher consists of the following:

  • Name (string): Name of the resource being referenced
  • Kind (string): Optional, defaults to the cluster wide CRD kinds of VirtualMachineClusterInstancetype or VirtualMachineClusterPreference if not provided
  • RevisionName (string) : Optional, name of a ControllerRevision containing a copy of the VirtualMachineInstancetypeSpec or VirtualMachinePreferenceSpec taken when the VirtualMachine is first created. See the Versioning section below for more details on how and why this is captured.
  • InferFromVolume (string): Optional, see the Inferring defaults from a Volume section below for more details.

Creating InstanceTypes, Preferences and VirtualMachines

It is possible to streamline the creation of instance types, preferences, and virtual machines with the usage of the virtctl command-line tool. To read more about it, please see the Creating VirtualMachines.

Versioning

Versioning of these resources is required to ensure the eventual VirtualMachineInstance created when starting a VirtualMachine does not change between restarts if any referenced instance type or set of preferences are updated during the lifetime of the VirtualMachine.

This is currently achieved by using ControllerRevision to retain a copy of the VirtualMachineInstancetype or VirtualMachinePreference at the time the VirtualMachine is created. A reference to these ControllerRevisions are then retained in the InstancetypeMatcher and PreferenceMatcher within the VirtualMachine for future use.

$ kubectl apply -f examples/csmall.yaml -f examples/vm-cirros-csmall.yaml
virtualmachineinstancetype.instancetype.kubevirt.io/csmall created
virtualmachine.kubevirt.io/vm-cirros-csmall created

$ kubectl get vm/vm-cirros-csmall -o json | jq .spec.instancetype
{
  "kind": "VirtualMachineInstancetype",
  "name": "csmall",
  "revisionName": "vm-cirros-csmall-csmall-72c3a35b-6e18-487d-bebf-f73c7d4f4a40-1"
}

$ kubectl get controllerrevision/vm-cirros-csmall-csmall-72c3a35b-6e18-487d-bebf-f73c7d4f4a40-1 -o json | jq .
{
  "apiVersion": "apps/v1",
  "data": {
    "apiVersion": "instancetype.kubevirt.io/v1beta1",
    "kind": "VirtualMachineInstancetype",
    "metadata": {
      "creationTimestamp": "2022-09-30T12:20:19Z",
      "generation": 1,
      "name": "csmall",
      "namespace": "default",
      "resourceVersion": "10303",
      "uid": "72c3a35b-6e18-487d-bebf-f73c7d4f4a40"
    },
    "spec": {
      "cpu": {
        "guest": 1
      },
      "memory": {
        "guest": "128Mi"
      }
    }
  },
  "kind": "ControllerRevision",
  "metadata": {
    "creationTimestamp": "2022-09-30T12:20:19Z",
    "name": "vm-cirros-csmall-csmall-72c3a35b-6e18-487d-bebf-f73c7d4f4a40-1",
    "namespace": "default",
    "ownerReferences": [
      {
        "apiVersion": "kubevirt.io/v1",
        "blockOwnerDeletion": true,
        "controller": true,
        "kind": "VirtualMachine",
        "name": "vm-cirros-csmall",
        "uid": "5216527a-1d31-4637-ad3a-b640cb9949a2"
      }
    ],
    "resourceVersion": "10307",
    "uid": "a7bc784b-4cea-45d7-8432-15418e1dd7d3"
  },
  "revision": 0
}


$ kubectl delete vm/vm-cirros-csmall
virtualmachine.kubevirt.io "vm-cirros-csmall" deleted

$ kubectl get controllerrevision/controllerrevision/vm-cirros-csmall-csmall-72c3a35b-6e18-487d-bebf-f73c7d4f4a40-1
Error from server (NotFound): controllerrevisions.apps "vm-cirros-csmall-csmall-72c3a35b-6e18-487d-bebf-f73c7d4f4a40-1" not found

Users can opt in to moving to a newer generation of an instance type or preference by removing the referenced revisionName from the appropriate matcher within the VirtualMachine object. This will result in fresh ControllerRevisions being captured and used.

The following example creates a VirtualMachine using an initial version of the csmall instance type before increasing the number of vCPUs provided by the instance type:

$ kubectl apply -f examples/csmall.yaml -f examples/vm-cirros-csmall.yaml
virtualmachineinstancetype.instancetype.kubevirt.io/csmall created
virtualmachine.kubevirt.io/vm-cirros-csmall created

$ kubectl get vm/vm-cirros-csmall -o json | jq .spec.instancetype
{
  "kind": "VirtualMachineInstancetype",
  "name": "csmall",
  "revisionName": "vm-cirros-csmall-csmall-3e86e367-9cd7-4426-9507-b14c27a08671-1"
}

$ virtctl start vm-cirros-csmall
VM vm-cirros-csmall was scheduled to start

$ kubectl get vmi/vm-cirros-csmall -o json | jq .spec.domain.cpu
{
  "cores": 1,
  "model": "host-model",
  "sockets": 1,
  "threads": 1
}

$ kubectl patch VirtualMachineInstancetype/csmall --type merge -p '{"spec":{"cpu":{"guest":2}}}'
virtualmachineinstancetype.instancetype.kubevirt.io/csmall patched

In order for this change to be picked up within the VirtualMachine, we need to stop the running VirtualMachine and clear the revisionName referenced by the InstancetypeMatcher:

$ virtctl stop vm-cirros-csmall
VM vm-cirros-csmall was scheduled to stop

$ kubectl patch vm/vm-cirros-csmall --type merge -p '{"spec":{"instancetype":{"revisionName":""}}}'
virtualmachine.kubevirt.io/vm-cirros-csmall patched

$ kubectl get vm/vm-cirros-csmall -o json | jq .spec.instancetype
{
  "kind": "VirtualMachineInstancetype",
  "name": "csmall",
  "revisionName": "vm-cirros-csmall-csmall-3e86e367-9cd7-4426-9507-b14c27a08671-2"
}

As you can see above, the InstancetypeMatcher now references a new ControllerRevision containing generation 2 of the instance type. We can now start the VirtualMachine again and see the new number of vCPUs being used by the VirtualMachineInstance:

$ virtctl start vm-cirros-csmall
VM vm-cirros-csmall was scheduled to start

$ kubectl get vmi/vm-cirros-csmall -o json | jq .spec.domain.cpu
{
  "cores": 1,
  "model": "host-model",
  "sockets": 2,
  "threads": 1
}

inferFromVolume

The inferFromVolume attribute of both the InstancetypeMatcher and PreferenceMatcher allows a user to request that defaults are inferred from a volume. When requested, KubeVirt will look for the following labels on the underlying PVC, DataSource or DataVolume to determine the default name and kind:

  • instancetype.kubevirt.io/default-instancetype
  • instancetype.kubevirt.io/default-instancetype-kind (optional, defaults to VirtualMachineClusterInstancetype)
  • instancetype.kubevirt.io/default-preference
  • instancetype.kubevirt.io/default-preference-kind (optional, defaults to VirtualMachineClusterPreference)

These values are then written into the appropriate matcher by the mutation webhook and used during validation before the VirtualMachine is formally accepted.

The validation can be controlled by the value provided to inferFromVolumeFailurePolicy in either the InstancetypeMatcher or PreferenceMatcher of a VirtualMachine.

The default value of Reject will cause the request to be rejected on failure to find the referenced Volume or labels on an underlying resource.

If Ignore was provided, the respective InstancetypeMatcher or PreferenceMatcher will be cleared on a failure instead.

Example with implicit default value of Reject:

$ kubectl apply -k https://github.com/kubevirt/common-instancetypes.git
[..]
$ virtctl image-upload pvc cirros-pvc --size=1Gi --image-path=./cirros-0.5.2-x86_64-disk.img
[..]
$ kubectl label pvc/cirros-pvc \
  instancetype.kubevirt.io/default-instancetype=server.tiny \
  instancetype.kubevirt.io/default-preference=cirros
[..]
$ kubectl apply -f - << EOF
---
apiVersion: cdi.kubevirt.io/v1beta1
kind: DataSource
metadata:
  name: cirros-datasource
spec:
  source:
    pvc:
      name: cirros-pvc
      namespace: default
---
apiVersion: kubevirt.io/v1
kind: VirtualMachine
metadata:
  name: cirros
spec:
  instancetype:
    inferFromVolume: cirros-volume
  preference:
    inferFromVolume: cirros-volume
  running: false
  dataVolumeTemplates:
    - metadata:
        name: cirros-datavolume
      spec:
        storage:
          resources:
            requests:
              storage: 1Gi
          storageClassName: local
        sourceRef:
          kind: DataSource
          name: cirros-datasource
          namespace: default
  template:
    spec:
      domain:
        devices: {}
      volumes:
        - dataVolume:
            name: cirros-datavolume
          name: cirros-volume
EOF
[..]
kubectl get vms/cirros -o json | jq '.spec.instancetype, .spec.preference'
{
  "kind": "virtualmachineclusterinstancetype",
  "name": "server.tiny",
  "revisionName": "cirros-server.tiny-76454433-3d82-43df-a7e5-586e48c71f68-1"
}
{
  "kind": "virtualmachineclusterpreference",
  "name": "cirros",
  "revisionName": "cirros-cirros-85823ddc-9e8c-4d23-a94c-143571b5489c-1"
}

Example with explicit value of Ignore:

$ virtctl image-upload pvc cirros-pvc --size=1Gi --image-path=./cirros-0.5.2-x86_64-disk.img
$ kubectl apply -f - << EOF
---
apiVersion: cdi.kubevirt.io/v1beta1
kind: DataSource
metadata:
  name: cirros-datasource
spec:
  source:
    pvc:
      name: cirros-pvc
      namespace: default
---
apiVersion: kubevirt.io/v1
kind: VirtualMachine
metadata:
  name: cirros
spec:
  instancetype:
    inferFromVolume: cirros-volume
    inferFromVolumeFailurePolicy: Ignore
  preference:
    inferFromVolume: cirros-volume
    inferFromVolumeFailurePolicy: Ignore
  running: false
  dataVolumeTemplates:
    - metadata:
        name: cirros-datavolume
      spec:
        storage:
          accessModes:
            - ReadWriteOnce
          resources:
            requests:
              storage: 1Gi
          storageClassName: local
        sourceRef:
          kind: DataSource
          name: cirros-datasource
          namespace: default
  template:
    spec:
      domain:
        devices: {}
      volumes:
        - dataVolume:
            name: cirros-datavolume
          name: cirros-volume
EOF
[..]
kubectl get vms/cirros -o json | jq '.spec.instancetype, .spec.preference'
null
null

common-instancetypes

The kubevirt/common-instancetypes provide a set of instancetypes and preferences to help create KubeVirt VirtualMachines.

See Deploy common-instancetypes on how to deploy them.

Examples

Various examples are available within the kubevirt repo under /examples. The following uses an example VirtualMachine provided by the containerdisk/fedora repo and replaces much of the DomainSpec with the equivalent instance type and preferences:

$ kubectl apply -f - << EOF
---
apiVersion: instancetype.kubevirt.io/v1beta1
kind: VirtualMachineInstancetype
metadata:
  name: cmedium
spec:
  cpu:
    guest: 1
  memory:
    guest: 1Gi
---
apiVersion: instancetype.kubevirt.io/v1beta1
kind: VirtualMachinePreference
metadata:
  name: fedora
spec:
  devices:
    preferredDiskBus: virtio
    preferredInterfaceModel: virtio
    preferredRng: {}
  features:
    preferredAcpi: {}
    preferredSmm: {}
  firmware:
    preferredUseEfi: true
    preferredUseSecureBoot: true    
---
apiVersion: kubevirt.io/v1
kind: VirtualMachine
metadata:
  creationTimestamp: null
  name: fedora
spec:
  instancetype:
    name: cmedium
    kind: virtualMachineInstancetype
  preference:
    name: fedora
    kind: virtualMachinePreference
  runStrategy: Always
  template:
    metadata:
      creationTimestamp: null
    spec:
      domain:
        devices: {}
      volumes:
      - containerDisk:
          image: quay.io/containerdisks/fedora:latest
        name: containerdisk
      - cloudInitNoCloud:
          userData: |-
            #cloud-config
            ssh_authorized_keys:
              - ssh-rsa AAAA...
        name: cloudinit
EOF

Version History

instancetype.kubevirt.io/v1alpha1 (Experimental)

  • Initial development version.

instancetype.kubevirt.io/v1alpha2 (Experimental)

  • This version captured complete VirtualMachine{Instancetype,ClusterInstancetype,Preference,ClusterPreference} objects within the created ControllerRevisions

  • This version is backwardly compatible with instancetype.kubevirt.io/v1alpha1.

instancetype.kubevirt.io/v1beta1