Skip to content

Memory Hotplug

Memory hotplug was introduced in KubeVirt version 1.1, enabling the dynamic resizing of the amount of memory available to a running VM.

Limitations

  • Memory hotplug is currently only supported on the x86_64 architecture.
  • Current hotplug implementation involves live-migration of the VM workload.

Configuration

Enable feature-gate

To use memory hotplug we need to add the VMLiveUpdateFeatures feature gate in the KubeVirt CR:

apiVersion: kubevirt.io/v1
kind: KubeVirt
spec:
  configuration:
    developerConfiguration:
      featureGates:
        - VMLiveUpdateFeatures

Configure the Workload Update Strategy

Configure LiveMigrate as workloadUpdateStrategy in the KubeVirt CR, since the current implementation of the hotplug process requires the VM to live-migrate.

apiVersion: kubevirt.io/v1
kind: KubeVirt
spec:
  workloadUpdateStrategy:
    workloadUpdateMethods:
    - LiveMigrate

Configure the VM rollout strategy

Finally, set the VM rollout strategy to LiveUpdate, so that the changes made to the VM object propagate to the VMI without a restart.
This is also done in the KubeVirt CR configuration:

apiVersion: kubevirt.io/v1
kind: KubeVirt
spec:
  configuration:
    vmRolloutStrategy: "LiveUpdate"

NOTE: If memory hotplug is enabled/disabled on an already running VM, a reboot is necessary for the changes to take effect.

More information can be found on the VM Rollout Strategies page.

[OPTIONAL] Set a cluster-wide maximum amount of memory

You can set the maximum amount of memory for the guest using a cluster level setting in the KubeVirt CR.

apiVersion: kubevirt.io/v1
kind: KubeVirt
spec:
  configuration:
    liveUpdateConfiguration:
      maxGuest: 8Gi

The VM-level configuration will take precedence over the cluster-wide one.

Memory Hotplug in Action

First we enable the VMLiveUpdateFeatures feature gate, set the rollout strategy to LiveUpdate and set LiveMigrate as workloadUpdateStrategy in the KubeVirt CR.

$ kubectl --namespace kubevirt patch kv kubevirt -p='[{"op": "add", "path": "/spec/configuration/developerConfiguration/featureGates", "value": ["VMLiveUpdateFeatures"]}]' --type='json'
$ kubectl --namespace kubevirt patch kv kubevirt -p='[{"op": "add", "path": "/spec/configuration/vmRolloutStrategy", "value": "LiveUpdate"}]' --type='json'
$ kubectl --namespace kubevirt patch kv kubevirt -p='[{"op": "add", "path": "/spec/workloadUpdateStrategy/workloadUpdateMethods", "value": ["LiveMigrate"]}]' --type='json'

Now we create a VM with memory hotplug enabled.

apiVersion: kubevirt.io/v1
kind: VirtualMachine
metadata:
  name: vm-cirros
spec:
  running: true
  template:
    spec:
      domain:
        memory:
          maxGuest: 2Gi
          guest: 128Mi
        devices:
          disks:
          - disk:
              bus: virtio
            name: containerdisk
      volumes:
      - containerDisk:
          image: registry:5000/kubevirt/alpine-container-disk-demo:devel
        name: containerdisk

The Virtual Machine will automatically start and once booted it will report the currently available memory to the guest in the status.memory field inside the VMI.

$ kubectl get vmi vm-cirros -o json | jq .status.memory
{
  "guestAtBoot": "128Mi",
  "guestCurrent": "128Mi",
  "guestRequested": "128Mi"
}

Since the Virtual Machine is now running we can patch the VM object to double the available guest memory so that we'll go from 128Mi to 256Mi.

$ kubectl patch vm vm-cirros -p='[{"op": "replace", "path": "/spec/template/spec/domain/memory/guest", "value": "256Mi"}]' --type='json'

After the hotplug request is processed and the Virtual Machine is live migrated, the new amount of memory should be available to the guest and visible in the VMI object.

$ kubectl get vmi vm-cirros -o json | jq .status.memory
{
  "guestAtBoot": "128Mi",
  "guestCurrent": "256Mi",
  "guestRequested": "256Mi"
}