Virtual machines

For creating a virtual machine, the VirtualMachine resource is used. Its parameters allow you to configure:

• Virtual machine class;
• Resources required for the virtual machine (CPU, memory, disks, and images);
• Node placement policies for the virtual machine in the cluster;
• Bootloader settings and optimal parameters for the guest OS;
• Virtual machine startup policy and change application policy;
• Initial configuration scripts (cloud-init);
• List of block devices.

Creating a virtual machine

CPU and coreFraction Settings

When creating a virtual machine, you can configure the amount of CPU resources it will use by specifying the cores and coreFraction parameters. The cores parameter defines the number of virtual CPU cores allocated to the VM. The coreFraction parameter sets the guaranteed minimum share of computational power allocated per core.

Allowed values for coreFraction may be defined in the VirtualMachineClass resource for a given range of cores, and only those values are permitted.

For example, if you set cores: 2, the VM will be allocated two virtual CPU cores, which correspond to two physical cores on the hypervisor. With coreFraction: 20%, the VM is guaranteed at least 20% of the processing power of each core, regardless of the hypervisor node load. If additional resources are available on the node, the VM can use up to 100% of each core’s capacity, achieving maximum performance. In this case, the VM is guaranteed 0.2 CPU per physical core but may utilize up to 2 CPUs if the node has idle resources.

If the coreFraction parameter is not specified, each virtual core receives 100% of the corresponding physical core’s processing power.

Configuration example:

spec:
  cpu:
    cores: 2
    coreFraction: 20%

This approach ensures stable VM performance even under high load in oversubscription scenarios, where more CPU cores are allocated to VMs than are physically available on the hypervisor.

The cores and coreFraction parameters are taken into account during VM placement. The guaranteed CPU share is used when selecting a node to ensure the necessary performance can be delivered for all VMs. If a node cannot satisfy the required guarantees, the VM will not be scheduled on it.

Below is a visualization of two virtual machines with different CPU configurations placed on the same node:

Creating VM

Below is an example of a simple virtual machine configuration that runs Ubuntu 22.04. The example uses a cloud-init script that installs the qemu-guest-agent and nginx services, as well as creates the user cloud with the password cloud.

The password in this example was generated using the command mkpasswd --method=SHA-512 --rounds=4096 -S saltsalt. You can change it to your own if needed.

Create a virtual machine with a disk:

d8 k apply -f - <<"EOF"
apiVersion: virtualization.deckhouse.io/v1alpha2
kind: VirtualMachine
metadata:
  name: linux-vm
spec:
  # Virtual machine class name.
  virtualMachineClassName: host
  # Cloud-init script block for provisioning the VM.
  provisioning:
    type: UserData
    # Example cloud-init script to create the user "cloud" with the password "cloud" and install the qemu-guest-agent and nginx services.
    userData: |
      #cloud-config
      package_update: true
      packages:
        - nginx
        - qemu-guest-agent
      run_cmd:
        - systemctl daemon-reload
        - systemctl enable --now nginx.service
        - systemctl enable --now qemu-guest-agent.service
      ssh_pwauth: True
      users:
      - name: cloud
        passwd: '$6$rounds=4096$saltsalt$fPmUsbjAuA7mnQNTajQM6ClhesyG0.yyQhvahas02ejfMAq1ykBo1RquzS0R6GgdIDlvS.kbUwDablGZKZcTP/'
        shell: /bin/bash
        sudo: ALL=(ALL) NOPASSWD:ALL
        lock_passwd: False
      final_message: "The system is finally up, after $UPTIME seconds"
  # VM resource settings.
  cpu:
    # Number of CPU cores.
    cores: 1
    # Request 10% of a physical core's CPU time.
    coreFraction: 10%
  memory:
    # Amount of RAM.
    size: 1Gi
  # List of disks and images used in the VM.
  blockDeviceRefs:
    # The order of disks and images in this block determines the boot priority.
    - kind: VirtualDisk
      name: linux-vm-root
EOF

After creation, the VirtualMachine resource can be in the following states:

• Pending — waiting for the readiness of all dependent resources required to start the virtual machine.
• Starting — the process of starting the virtual machine is in progress.
• Running — the virtual machine is running.
• Stopping — the process of stopping the virtual machine is in progress.
• Stopped — the virtual machine is stopped.
• Terminating — the virtual machine is being deleted.
• Migrating — the virtual machine is in the process of online migration to another node.

Check the state of the virtual machine after creation:

d8 k get vm linux-vm

Example output:

NAME       PHASE     NODE           IPADDRESS     AGE
linux-vm   Running   virtlab-pt-2   10.66.10.12   11m

After creation, the virtual machine will automatically receive an IP address from the range specified in the module settings (block virtualMachineCIDRs).

Virtual Machine resource configuration and sizing policy

The sizing policy in a VirtualMachineClass, defined under .spec.sizingPolicies, sets the rules for configuring virtual machine resources, including the number of CPU cores, memory size, and core usage fraction (coreFraction). This policy is optional. If it is not defined, you can specify arbitrary values for VM resources without strict constraints. However, if a sizing policy is present, the VM configuration must strictly conform to it. Otherwise, the configuration cannot be saved.

The policy divides the number of CPU cores into ranges, such as 1–4 cores or 5–8 cores. For each range, you can specify how much memory is allowed per core and/or which coreFraction values are permitted.

If a VM’s configuration (CPU cores, memory, or coreFraction) does not match the policy, the status will include a condition: type: SizingPolicyMatched, status: False.

If you update the sizing policy in the VirtualMachineClass, existing virtual machines may need to be adjusted to comply with the new policy. VMs that do not match the updated policy will continue to run, but their configurations cannot be modified until they are updated to comply.

Example:

spec:
  sizingPolicies:
    - cores:
        min: 1
        max: 4
      memory:
        min: 1Gi
        max: 8Gi
      coreFractions: [5, 10, 20, 50, 100]
    - cores:
        min: 5
        max: 8
      memory:
        min: 5Gi
        max: 16Gi
      coreFractions: [20, 50, 100]

If a VM is configured with 2 cores, it falls into the 1–4 cores range. In that case, the memory size must be between 1 GiB and 8 GiB, and the coreFraction must be one of: 5%, 10%, 20%, 50%, or 100%. If it has 6 cores, it falls into the 5–8 cores range, where memory must be between 5 GiB and 16 GiB, and coreFraction must be 20%, 50%, or 100%.

In addition to VM sizing, the policy also helps enforce the desired maximum CPU oversubscription. For example, specifying coreFraction: 20% in the policy ensures that each VM gets at least 20% of CPU time per core, effectively allowing up to 5:1 oversubscription.

Virtual Machine Life Cycle

A virtual machine (VM) goes through several phases in its existence, from creation to deletion. These stages are called phases and reflect the current state of the VM. To understand what is happening with the VM, you should check its status (.status.phase field), and for more detailed information — .status.conditions block. All the main phases of the VM life cycle, their meaning and peculiarities are described below.

• Pending — waiting for resources to be ready

  A VM has just been created, restarted or started after a shutdown and is waiting for the necessary resources (disks, images, ip addresses, etc.) to be ready.

  • Possible problems:
    • Dependent resources are not ready: disks, images, VM classes, secret with initial configuration script, etc.

  • Diagnostics: In .status.conditions you should pay attention to *Ready conditions. By them you can determine what is blocking the transition to the next phase, for example, waiting for disks to be ready (BlockDevicesReady) or VM class (VirtualMachineClassReady).

      d8 k get vm <vm-name> -o json | jq '.status.conditions[] | select(.type | test(".*Ready"))'
    

• Starting — starting the virtual machine

  All dependent VM resources are ready and the system is attempting to start the VM on one of the cluster nodes.

  • Possible problems:
    • There is no suitable node to start.
    • There is not enough CPU or memory on suitable nodes.
    • Namespace or project quotas have been exceeded.
  • Diagnostics:

    • If the startup is delayed, check .status.conditions, the type: Running condition

      d8 k get vm <vm-name> -o json | jq '.status.conditions[] | select(.type=="Running")'
      

• Running — the virtual machine is running

  The VM is successfully started and running.

  • Features:
    • When qemu-guest-agent is installed in the guest system, the AgentReady condition will be true and .status.guestOSInfo will display information about the running guest OS.
    • The type: FirmwareUpToDate, status: False condition informs that the VM firmware needs to be updated.
    • Condition type: ConfigurationApplied, status: False informs that the VM configuration is not applied to the running VM.
    • The type: AwaitingRestartToApplyConfiguration, status: True condition displays information about the need to manually reboot the VM because some configuration changes cannot be applied without rebooting the VM.
  • Possible problems:
    • An internal failure in the VM or hypervisor.
  • Diagnosis:

    • Check .status.conditions, condition type: Running.

      d8 k get vm <vm-name> -o json | jq '.status.conditions[] | select(.type=="Running")'
      

• Stopping — The VM is stopped or rebooted.

• Stopped — The VM is stopped and is not consuming computational resources

• Terminating — the VM is deleted.

  This phase is irreversible. All resources associated with the VM are released, but are not automatically deleted.

• Migrating — live migration of a VM

  The VM is migrated to another node in the cluster (live migration).

  • Features:
    • VM migration is supported only for non-local disks, the type: Migratable condition displays information about whether the VM can migrate or not.
  • Possible issues:
    • Incompatibility of processor instructions (when using host or host-passthrough processor types).
    • Difference in kernel versions on hypervisor nodes.
    • Not enough CPU or memory on eligible nodes.
    • Namespace or project quotas have been exceeded.
  • Diagnostics:
    • Check the .status.conditions condition type: Migrating as well as the .status.migrationState block

    d8 k get vm <vm-name> -o json | jq '.status | {condition: .conditions[] | select(.type=="Migrating"), migrationState}'
    

The type: SizingPolicyMatched, status: False condition indicates that the resource configuration does not comply with the sizing policy of the VirtualMachineClass being used. If the policy is violated, it is impossible to save VM parameters without making the resources conform to the policy.

Conditions display information about the state of the VM, as well as on problems that arise. You can understand what is wrong with the VM by analyzing them:

d8 k get vm fedora -o json | jq '.status.conditions[] | select(.message != "")'

Guest OS Agent

To improve VM management efficiency, it is recommended to install the QEMU Guest Agent, a tool that enables communication between the hypervisor and the operating system inside the VM.

How will the agent help?

• It will provide consistent snapshots of disks and VMs.

• Will provide information about the running OS, which will be reflected in the status of the VM. Example:

  status:
    guestOSInfo:
      id: fedora
      kernelRelease: 6.11.4-301.fc41.x86_64
      kernelVersion: '#1 SMP PREEMPT_DYNAMIC Sun Oct 20 15:02:33 UTC 2024'
      machine: x86_64
      name: Fedora Linux
      prettyName: Fedora Linux 41 (Cloud Edition)
      version: 41 (Cloud Edition)
      versionId: “41”
  

• Will allow tracking that the OS has actually booted:

  d8 k get vm -o wide
  

  Sample output (AGENT column):

  NAME     PHASE     CORES   COREFRACTION   MEMORY   NEED RESTART   AGENT   MIGRATABLE   NODE           IPADDRESS    AGE
  fedora   Running   6       5%             8000Mi   False          True    True         virtlab-pt-1   10.66.10.1   5d21h
  

How to install QEMU Guest Agent:

For Debian-based OS:

sudo apt install qemu-guest-agent

For Centos-based OS:

sudo yum install qemu-guest-agent

Starting the agent service:

sudo systemctl enable --now qemu-guest-agent

Automatic CPU Topology Configuration

The CPU topology of a virtual machine (VM) determines how the CPU cores are allocated across sockets. This is important to ensure optimal performance and compatibility with applications that may depend on the CPU configuration. In the VM configuration, you specify only the total number of processor cores, and the topology (the number of sockets and cores in each socket) is automatically calculated based on this value.

The number of processor cores is specified in the VM configuration as follows:

spec:
  cpu:
    cores: 1

Next, the system automatically determines the topology depending on the specified number of cores. The calculation rules depend on the range of the number of cores and are described below.

• If the number of cores is between 1 and 16 (1 ≤ .spec.cpu.cores ≤ 16):
  • 1 socket is used.
  • The number of cores in the socket is equal to the specified value.
  • Change step: 1 (you can increase or decrease the number of cores one at a time).
  • Valid values: any integer from 1 to 16 inclusive.
  • Example: If .spec.cpu.cores = 8, topology: 1 socket with 8 cores.
• If the number of cores is from 17 to 32 (16 < .spec.cpu.cores ≤ 32):
  • 2 sockets are used.
  • Cores are evenly distributed between sockets (the number of cores in each socket is the same).
  • Change step: 2 (total number of cores must be even).
  • Allowed values: 18, 20, 22, 24, 26, 28, 30, 32.
  • Limitations: minimum 9 cores per socket, maximum 16 cores per socket.
  • Example: If .spec.cpu.cores = 20, topology: 2 sockets with 10 cores each.
• If the number of cores is between 33 and 64 (32 < .spec.cpu.cores ≤ 64):
  • 4 sockets are used.
  • Cores are evenly distributed among the sockets.
  • Step change: 4 (the total number of cores must be a multiple of 4).
  • Allowed values: 36, 40, 44, 48, 52, 56, 60, 64.
  • Limitations: minimum 9 cores per socket, maximum 16 cores per socket.
  • Example: If .spec.cpu.cores = 40, topology: 4 sockets with 10 cores each.
• If the number of cores is greater than 64 (.spec.cpu.cores > 64):
  • 8 sockets are used.
  • Cores are evenly distributed among the sockets.
  • Step change: 8 (the total number of cores must be a multiple of 8).
  • Valid values: 72, 80, 88, 88, 96, and so on up to 248
  • Limitations: minimum 9 cores per socket.
  • Example: If .spec.cpu.cores = 80, topology: 8 sockets with 10 cores each.

The change step indicates by how much the total number of cores can be increased or decreased so that they are evenly distributed across the sockets.

The maximum possible number of cores is 248.

The current VM topology (number of sockets and cores in each socket) is displayed in the VM status in the following format:

status:
  resources:
    cpu:
      coreFraction: 10%
      cores: 1
      requestedCores: "1"
      runtimeOverhead: "0"
      topology:
        sockets: 1
        coresPerSocket: 1

Connecting to a virtual machine

There are several ways to connect to a virtual machine:

• Remote management protocol (such as SSH), which must be preconfigured on the virtual machine.
• Serial console.
• VNC protocol.

Example of connecting to a virtual machine using the serial console:

d8 v console linux-vm

Example output:

Successfully connected to linux-vm console. The escape sequence is ^]

linux-vm login: cloud
Password: cloud

To exit the serial console, press Ctrl+].

Example command to connect via VNC:

d8 v vnc linux-vm

Example command to connect via SSH:

d8 v ssh cloud@linux-vm --local-ssh

Startup policy and virtual machine state management

The startup policy of a virtual machine is designed for automated management of the virtual machine’s state. It is defined as the .spec.runPolicy parameter in the virtual machine’s specification. The following policies are supported:

• AlwaysOnUnlessStoppedManually — (default) the VM remains running after creation. If a failure occurs, the VM is automatically restarted. Stopping the VM is only possible by calling the d8 v stop command or creating the corresponding operation.
• AlwaysOn — the VM remains running after creation, even if it is shut down by the OS. If a failure occurs, the VM is automatically restarted.
• Manual — after creation, the VM state is managed manually by the user using commands or operations.
• AlwaysOff — the VM remains off after creation. Turning it on via commands or operations is not possible.

The state of the virtual machine can be managed using the following methods:

• Creating a VirtualMachineOperation (vmop) resource.
• Using the d8 utility with the corresponding subcommand.

The VirtualMachineOperation resource declaratively defines an action that should be performed on the virtual machine.

Example operation to perform a reboot on the virtual machine named linux-vm:

d8 k apply -f - <<EOF
apiVersion: virtualization.deckhouse.io/v1alpha2
kind: VirtualMachineOperation
metadata:
  name: restart-linux-vm-$(date +%s)
spec:
  virtualMachineName: linux-vm
  # Type of operation being applied = Restart operation.
  type: Restart
EOF

You can view the result of the action using the following command:

d8 k get virtualmachineoperation
# or
d8 k get vmop

A similar action can be performed using the d8 utility:

d8 v restart  linux-vm

The possible operations:

Changing the configuration of a virtual machine

The configuration of a virtual machine can be modified at any time after the VirtualMachine resource is created. However, how these changes are applied depends on the current phase of the virtual machine and the nature of the changes.

You can make changes to the virtual machine’s configuration using the following command:

d8 k edit vm linux-vm

If the virtual machine is in a stopped state (.status.phase: Stopped), the changes will take effect as soon as it is started.

If the virtual machine is running (.status.phase: Running), the method of applying the changes depends on their type:

Let’s consider an example of changing the virtual machine’s configuration:

Suppose we want to change the number of CPU cores. Currently, the virtual machine is running and using one core, which can be confirmed by connecting to it via the serial console and running the nproc command.

d8 v ssh cloud@linux-vm --local-ssh --command "nproc"

Example output:

1

Apply the following patch to the virtual machine to change the number of CPU cores from 1 to 2.

d8 k patch vm linux-vm --type merge -p '{"spec":{"cpu":{"cores":2}}}'

Example output:

virtualmachine.virtualization.deckhouse.io/linux-vm patched

The configuration changes have been made, but they have not been applied to the virtual machine yet. Verify this by running the following command again:

d8 v ssh cloud@linux-vm --local-ssh --command "nproc"

Example output:

1

To apply this change, a restart of the virtual machine is required. Run the following command to see the changes that are pending application (which require a restart):

d8 k get vm linux-vm -o jsonpath="{.status.restartAwaitingChanges}" | jq .

Example output:

[
  {
    "currentValue": 1,
    "desiredValue": 2,
    "operation": "replace",
    "path": "cpu.cores"
  }
]

Run the following command:

d8 k get vm linux-vm -o wide

Example output:

NAME        PHASE     CORES   COREFRACTION   MEMORY   NEED RESTART   AGENT   MIGRATABLE   NODE           IPADDRESS     AGE
linux-vm   Running   2       100%           1Gi      True           True    True         virtlab-pt-1   10.66.10.13   5m16s

In the NEED RESTART column, we see True, which indicates that a restart is required to apply the changes.

Let’s restart the virtual machine:

d8 v restart linux-vm

After the restart, the changes will be applied, and the .status.restartAwaitingChanges block will be empty.

Run the following command to verify:

d8 v ssh cloud@linux-vm --local-ssh --command "nproc"

Example output:

2

The default behavior is to apply changes to the virtual machine through a “manual” restart. If you want to apply the changes immediately and automatically, you need to change the change application policy:

spec:
  disruptions:
    restartApprovalMode: Automatic

Initial configuration scripts

Initial configuration scripts are used for the initial setup of a virtual machine when it starts.

The following types of initialization scripts are supported:

• CloudInit.
• Sysprep.

A CloudInit script can be embedded directly within the VM specification, but this script is limited to a maximum length of 2048 bytes:

spec:
  provisioning:
    type: UserData
    userData: |
      #cloud-config
      package_update: true
      ...

For longer initialization scripts or when private data is involved, the initialization script for the virtual machine can be created in a Secret resource. Below is an example of a Secret with a CloudInit script:

apiVersion: v1
kind: Secret
metadata:
  name: cloud-init-example
data:
  userData: <base64 data>
type: provisioning.virtualization.deckhouse.io/cloud-init

Here is a fragment of the virtual machine configuration when using a CloudInit initialization script stored in a Secret resource:

spec:
  provisioning:
    type: UserDataRef
    userDataRef:
      kind: Secret
      name: cloud-init-example

Note: The value of the .data.userData field must be Base64 encoded.

For configuring virtual machines running Windows using Sysprep, only the Secret option is supported.

Here is an example of a secret with a Sysprep script:

apiVersion: v1
kind: Secret
metadata:
  name: sysprep-example
data:
  unattend.xml: <base64 data>
type: provisioning.virtualization.deckhouse.io/sysprep

Note: The value of the .data.unattend.xml field must be Base64 encoded.

Here is the configuration fragment for a virtual machine using the Sysprep initialization script stored in a Secret resource:

spec:
  provisioning:
    type: SysprepRef
    sysprepRef:
      kind: Secret
      name: sysprep-example

Placement of virtual machines on nodes

The following methods can be used to manage the placement of virtual machines (placement parameters) across nodes:

• Simple label selection (nodeSelector) — the basic method for selecting nodes with specified labels.
• Preferred selection (Affinity):
• nodeAffinity — specifies priority nodes for placement.
  • virtualMachineAndPodAffinity — defines workload co-location rules for VMs or containers.
• Co-location avoidance (AntiAffinity):
• virtualMachineAndPodAntiAffinity — defines workload rules for VMs or containers to be placed on the same node.

All of the above parameters (including the .spec.nodeSelector parameter from VirtualMachineClass) are applied together when scheduling VMs. If at least one condition cannot be met, the VM will not be started. To minimize risks, we recommend:

• Creating consistent placement rules.
• Checking the compatibility of rules before applying them.
• Consider the types of conditions:
• Strict (requiredDuringSchedulingIgnoredDuringExecution) — require strict compliance.
• Soft (preferredDuringSchedulingIgnoredDuringExecution) — allow partial compliance.
• Use combinations of labels instead of single restrictions. For example, instead of required for a single label (e.g. env=prod), use several preferred conditions.
• Consider the order in which interdependent VMs are launched. When using Affinity between VMs (for example, the backend depends on the database), launch the VMs referenced by the rules first to avoid lockouts.
• Plan backup nodes for critical workloads. For VMs with strict requirements (e.g., AntiAffinity), provide backup nodes to avoid downtime in case of failure or maintenance.
• Consider existing taints on nodes.

Simple label binding — nodeSelector

nodeSelector is the simplest way to control the placement of virtual machines using a set of labels. It allows you to specify which nodes can run virtual machines by selecting nodes with the required labels.

spec:
  nodeSelector:
    disktype: ssd

In this example, there are three nodes in the cluster: two with fast disks (disktype=ssd) and one with slow disks (disktype=hdd). The virtual machine will only be placed on nodes that have the disktype label with the value ssd.

Preferred affinity

Placement requirements can be:

• Strict (requiredDuringSchedulingIgnoredDuringExecution) — The VM is placed only on nodes that meet the condition.
• Soft (preferredDuringSchedulingIgnoredDuringExecution) — The VM is placed on suitable nodes, if possible.

nodeAffinity — determines on which nodes a VM can be launched using tag expressions.

Example of using nodeAffinity with a strict rule:

spec:
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
          - matchExpressions:
              - key: disktype
                operator: In
                values:
                  - ssd

In this example, there are three nodes in the cluster, two with fast disks (disktype=ssd) and one with slow disks (disktype=hdd). The virtual machine will only be deployed on nodes that have the disktype label with the value ssd.

If you use a soft requirement (preferredDuringSchedulingIgnoredDuringExecution), then if there are no resources to start the VM on nodes with disks labeled disktype=ssd, it will be scheduled on a node with disks labeled disktype=hdd.

virtualMachineAndPodAffinity controls the placement of virtual machines relative to other virtual machines. It allows you to specify a preference for placing virtual machines on the same nodes where certain virtual machines are already running.

Example of a soft rule:

spec:
  affinity:
    virtualMachineAndPodAffinity:
      preferredDuringSchedulingIgnoredDuringExecution:
        - weight: 1
          virtualMachineAndPodAffinityTerm:
            labelSelector:
              matchLabels:
                server: database
            topologyKey: "kubernetes.io/hostname"

In this example, the virtual machine will be placed on nodes that do not have any virtual machine labeled with server: database on the same node, as the goal is to avoid co-location of certain virtual machines.

Avoiding Co-Location — AntiAffinity

AntiAffinity is the opposite of Affinity, and it allows setting requirements to avoid placing virtual machines on the same nodes. This is useful for load distribution or ensuring fault tolerance.

Placement requirements can be strict or soft:

• Strict (requiredDuringSchedulingIgnoredDuringExecution) — The VM is scheduled only on nodes that meet the condition.
• Soft (preferredDuringSchedulingIgnoredDuringExecution) — The VM is scheduled on suitable nodes if possible.

The terms Affinity and AntiAffinity are applicable only to the relationship between virtual machines. For nodes, the corresponding constraints are referred to as nodeAffinity. In nodeAffinity, there is no direct opposite term like in virtualMachineAndPodAffinity. However, you can create opposing conditions by using negative operators in label expressions. To emphasize excluding certain nodes, you can use nodeAffinity with operators like NotIn.

Example using virtualMachineAndPodAntiAffinity:

spec:
  affinity:
    virtualMachineAndPodAntiAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        - labelSelector:
            matchLabels:
              server: database
          topologyKey: "kubernetes.io/hostname"

In this example, the created virtual machine will not be placed on the same node as the virtual machine with the label server: database.

Static and dynamic Block Devices

Block devices can be divided into two types based on how they are connected: static and dynamic (hotplug).

Static Block Devices

Block devices and their features are presented in the table:

Static block devices are specified in the virtual machine specification in the .spec.blockDeviceRefs block as a list. The order of devices in this list determines the sequence in which they are loaded. Thus, if a disk or image is specified first, the bootloader will first try to boot from it. If this fails, the system will move to the next device in the list and try to boot from it. And so on until the first bootloader is detected.

Changing the composition and order of devices in the .spec.blockDeviceRefs block is only possible with a reboot of the virtual machine.

A fragment of the VirtualMachine configuration with a statically connected disk and project image:

spec:
blockDeviceRefs:
- kind: VirtualDisk
name: <virtual-disk-name>
- kind: VirtualImage
name: <virtual-image-name>

Dynamic Block Devices

Dynamic block devices can be connected and disconnected from a running virtual machine without requiring a reboot.

To attach dynamic block devices, the resource VirtualMachineBlockDeviceAttachment (vmbda) is used. Currently, only VirtualDisk is supported for attachment as a dynamic block device.

Create the following resource to attach an empty disk blank-disk to the virtual machine linux-vm:

d8 k apply -f - <<EOF
apiVersion: virtualization.deckhouse.io/v1alpha2
kind: VirtualMachineBlockDeviceAttachment
metadata:
  name: attach-blank-disk
spec:
  blockDeviceRef:
    kind: VirtualDisk
    name: blank-disk
  virtualMachineName: linux-vm
EOF

After creating the VirtualMachineBlockDeviceAttachment, it can be in the following states:

• Pending — waiting for all dependent resources to be ready.
• InProgress — the device attachment process is ongoing.
• Attached — the device is successfully attached.

Diagnosing problems with a resource is done by analyzing the information in the .status.conditions block

Check the state of your resource:

d8 k get vmbda attach-blank-disk

Example output:

NAME                PHASE      VIRTUAL MACHINE NAME   AGE
attach-blank-disk   Attached   linux-vm              3m7s

Connect to the virtual machine and verify that the disk is attached:

d8 v ssh cloud@linux-vm --local-ssh --command "lsblk"

Example output:

NAME    MAJ:MIN RM  SIZE RO TYPE MOUNTPOINTS
sda       8:0    0   10G  0 disk <--- statically attached disk linux-vm-root
|-sda1    8:1    0  9.9G  0 part /
|-sda14   8:14   0    4M  0 part
`-sda15   8:15   0  106M  0 part /boot/efi
sdb       8:16   0    1M  0 disk <--- cloudinit
sdc       8:32   0 95.9M  0 disk <--- dynamically attached disk blank-disk

To detach the disk from the virtual machine, delete the previously created resource:

d8 k delete vmbda attach-blank-disk

Attaching images is done by analogy. To do this, specify VirtualImage or ClusterVirtualImage and the image name as kind:

d8 k apply -f - <<EOF
apiVersion: virtualization.deckhouse.io/v1alpha2
kind: VirtualMachineBlockDeviceAttachment
metadata:
  name: attach-ubuntu-iso
spec:
  blockDeviceRef:
    kind: VirtualImage # or ClusterVirtualImage
    name: ubuntu-iso
  virtualMachineName: linux-vm
EOF

Organizing interaction with virtual machines

Virtual machines can be accessed directly via their fixed IP addresses. However, this approach has limitations — direct use of IP addresses requires manual management, complicates scaling, and makes the infrastructure less flexible. An alternative is services—a mechanism that abstracts access to VMs by providing logical entry points instead of binding to physical addresses.

Services simplify interaction with both individual VMs and groups of similar VMs. For example, the ClusterIP service type creates a fixed internal address that can be used to access both a single VM and a group of VMs, regardless of their actual IP addresses. This allows other system components to interact with resources through a stable name or IP, automatically directing traffic to the right machines.

Services also serve as a load balancing tool — they distribute requests evenly among all connected machines, ensuring fault tolerance and ease of expansion without the need to reconfigure clients.

For scenarios where direct access to specific VMs within the cluster is important (for example, for diagnostics or cluster configuration), headless services can be used. Headless services do not assign a common IP, but instead link the DNS name to the real addresses of all connected machines. A request to such a name returns a list of IPs, allowing you to select the desired VM manually while maintaining the convenience of predictable DNS records.

For external access, services are supplemented with mechanisms such as NodePort, which opens a port on a cluster node, LoadBalancer, which automatically creates a cloud load balancer, or Ingress, which manages HTTP/HTTPS traffic routing.

All these approaches are united by their ability to hide the complexity of the infrastructure behind simple interfaces: clients work with a specific address, and the system itself decides how to route the request to the desired VM, even if its number or status changes.

The service name is formed as <service-name>.<namespace or project name>.svc.<clustername>, or more briefly: <service-name>.<namespace or project name>.svc. For example, if your service name is http and the namespace is default, the full DNS name will be http.default.svc.cluster.local.

The VM’s membership in the service is determined by a set of labels. To set labels on a VM in the context of infrastructure management, use the following command:

d8 k label vm <vm-name> label-name=label-value

Example:

d8 k label vm linux-vm app=nginx

Example output:

virtualmachine.virtualization.deckhouse.io/linux-vm labeled

Headless service

A headless service allows you to easily route requests within a cluster without the need for load balancing. Instead, it simply returns all IP addresses of virtual machines connected to this service.

Even if you use a headless service for only one virtual machine, it is still useful. By using a DNS name, you can access the machine without depending on its current IP address. This simplifies management and configuration because other applications within the cluster can use this DNS name to connect instead of using a specific IP address, which may change.

Example of creating a headless service:

d8 k apply -f - <<EOF
apiVersion: v1
kind: Service
metadata:
  name: http
  namespace: default
spec:
  clusterIP: None
  selector:
    # Label by which the service determines which virtual machine to direct traffic to.
    app: nginx
EOF

After creation, the VM or VM group can be accessed by name: http.default.svc

ClusterIP service

ClusterIP is a standard service type that provides an internal IP address for accessing the service within the cluster. This IP address is used to route traffic between different components of the system. ClusterIP allows virtual machines to interact with each other through a predictable and stable IP address, which simplifies internal communication within the cluster.

Example ClusterIP configuration:

d8 k apply -f - <<EOF
apiVersion: v1
kind: Service
metadata:
  name: http
spec:
  selector:
    # Label by which the service determines which virtual machine to route traffic to.
    app: nginx
EOF

Publish virtual machine services using a service with the NodePort type

NodePort is an extension of the ClusterIP service that provides access to the service through a specified port on all nodes in the cluster. This makes the service accessible from outside the cluster through a combination of the node’s IP address and port.

NodePort is suitable for scenarios where direct access to the service from outside the cluster is required without using a external load balancer.

Create the following service:

d8 k apply -f - <<EOF
apiVersion: v1
kind: Service
metadata:
  name: linux-vm-nginx-nodeport
spec:
  type: NodePort
  selector:
    # label by which the service determines which virtual machine to direct traffic to
    app: nginx
  ports:
    - protocol: TCP
      port: 80
      targetPort: 80
      nodePort: 31880
EOF

In this example, a service with the type NodePort will be created that opens external port 31880 on all nodes in your cluster. This port will forward incoming traffic to internal port 80 on the virtual machine where the Nginx application is running.

If you do not explicitly specify the nodePort value, an arbitrary port will be assigned to the service, which can be viewed in the service status immediately after its creation.

Publishing virtual machine services using a service with the LoadBalancer service type

LoadBalancer is a type of service that automatically creates an external load balancer with a static IP address. This balancer distributes incoming traffic among virtual machines, ensuring the service’s availability from the Internet.

d8 k apply -f - <<EOF
apiVersion: v1
kind: Service
metadata:
  name: linux-vm-nginx-lb
spec:
  type: LoadBalancer
  selector:
    # label by which the service determines which virtual machine to direct traffic to
    app: nginx
  ports:
    - protocol: TCP
      port: 80
      targetPort: 80
EOF

Publish virtual machine services using Ingress

Ingress allows you to manage incoming HTTP/HTTPS requests and route them to different servers within your cluster. This is the most appropriate method if you want to use domain names and SSL termination to access your virtual machines.

To publish a virtual machine service through Ingress, you must create the following resources:

An internal service to bind to Ingress. Example:

d8 k apply -f - <<EOF
apiVersion: v1
kind: Service
metadata:
  name: linux-vm-nginx
spec:
  selector:
    # label by which the service determines which virtual machine to direct traffic to
    app: nginx
  ports:
    - protocol: TCP
      port: 80
      targetPort: 80
EOF

And an Ingress resource for publishing. Example:

d8 k apply -f - <<EOF
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: linux-vm
spec:
  rules:
    - host: linux-vm.example.com
      http:
        paths:
          - path: /
            pathType: Prefix
            backend:
              service:
                name: linux-vm-nginx
                port:
                  number: 80
EOF

Live migration of virtual machines

Live virtual machine (VM) migration is the process of moving a running VM from one physical host to another without shutting it down. This feature plays a key role in the management of virtualized infrastructure, ensuring application continuity during maintenance, load balancing, or upgrades.

How live migration works

The live migration process involves several steps:

1. Creation of a new VM instance

   A new VM is created on the target host in a suspended state. Its configuration (CPU, disks, network) is copied from the source node.

2. Primary Memory Transfer

   The entire RAM of the VM is copied to the target node over the network. This is called primary transfer.

3. Change Tracking (Dirty Pages)

   While memory is being transferred, the VM continues to run on the source node and may change some memory pages. These pages are called dirty pages and the hypervisor marks them.

4. Iterative synchronization.

   After the initial transfer, only the modified pages are resent. This process is repeated in several cycles:

   • The higher the load on the VM, the more “dirty” pages appear, and the longer the migration takes.
   • With good network bandwidth, the amount of unsynchronized data gradually decreases.

5. Final synchronization and switching.

   When the number of dirty pages becomes minimal, the VM on the source node is suspended (typically for 100 milliseconds):

   • The remaining memory changes are transferred to the target node.
   • The state of the CPU, devices, and open connections are synchronized.
   • The VM is started on the new node and the source copy is deleted.

AutoConverge mechanism

If the network struggles to handle data transfer and the number of “dirty” pages keeps growing, the AutoConverge mechanism can be useful. It helps complete migration even with low network bandwidth.

The working principles of AutoConverge mechanism:

1. VM CPU slowdown.

   The hypervisor gradually reduces the CPU frequency of the source VM. This reduces the rate at which new “dirty” pages appear. The higher the load on the VM, the greater the slowdown.

2. Synchronization acceleration.

   Once the data transfer rate exceeds the memory change rate, final synchronization is started and the VM switches to the new node.

3. Automatic termination.

   Final synchronization is started when the data transfer rate exceeds the memory change rate.

AutoConverge is a kind of “insurance” that ensures that the migration completes even if the network struggles to handle data transfer. However, CPU slowdown can affect the performance of applications running on the VM, so its use should be monitored.

Configuring Migration Policy

To configure migration behavior, use the .spec.liveMigrationPolicy parameter in the VM configuration. The following options are available:

• AlwaysSafe: Migration is performed without slowing down the CPU (AutoConverge is not used). Suitable for cases where maximizing VM performance is important but requires high network bandwidth.
• PreferSafe (used as the default policy): By default, migration runs without AutoConverge, but CPU slowdown can be enabled manually if the migration fails to complete. This is done by using the VirtualMachineOperation resource with type=Evict and force=true.
• AlwaysForced: Migration always uses AutoConverge, meaning the CPU is slowed down when necessary. This ensures that the migration completes even if the network is bad, but may degrade VM performance.
• PreferForced: By default migration goes with AutoConverge, but slowdown can be manually disabled via VirtualMachineOperation with the parameter type=Evict and force=false.

Migration Types

Migration can be performed manually by the user, or automatically by the following system events:

• Updating the “firmware” of a virtual machine.
• Redistribution of load in the cluster.
• Transferring a node into maintenance mode (Node drain).
• When you change VM placement settings (not available in Community edition).

The trigger for live migration is the appearance of the VirtualMachineOperations resource with the Evict type.

The table shows the VirtualMachineOperations resource name prefixes with the Evict type that are created for live migrations caused by system events:

This resource can be in the following states:

• Pending — the operation is pending.
• InProgress — live migration is in progress.
• Completed — live migration of the virtual machine has been completed successfully.
• Failed — the live migration of the virtual machine has failed.

Diagnosing problems with a resource is done by analyzing the information in the .status.conditions block.

You can view active operations using the command:

d8 k get vmop

Example output:

NAME                    PHASE       TYPE    VIRTUALMACHINE      AGE
firmware-update-fnbk2   Completed   Evict   static-vm-node-00   148m

You can interrupt any live migration while it is in the Pending, InProgress phase by deleting the corresponding VirtualMachineOperations resource.

How to perform a live migration of a virtual machine using VirtualMachineOperations

Let’s look at an example. Before starting the migration, view the current status of the virtual machine:

d8 k get vm

Example output:

NAME       PHASE     NODE           IPADDRESS     AGE
linux-vm   Running   virtlab-pt-1   10.66.10.14   79m

The virtual machine is running on the virtlab-pt-1 node.

To migrate a virtual machine from one host to another, taking into account the virtual machine placement requirements, the command is used:

d8 v evict -n <namespace> <vm-name>

execution of this command leads to the creation of the VirtualMachineOperations resource. You can also start the migration by creating a VirtualMachineOperations (vmop) resource with the Evict type manually:

d8 k apply -f - <<EOF
apiVersion: virtualization.deckhouse.io/v1alpha2
kind: VirtualMachineOperation
metadata:
  name: evict-linux-vm-$(date +%s)
spec:
  # name of the virtual machine
  virtualMachineName: linux-vm
  # operation for migration
  type: Evict
EOF

To track the migration of a virtual machine immediately after the vmop resource is created, run the command:

d8 k get vm -w

Example output:

NAME       PHASE       NODE           IPADDRESS     AGE
linux-vm   Running     virtlab-pt-1   10.66.10.14   79m
linux-vm   Migrating   virtlab-pt-1   10.66.10.14   79m
linux-vm   Migrating   virtlab-pt-1   10.66.10.14   79m
linux-vm   Running     virtlab-pt-2   10.66.10.14   79m

You can also perform the migration using the following command:

d8 v evict <vm-name>

Live migration of virtual machine when changing placement parameters (not available in CE edition)

Let’s consider the migration mechanism on the example of a cluster with two node groups (NodeGroups): green and blue. Suppose a virtual machine (VM) is initially running on a node in the green group and its configuration contains no placement restrictions.

Step 1: Add the placement parameter Let’s specify in the VM specification the requirement for placement in the green group :

spec:
  nodeSelector:
    node.deckhouse.io/group: green

After saving the changes, the VM will continue to run on the current node, since the nodeSelector condition is already met.

Step 2: Change the placement parameter Let’s change the placement requirement to group blue :

spec:
  nodeSelector:
    node.deckhouse.io/group: blue

Now the current node (groups green) does not match the new conditions. The system will automatically create a VirtualMachineOperations object of type Evict, which will initiate a live migration of the VM to an available node in group blue .

Maintenance mode

When performing work on nodes with running virtual machines, there is a risk of disrupting their functionality. To avoid this, the node can be put into maintenance mode and the virtual machines can be migrated to other free nodes. To do this, run the following command:

d8 k drain <nodename> --ignore-daemonsets --delete-emptydir-dat

where <nodename> is the node on which the work is supposed to be performed and which must be freed from all resources (including system resources).

If you need to evict only virtual machines from a node, run the following command:

d8 k drain <nodename> --pod-selector vm.kubevirt.internal.virtualization.deckhouse.io/name --delete-emptydir-data

After running the d8 k drain commands, the node will go into maintenance mode and virtual machines will not be able to start on it. To take it out of maintenance mode, run the following command:

d8 k uncordon <nodename>

IP Addresses of virtual machines

The .spec.settings.virtualMachineCIDRs block in the virtualization module configuration specifies a list of subnets for assigning IP addresses to virtual machines (a shared pool of IP addresses). All addresses in these subnets are available for use, except for the first (network address) and the last (broadcast address).

The VirtualMachineIPAddressLease (vmipl) resource is a cluster-wide resource that manages the temporary allocation of IP addresses from the shared pool specified in virtualMachineCIDRs.

To view the list of temporarily allocated IP addresses (vmipl), use the following command:

d8 k get vmipl

Example output:

NAME             VIRTUALMACHINEIPADDRESS                             STATUS   AGE
ip-10-66-10-14   {"name":"linux-vm-7prpx","namespace":"default"}     Bound    12h

The VirtualMachineIPAddress (vmip) resource is a project or namespace resource responsible for reserving allocated IP addresses and binding them to virtual machines. IP addresses can be assigned automatically or upon request.

To check the assigned IP address, you can use the following command:

d8 k get vmip

Example output:

NAME             ADDRESS       STATUS     VM         AGE
linux-vm-7prpx   10.66.10.14   Attached   linux-vm   12h

The algorithm for automatically assigning an IP address to a virtual machine works as follows:

• The user creates a virtual machine with the name <vmname>.
• The module controller automatically creates a vmip resource with the name <vmname>-<hash> to request an IP address and associate it with the virtual machine.
• A vmipl lease resource is created for this, which selects a random IP address from the general pool.
• Once the vmip resource is created, the virtual machine is assigned the IP address.

By default, the IP address for the virtual machine is automatically assigned from the subnets defined in the module and is bound to it until the virtual machine is deleted. After the virtual machine is deleted, the vmip resource is also removed, but the IP address temporarily remains bound to the project/namespace and can be requested again.

Requesting the required IP address

Create the vmip resource:

d8 k apply -f - <<EOF
apiVersion: virtualization.deckhouse.io/v1alpha2
kind: VirtualMachineIPAddress
metadata:
  name: linux-vm-custom-ip
spec:
  staticIP: 10.66.20.77
  type: Static
EOF

Create a new or modify an existing virtual machine and explicitly specify the required vmip resource in the specification:

spec:
  virtualMachineIPAddressName: linux-vm-custom-ip

Retaining the IP address assigned to a virtual machine

To prevent the automatically assigned IP address of a virtual machine from being deleted along with the virtual machine itself, follow these steps.

Obtain the vmip resource name for the specified virtual machine:

d8 k get vm linux-vm -o jsonpath="{.status.virtualMachineIPAddressName}"

Example output:

linux-vm-7prpx

Remove the .metadata.ownerReferences blocks from the found resource:

d8 k patch vmip linux-vm-7prpx --type=merge --patch '{"metadata":{"ownerReferences":null}}'

After deleting the virtual machine, the vmip resource will persist and can be used for a newly created virtual machine:

spec:
  virtualMachineIPAddressName: linux-vm-7prpx

Even if the vmip resource is deleted, it remains leased to the current project/namespace for another 10 minutes, and there is an option to re-lease it upon request:

d8 k apply -f - <<EOF
apiVersion: virtualization.deckhouse.io/v1alpha2
kind: VirtualMachineIPAddress
metadata:
  name: linux-vm-custom-ip
spec:
  staticIP: 10.66.20.77
  type: Static
EOF