Alauda Container Platform

Concepts and Glossary

This page is for administrators who are fluent in Kubernetes but new to virtualization. The rest of the manual assumes the terms below; each is defined once here, with the closest Kubernetes analogy, so the how-to pages stay short.

TOC

How it fits togetherCompute objectsImages and disksNetworkingMigration and availabilityBackup and recoveryWhere to start

How it fits together

Virtualization on this platform is KubeVirt and CDI, deployed and reconciled by one operator:

  • HyperConverged (HCO) — the HyperConverged custom resource is the single source of truth. The operator installs and continuously reconciles KubeVirt and CDI from it, so cluster-wide settings (for example live-migration limits) are set on HyperConverged, not directly on the KubeVirt resource. Analogy: a top-level operator CR, like an Installation/Subscription that owns everything below it.
  • KubeVirt — runs the virtual machines. Its controllers are virt-controller (cluster-level, schedules VMs), virt-handler (a DaemonSet, one per node, like the kubelet for VMs), and virt-launcher (one pod per running VM that wraps the QEMU process).
  • CDI (Containerized Data Importer) — imports and clones disk images into PVCs. It is the engine behind bootable volumes and VM disks.

A running virtual machine is therefore an ordinary Pod (the virt-launcher pod) on an ordinary node, holding an ordinary PVC for its disk and getting an ordinary Pod IP — which is why your existing Kubernetes tooling (RBAC, NetworkPolicy, scheduling, monitoring) still applies.

Compute objects

TermWhat it isKubernetes analogy
VirtualMachine (kubevirt.io/v1)The desired-state definition of a VM (CPU, memory, disks, network, whether it should be running).A Deployment — the spec you create and edit.
VirtualMachineInstance (VMI)The running instance produced when a VirtualMachine is started. Lifecycle subresources (console, VNC, pause, migrate) act on the VMI.A Pod — created from the spec, exists only while running.
virt-launcher podThe pod that wraps one running VM's QEMU process.The Pod that actually executes the workload.
spec.runningBoolean power switch on the VirtualMachine (true starts it, false stops it). This platform uses this field.replicas: 1 vs 0.
spec.runStrategyThe upstream alternative to spec.running (Always / RerunOnFailure / Manual / Halted) for finer restart behavior.A restart policy.
instancetypeA reusable, cluster-wide CPU/memory profile a VM can reference instead of inline resources.A sizing preset / LimitRange-like template.
preferenceA reusable profile of OS-specific device/firmware defaults (bus types, clock, features).A device/firmware "profile" applied to the VM.
QEMU guest agentAn agent installed inside the guest OS. Enables IP reporting, graceful shutdown, and filesystem freeze for application-consistent snapshots. Without it, the platform cannot see the guest IP and online snapshots are only crash-consistent.A sidecar/agent inside the workload that reports health and responds to signals. See Installing Guest Tools.
vTPMA virtual TPM 2.0 device (spec.template.spec.domain.devices.tpm). Required by Windows 11 and BitLocker.A virtual security device attached to the VM.

Images and disks

A VM boots from a disk that is cloned from a reusable image. The chain is:

remote image (HTTP / registry / S3)  ──CDI import──▶  PVC (a DataVolume)  ──published as──▶  DataSource (a "bootable volume")  ──cloned at VM create──▶  the VM's boot disk
TermWhat it isKubernetes analogy
qcow2 / rawDisk-image file formats. qcow2 is sparse/copy-on-write; raw is a flat image. This is the content you import, not a container image.A disk image artifact, like an OS cloud image.
containerDiskA VM disk packaged inside a container image and pulled like one. Note: a normal application container image is not a containerDisk and will not boot.An OCI image whose payload is a bootable disk.
CDI / DataVolumeA DataVolume (cdi.kubevirt.io/v1beta1) declares "import this image into a PVC"; CDI does the import and creates the backing PVC.A PVC plus an init job that populates it.
DataSource / bootable volumeA DataSource (cdi.kubevirt.io/v1beta1) points at a golden-image PVC, also called a Bootable Volume; creating a VM clones it into the VM's boot disk.A read-only "golden" PVC used as a template. See Bootable Volumes.
volumeMode: Block vs FilesystemHow the PVC is presented. Block performs closer to bare metal and is preferred for VM disks; Filesystem also works.The PVC volumeMode you already know.
StorageProfile / cloneStrategyA CDI-published object per StorageClass describing supported access/volume modes and how clones are made (csi-clone, snapshot, or host-assisted copy).Capabilities metadata for a StorageClass. See Managing Virtual Disks.

A fresh cluster has no bootable volumes until an administrator imports one (or publishes golden images into kube-public). Start at Bootable Volumes.

Networking

By default a VM runs in a virt-launcher Pod and uses that Pod's network — so it gets a Pod IP and is subject to the same CNI and NetworkPolicy as any Pod. The create form's Network Mode maps to a KubeVirt binding:

UI labelYAML bindingBehavior
Bridgedbridge: {}The VM takes over the Pod's network interface and uses the Pod IP directly. Required for some protocols; not live-migration friendly by default — live migration over a bridge requires the kubevirt.io/allow-pod-bridge-network-live-migration annotation and a supported CNI (see Live Migration).
NATmasquerade: {}The VM gets a private internal IP and is NAT'd out through the Pod IP. The default; live-migration friendly.

"Container group" in the UI means the Kubernetes Pod that backs the VM.

For a routable IP on the physical network (instead of a Pod IP), use a Kube-OVN Underlay subnet or an auxiliary NIC. Secondary NICs attach via a NetworkAttachmentDefinition (Multus).

Migration and availability

TermWhat it isKubernetes analogy
Live migrationMoving a running VM to another node with no downtime, by copying its memory. Pre-copy copies memory iteratively while the VM runs; post-copy finishes copying after the VM is already on the target.Rescheduling a Pod without restarting the process — which Pods cannot normally do. See Live Migration.
VirtualMachineInstanceMigrationThe request object that triggers one live migration; deleting it cancels an in-progress migration.A one-shot Job that moves the VMI.
MigrationPolicyCluster-scoped overrides (per namespace/VM) for migration tuning. Cluster defaults live on HyperConverged.spec.liveMigrationConfig.A policy CR selected by labels.
evictionStrategyWhat happens on node drain: LiveMigrate, LiveMigrateIfPossible, or None. A non-migratable VM (RWO storage or device passthrough) with LiveMigrate blocks the drain — and therefore cluster upgrades.A PodDisruptionBudget-like constraint, but it migrates instead of just blocking. See High Availability.
Live migration requires shared storageA VM can only live-migrate when all its disk PVCs use ReadWriteMany (RWX) access mode (for example CephRBD in Block mode).RWX vs RWO PVCs — the same concept.
Storage migrationMoving a running VM's disks between StorageClasses online.Re-homing a PVC to a different StorageClass without downtime.

Backup and recovery

TermWhat it isKubernetes analogy
VirtualMachineSnapshot / VirtualMachineRestorePoint-in-time snapshot of a VM's disks (and config) and its restore object. Backed by CSI VolumeSnapshot.A VolumeSnapshot, scoped to the whole VM.
crash-consistent vs application-consistentWithout the guest agent a snapshot is crash-consistent (like pulling the power). With the guest agent the filesystem is quiesced first, giving an application-consistent snapshot.The fsfreeze distinction familiar from storage backups.

Where to start

A first-time path through the manual:

  1. Install — enable virtualization and deploy the HCO operator.
  2. Bootable Volumes — import a usable OS image (a fresh cluster has none).
  3. Creating Virtual Machines — create your first VM from that image.
  4. Serial Console — log in (cloud images have no default password; use the SSH key or password you set at create time).
  5. Managing Virtual Disks — attach a data disk.

The overview also has a "Console feature → API object" mapping table, and every how-to page carries a copy-pasteable Using the API block.