About DataVolumes

DataVolume objects are custom resources that are provided by the Containerized Data Importer (CDI) project. DataVolumes orchestrate import, clone, and upload operations that are associated with an underlying PersistentVolumeClaim (PVC). DataVolumes are integrated with KubeVirt, and they prevent a virtual machine from being started before the PVC has been prepared.

Understanding scratch space

The Containerized Data Importer (CDI) requires scratch space (temporary storage) to complete some operations, such as importing and uploading virtual machine images. During this process, the CDI provisions a scratch space PVC equal to the size of the PVC backing the destination DataVolume (DV). The scratch space PVC is deleted after the operation completes or aborts.

The CDIConfig object allows you to define which StorageClass to use to bind the scratch space PVC by setting the scratchSpaceStorageClass in the spec: section of the CDIConfig object.

If the defined StorageClass does not match a StorageClass in the cluster, then the default StorageClass defined for the cluster is used. If there is no default StorageClass defined in the cluster, the StorageClass used to provision the original DV or PVC is used.

The CDI requires requesting scratch space with a file volume mode, regardless of the PVC backing the origin DataVolume. If the origin PVC is backed by block volume mode, you must define a StorageClass capable of provisioning file volume mode PVCs.

Manual provisioning

If there are no storage classes, the CDI will use any PVCs in the project that match the size requirements for the image. If there are no PVCs that match these requirements, the CDI import Pod will remain in a Pending state until an appropriate PVC is made available or until a timeout function kills the Pod.

Defining a StorageClass in the CDI configuration

Define a StorageClass in the CDI configuration to dynamically provision scratch space for CDI operations.

Procedure
  • Use the oc client to edit the cdiconfig/config and add or edit the spec: scratchSpaceStorageClass to match a StorageClass in the cluster.

    $ oc edit cdiconfig/config
    API Version:  cdi.kubevirt.io/v1alpha1
    kind: CDIConfig
    metadata:
      name: config
    ...
    spec:
      scratchSpaceStorageClass: "<storage_class>"
    ...

CDI operations that require scratch space

Type Reason

Registry imports

The CDI must download the image to a scratch space and extract the layers to find the image file. The image file is then passed to QEMU-IMG for conversion to a raw disk.

Upload image

QEMU-IMG does not accept input from STDIN. Instead, the image to upload is saved in scratch space before it can be passed to QEMU-IMG for conversion.

HTTP imports of archived images

QEMU-IMG does not know how to handle the archive formats CDI supports. Instead, the image is unarchived and saved into scratch space before it is passed to QEMU-IMG.

HTTP imports of authenticated images

QEMU-IMG inadequately handles authentication. Instead, the image is saved to scratch space and authenticated before it is passed to QEMU-IMG.

HTTP imports of custom certificates

QEMU-IMG inadequately handles custom certificates of HTTPS endpoints. Instead, the CDI downloads the image to scratch space before passing the file to QEMU-IMG.

CDI supported operations matrix

This matrix shows the supported CDI operations for content types against endpoints, and which of these operations requires scratch space.

Content types HTTP HTTPS HTTP basic auth Registry S3 Bucket Upload

KubeVirt(QCOW2)

✓ QCOW2
✓ GZ*
✓ XZ*

✓ QCOW2**
✓ GZ*
✓ XZ*

✓ QCOW2
✓ GZ*
✓ XZ*

✓ QCOW2*
□ GZ
□ XZ

✓ QCOW2*
✓ GZ*
✓ XZ*

✓ QCOW2*
✓ GZ*
✓ XZ*

KubeVirt (RAW)

✓ RAW
✓ GZ
✓ XZ

✓ RAW
✓ GZ
✓ XZ

✓ RAW
✓ GZ
✓ XZ

✓ RAW*
□ GZ
□ XZ

✓ RAW
✓ GZ
✓ XZ

✓ RAW*
✓ GZ*
✓ XZ*

Archive+

✓ TAR

✓ TAR

✓ TAR

□ TAR

□ TAR

□ TAR

✓ Supported operation

□ Unsupported operation

* Requires scratch space

** Requires scratch space if a custom certificate authority is required

+ Archive does not support block mode DVs

Additional resources
  • See the Dynamic provisioning section for more information on StorageClasses and how these are defined in the cluster.