Container Storage Interface Snapshot Support in Velero

Integrating Container Storage Interface (CSI) snapshot support into Velero enables Velero to backup and restore CSI-backed volumes using the Kubernetes CSI Snapshot APIs.

By supporting CSI snapshot APIs, Velero can support any volume provider that has a CSI driver, without requiring a Velero-specific plugin to be available. This page gives an overview of how to add support for CSI snapshots to Velero through CSI plugins. For more information about specific components, see the plugin repo.


  1. Your cluster is Kubernetes version 1.20 or greater.
  2. Your cluster is running a CSI driver capable of support volume snapshots at the v1 API level.
  3. When restoring CSI VolumeSnapshots across clusters, the name of the CSI driver in the destination cluster is the same as that on the source cluster to ensure cross cluster portability of CSI VolumeSnapshots

NOTE: Not all cloud provider’s CSI drivers guarantee snapshot durability, meaning that the VolumeSnapshot and VolumeSnapshotContent objects may be stored in the same object storage system location as the original PersistentVolume and may be vulnerable to data loss. You should refer to your cloud provider’s documentation for more information on configuring snapshot durability. Since v0.3.0 the velero team will provide official support for CSI plugin when they are used with AWS and Azure drivers.

Installing Velero with CSI support

To integrate Velero with the CSI volume snapshot APIs, you must enable the EnableCSI feature flag and install the Velero CSI plugins on the Velero server.

Both of these can be added with the velero install command.

velero install \
--features=EnableCSI \
--plugins=<object storage plugin>,velero/velero-plugin-for-csi:v0.6.0 \

To include the status of CSI objects associated with a Velero backup in velero backup describe output, run velero client config set features=EnableCSI. See Enabling Features for more information about managing client-side feature flags. You can also view the image on Docker Hub.

Implementation Choices

This section documents some of the choices made during implementation of the Velero CSI plugins:

  1. VolumeSnapshots created by the Velero CSI plugins are retained only for the lifetime of the backup even if the DeletionPolicy on the VolumeSnapshotClass is set to Retain. To accomplish this, during deletion of the backup the prior to deleting the VolumeSnapshot, VolumeSnapshotContent object is patched to set its DeletionPolicy to Delete. Deleting the VolumeSnapshot object will result in cascade delete of the VolumeSnapshotContent and the snapshot in the storage provider.
  2. VolumeSnapshotContent objects created during a velero backup that are dangling, unbound to a VolumeSnapshot object, will be discovered, using labels, and deleted on backup deletion.
  3. The Velero CSI plugins, to backup CSI backed PVCs, will choose the VolumeSnapshotClass in the cluster based on the following logic:
    1. Default Behavior: You can simply create a VolumeSnapshotClass for a particular driver and put a label on it to indicate that it is the default VolumeSnapshotClass for that driver. For example, if you want to create a VolumeSnapshotClass for the CSI driver for taking snapshots of disks created with based storage classes, you can create a VolumeSnapshotClass like this:

      kind: VolumeSnapshotClass
        name: test-snapclass

      Note: For each driver type, there should only be 1 VolumeSnapshotClass with the label "true".

    2. Choose VolumeSnapshotClass for a particular Backup Or Schedule: If you want to use a particular VolumeSnapshotClass for a particular backup or schedule, you can add a annotation to the backup or schedule to indicate which VolumeSnapshotClass to use. For example, if you want to use the VolumeSnapshotClass test-snapclass for a particular backup for snapshotting PVCs of, you can create a backup like this:

      kind: Backup
        name: test-backup
          - default

      Note: Please ensure all your annotations are in lowercase. And follow the following format:<driver name> = <VolumeSnapshotClass Name>

    3. Choosing VolumeSnapshotClass for a particular PVC: If you want to use a particular VolumeSnapshotClass for a particular PVC, you can add a annotation to the PVC to indicate which VolumeSnapshotClass to use. This overrides any annotation added to backup or schedule. For example, if you want to use the VolumeSnapshotClass test-snapclass for a particular PVC, you can create a PVC like this:

      apiVersion: v1
      kind: PersistentVolumeClaim
        name: test-pvc
          - ReadWriteOnce
              storage: 1Gi
  4. The VolumeSnapshot objects will be removed from the cluster after the backup is uploaded to the object storage, so that the namespace that is backed up can be deleted without removing the snapshot in the storage provider if the DeletionPolicy is Delete.

How it Works - Overview

Velero’s CSI support does not rely on the Velero VolumeSnapshotter plugin interface.

Instead, Velero uses a collection of BackupItemAction plugins that act first against PersistentVolumeClaims.

When this BackupItemAction sees PersistentVolumeClaims pointing to a PersistentVolume backed by a CSI driver, it will choose the VolumeSnapshotClass with the same driver name that has the label to create a CSI VolumeSnapshot object with the PersistentVolumeClaim as a source. This VolumeSnapshot object resides in the same namespace as the PersistentVolumeClaim that was used as a source.

From there, the CSI external-snapshotter controller will see the VolumeSnapshot and create a VolumeSnapshotContent object, a cluster-scoped resource that will point to the actual, disk-based snapshot in the storage system. The external-snapshotter plugin will call the CSI driver’s snapshot method, and the driver will call the storage system’s APIs to generate the snapshot. Once an ID is generated and the storage system marks the snapshot as usable for restore, the VolumeSnapshotContent object will be updated with a status.snapshotHandle and the status.readyToUse field will be set.

Velero will include the generated VolumeSnapshot and VolumeSnapshotContent objects in the backup tarball, as well as upload all VolumeSnapshots and VolumeSnapshotContents objects in a JSON file to the object storage system. Note that only Kubernetes objects are uploaded to the object storage, not the data in snapshots.

When Velero synchronizes backups into a new cluster, VolumeSnapshotContent objects and the VolumeSnapshotClass that is chosen to take snapshot will be synced into the cluster as well, so that Velero can manage backup expiration appropriately.

The DeletionPolicy on the VolumeSnapshotContent will be the same as the DeletionPolicy on the VolumeSnapshotClass that was used to create the VolumeSnapshot. Setting a DeletionPolicy of Retain on the VolumeSnapshotClass will preserve the volume snapshot in the storage system for the lifetime of the Velero backup and will prevent the deletion of the volume snapshot, in the storage system, in the event of a disaster where the namespace with the VolumeSnapshot object may be lost.

When the Velero backup expires, the VolumeSnapshot objects will be deleted and the VolumeSnapshotContent objects will be updated to have a DeletionPolicy of Delete, to free space on the storage system.

For more details on how each plugin works, see the CSI plugin repo’s documentation.

Note: The AWS, Microsoft Azure, and Google Cloud Platform (GCP) Velero plugins version 1.4 and later are able to snapshot and restore persistent volumes provisioned by a CSI driver via the APIs of the cloud provider, without having to install Velero CSI plugins. See the AWS, Microsoft Azure, and Google Cloud Platform (GCP) Velero plugin repo for more information on supported CSI drivers.

Getting Started

To help you get started, see the documentation.