Restoring block device from S3 using Kopia doesn't create deduplicated PVCs

[SkalaNetworks]

Hi, I'm using Velero to backup Kubevirt VMs.

The VMs are created using PVCs of type "Block" using Ceph RBD (provisioned through Rook/Ceph-CSI).
If I create a VM with 1G of data inside (but the PVC itself is 100G), the size of the S3 data when using CSI data movement is 1G, which is expected (deduplication...)

The problem arises when restoring. Kopia will restore the 1G of data from the S3 into a 100G PVC. I expect the actual copied data to be equal to 1G. But PVC inflates until it reaches its maximum size of 100G.

This is a pretty big problem:

• the data is thin provisionned on the source
• it is sparsified/deduplicated/compressed/whatever using Kopia in the S3 bucket
• it is inflated to its maximum size once restored

How can I deal with this? I used "rbd sparsify" on the Ceph RBD image, and after a few minutes it removed all the "zeroes" copied by Kopia, but that's not a really good workflow.

I tried using the WriteSparseFiles option on the restore CR, but it didn't yield any particular result. It seems like it only works on filesystems, and not block devices.

Any idea? Thanks!

[SkalaNetworks]

Related issue? #9528

[Lyndon-Li]

We are not going to enhance the existing block device backup/restore functionality through the existing file system data mover, the new block data mover brought with #9528 will replace the existing functionality.