Linstor datastore Marketplace export requires full virtual image size despite thin provisioning

Integration Support Marketplace Appliances
1

I am using OpenNebula with a LINSTOR-backed image datastore and I am trying to upload/export an image to the OpenNebula Marketplace.

The image is thin-provisioned in the LINSTOR/LVM thinpool. The OpenNebula image is registered as a 600G image, but the actual allocated data in the LVM thinpool is much smaller, around 24G.

Example from oneimage show:

IMAGE ID      : 172
NAME          : example-thin-image
DATASTORE     : LINSTOR image datastore
FORMAT        : qcow2
SIZE          : 600G
STATE         : rdy
SOURCE        : OpenNebula-Image-172

The LINSTOR/LVM thin volume shows something like:

OpenNebula-Image-172_00000  LSize <600G  Data% ~4%

So the actual consumed storage in the thinpool is much lower than the virtual image size.

However, when I try to export/upload the image to the Marketplace, the export appears to create a full-size staging file. The export failed with an error similar to:

RuntimeError: Unable to dd export image file
ERROR: bash: Command "could not dd image" failed

On the datastore bridge host, I found a partial temporary file under /var/tmp that had already grown to several hundred GB before the filesystem filled up:

/var/tmp/OpenNebula-Image-172-xxx  ~398G

The bridge host OS filesystem then reached 100% usage, causing the Marketplace export to fail.

My understanding is that the LINSTOR datastore export path is effectively doing something like:

LINSTOR/DRBD block device -> dd -> regular staging file -> Marketplace upload

Because of that, the export file grows toward the full virtual block device size, even though the backing image is thin-provisioned.

My questions are:

  1. Is this expected behavior for Marketplace export from a LINSTOR image datastore?

  2. Does the Marketplace upload/export process require staging space equal to the full virtual image size?

  3. Is there a supported way to make the export sparse-aware, for example using dd conv=sparse, or to export as a compressed/sparse qcow2 instead?

  4. Is there a recommended workflow for uploading thin-provisioned LINSTOR images to the Marketplace without consuming the full virtual image size in staging storage and object storage?

  5. Would the recommended approach be to keep/use the original qcow2 file and upload that to the Marketplace instead of exporting from the LINSTOR block device?

In short: the image is thin-provisioned and only consumes a small amount of physical storage in LINSTOR, but Marketplace export appears to require the full virtual size as staging space. I would like to know whether this is expected and what the recommended best practice is.

Thanks.