Add support for incremental backups in Ceph

[rsmontero]

Description
This feature is to implement incremental backups in Ceph. The overall process would be:

• Create a snapshot to mark the start of the increment
• Use rbd diff to get the blocks changed since the snapshot was created
• Read blocks and write them to a qcow2 file
• Store the qcow2 increment files in the backup server

Interface Changes
None

Progress Status

• Code committed
• Testing - QA
• Documentation (Release notes - resolved issues, compatibility, known issues)

[nachowork90]

I was doing the same thing and I was thinking in create a driver for Backy or Benji, in order to build a more profesional tools, because the conversion to qcow2 have better portability than the raw image, but dealing with the raw images can speed up the process! Backy/benji simplifies the whole process because they use the xml that represent the rbd diff!

[Franco-Sparrow]

Hi @rsmontero

I think that a better approach for Ceph Incremental Backups could be achieved using known projects like Backy2. Store the increments in qcow2 files will require to convert from raw to qcow2, so a backup uising Ceph will take more time. An rsync host could be prepared with backy2 and sharing its storage with the opennebula cloud. Backy2 driver should be prepared to create, delete, ls and some other actions required for a backup.

This is just my humble opinion, which is aligned with @nachowork90 proposal.

[rsmontero]

Hi @Franco-Sparrow

Agree, in fact the final implementation in 6.10.1 does not stores deltas in qcow2 format but in native rbd diff format. Don't know the internals of other tools but directly implementing this incremental backups on top of the OpenNebula backup framework is aimed to provide the best experience for our user base.

Thanks for your comments