Are volume snapshot backups supposed to be moved to s3?

[gmerigo]

I've set up oadp by the book, currently I'm testing on OCP4.6 on AWS, using gp2-csi volumes.
I tried manually and the snapshots are ok.

Here is the yaml file (I uploaded in .txt otherwise Github doesn't let me upload it)

konveyor.openshift.io_v1alpha1_velero_cr.yaml.txt

I'm looking into the s3 and I don't see any snapshot content (I expect the snapshot content for a new volume to be around 200MB as it contains a default mysql persistent volume witch get initialized from an empty one).

My backup is as follows:

apiVersion: velero.io/v1
kind: Backup
metadata:
  namespace: oadp-operator
  name: backuptestdb
spec: 
  includedNamespaces:
    - backuptest
  storageLocation: default
  defaultVolumesToRestic: false
  ttl: 720h0m0s

Note that the restore works, but in my opinion only because the volumesnapshotclass is set to Retain, so the snapshot is retained on the volume storage even after I delete the backuptest project and it's restored from there. I set the class to Retain because of a velero warning that told me so.

Is there any way to move the volume backup to s3 and having the volumesnapshot class set to Delete so the restore happens directly from “offline” media such as the s3 bucket and not from the snapshot? I'm feeling I'm missing some important information.

I would rather not use restic as it has high resource requirements when volumes get large since we can leverage EBS snapshots.

[dymurray]

@gmerigo Good question. What you are describing is a concept of a "data mover" and is not explicitly handled by Velero at this point in time. You did everything correctly and the Retain issue is well documented in Velero.

Velero offloads the handling of snapshots to the volume provider itself and tries to remain agnostic from it. OADP intends to enhance this experience by supplying a data mover tool that can copy snapshots into a common consumable format that can be stored in some off-cluster storage. I believe this is what you would be looking for in this use case.

[dymurray]

We wrote up a design for this awhile back here: https://github.com/konveyor/data-mover/blob/master/docs/design/initial-design.md but we are also investigating other tools such as Scribe to see if we can leverage them to handle this for us.

You may find that Scribe does what you need it to do right now... but ideally you want something that handles this automatically for you at snapshot time.

[dymurray]

I am closing this issue for now as the short answer is no, but hopefully in the near future it will be supported.