[Question/Feature Request] Is there a way to send incremental snapshots between two nodes with zfs-local-*?

[pando85]

I checked the code and I think that it is possible (not tested yet) to send snapshots from one node to another with the Clone implementation for CSI.

Would it be possible to keep in sync does snapshots with the source?

I guess that we will need custom CRDs for this.

My use case would be to have backups in a different ZFS pool or even move loads from one node to another transparently.

[travisghansen]

I’ve considered something along these lines but never drew up a proposal. Happy to review some ideas of you send them over.

Alternatively I have considered a setup using something like syncthing (not with zfs, but the local-hostpath) to keep data across nodes in sync and allow workloads to float around.

[pando85]

It is difficult for me because I don't know the interns about distributed provisioning in K8s. It would be great if we could do it in a way more integrated with the CSI.

My first approach is to do something like: https://github.com/backube/volsync . Having a ReplicationSource for sending K8s snapshots to the target node. And then, use ReplicationDestination to create a volume in that node based on the ReplicationSource used previously.

This like the first idea that comes to my mind. I want to be as integrated as it can with K8s CSI but I think that we don't have control for sending snapshots to other node or selecting in which node do you want a clone.

Also, this could be integrated here or be something independent. Again, my first thought is having it here because it is going to be coupled.

[travisghansen]

Are you on k8s slack or discord or similar to have a more thorough conversation about the topic? You can email me details if you wish and I can reach out.

[pando85]

I tried to contact you in those channels and I couldn't.
What does it means closed as completed?

[travisghansen]

Sure! Just doing some spring cleaning of issues etc. Shout out on discord or whatever and we can further discuss.