Once again about incremental zfs backups -- do we have them? Or can we have them? ))

Dear Devs (@stgraber and @brauner , and everyone I missed) )) –

– to follow up on 3073 (Future incremental copy - (lxc refresh) - ZFS backend), let me check the present state of availability of incremental backups for lxc/lxd :

  • 3073 – the most popular post that I can find on linuxcontainers.org that we can consider a feature request )
    • 3225 – one another, probably there are more

Now lxc copy --refresh apparently uses rsync only :

This could be noticeably inefficient compared to a zfs send | recv – and as far as I know, there is at least one successful attempt to implement this feature in the userland:

To add, I have also devised a similar solution based on Florian’s work (@usrflo ) and a look at lxd sql global .dump . This procedure makes lxd aware of any new zfs snapshots that were added on a particular container:

In essence, it compares two lists – (1) current zfs snapshots of a container and (2) the list of snapshots as known by lxd and stored in instances_snapshots, and then tries to recreate missing records using some lxd sql magic.

It “works for me” ™, but I am concerned that with lxd now following a rolling-release scheme via snap (another issue that I shall mention separately), this lxd sql trickery can break any moment you guys decide to make any improvements to the sql .schema )

So I wonder if there is a way to have incremental zfs sync being less fragile )

I’m not sure – may be we already have something like this :

Otherwise, I wonder how hard it could be to add an additional feature to extend --optimized-storage to do incremental backups, so that they could be then incrementally restored on a container that was imported earlier?

For the reference:

If you think that this is feasible, I can draft a feature request on Github – I am thinking of something like --incremental-since $snapname or --incremental-since $start-date for lxc export and then making lxc import to be smart enough to detect an incremental update (may be requiring an --incremental flag from the user so there’s no misunderstanding).

Upd. Adding @my5t3ry to the list of tagged users; reason: a related discussion at github issue 7902.

3 Likes

Added a quick comment regarding the gist I have referenced in the post:

1 Like