Is zfs_autobackup right for me?

[danboid]

I have read the documentation through a couple of times now but it leaves me uncertain about how zfs_autobackup prefers / handles the creation of snapshots?

It sounds like it can (must?) be used without using another program such as zfs-auto-snapshot or sanoid to create automated snapshots but what if the user wants to use z-a-s (read: I am already using and trust z-a-s) on the source machine instead of zfs_autobackup creating its own snapshots first every time it is run? I can't see an option to only transfer existing snapshots?

I want to use zfs-auto-snaphot on my source machine to create the snapshots but in the zfs_autobackup docs it seems to imply that zfs_autobackup will create a new snapshot (in every labelled dataset?) every time is run in a non-test mode? Can this be disabled so that it never creates new snapshots on the source?

I don't want zfs_autobackup to create any snapshots at all on the source machine, I just want it to transfer all of the latest snapshots from all of the datasets on the source machine onto the destination machine and additionally I want it to thin out all snapshots in all datasets on the destination machine only so that only keeps 90 days / 12 weeks worth of snapshots in all of the datasets. z-a-s handles the destroying of snapshots on the source side so no need for zfs_autobackup to do any thinning on the source.

I am hoping the thinning works reliably on the destinstion side. sanoid's auto-pruning didn't work at all for me which is why I'm looking into this as an alternative.

Another question is regarding --strip-path=1. The docs for this option say:

Select the whole source-pool. In this case: zfs set autobackup:offsite1=true rpool
Use --strip-path=1
Specify target-pool as target-path. In this case: data
Use the --force option the first time to overwrite the existing target pool. (New in v3.1.2)

I want to recursively replicate all of the datasets in /pool/home on the source into /pool2/home on the destination so it sounds like I need to use --strip-path=1.

Why then should I need to label/select the whole pool on the source if its only /pool/home that I want to recursively replicate, not the whole pool? Do I really need to go to the trouble of setting all of the other datasets in the source pool to false just so I can use --strip-path=1 on one dataset, the one I actually want to replicate without its path.

Finally, I presume that zfs_autobackup can work, on the destination side at least, as non-root users if we delegate the required permissions to the correct user? What are the minimum set of zfs permissions required for it to work on either side?

Thanks for your help!

[psy0rz]

thats a lot of questions, and some of them you have to figure out and --test yourself. :)

you can use --no-snapshot to not create snapshots

you can --keep-source=10000 to not delete snapshots from the source

also the snapshots should have to correct format, or use --snapshot-format. otherwise use --other-snapshot and prune them yourself on the target.

this all sounds like too much of a hassle to be honest..just let zfs-autobackup create its own snapshots to make your life easier. You can still use zfs-auto-snapshot, zfs-autobackup will just ignore those.

[danboid]

Done my second test now it all looks good except...

When I enable verbose mode (-v), it first lists the selected datasets then in the

Snapshotting

Section it says No changes since for EVERY dataset which is bollox because when it gets to

Synchronising

We can see that several datasets did have changes because several receive incremental updates.

It looks like the snapshotting section isn't working or at least isn't reporting the correct status of modified datasets?

[danboid]

My initial 20 questions feet to the fire bit is over with now BTW, I hope you enjoyed it?

Thanks for being prompt with your replies, I will likely tweak your docs soon in return. They are already good actually which is another reason to choose the autobackup.

What do you call it anyway? zab?

[psy0rz]

It skips snapshot creation if there are no changes, but it will still sync snapshots there are already created.

[psy0rz]

I usually call it zfs-autobackup. :)

The questions are fine, whenever possible i try to improve either the program or documentation, to prevent such questions in the future.

[danboid]

I have added the explantion of --no-snapshot to the manual.

[danboid]

If a user is doing using zab for pull backups and didn't want to let zfs-autobackup create any snapshots on the source machine, does zfs-autobackup still work fine for replicating existing snapshots to the target? Obviously zfs-autobackup's source snapshot trimming command becomes useless in such cases.

[danboid]

I'll be integrating your responses into the documentation shortly.

[danboid]

I've finished switching us from syncoid to zfs-autobackup now for our GROUNDBREAKING ZFS LTSP server :D

[viharm]

Great discussion. Thanks @danboid & @psy0rz .