I'd suggest doing a bit of reading and understanding of snapshots if you are actually doing this in production.
A snapshot doesn't contain anything really. It just marks a specific point in time and knows what data was on the dataset at that point.
If you send a snapshot, by default it contains the entire contents of the dataset at that point in time. You could restore just that single snapshot to a new pool. This would be a pretty bad way to back up a machine though as you'd be sending the entire dataset every time.
Normally what you would do is the following -
- Take a snapshot of the main pool and send it to the backup machine
If you have 10TB of data, this will send 10TB of data
zfs send dataset@snapshot1
- Take a second snapshot and send just the differences since the first snapshot.
If only 10GB of data has changed, this will only send 10GB of data
zfs send -i snapshot1 dataset@snapshot2
- Continue to do this, always sending the differences to the previous snapshot.
Now, if you're storing all these snapshots somewhere in files, snapshot1 will contain most of the data. snapshot2 will contain just the changes since snapshot1. snapshot3 will contain just the changes since snapshot2, and so on. In order to restore from this you would need to receive from each snapshot in turn. This would be a lot of hassle.
What most people actually do is to send the original snapshot to another ZFS pool. I tend to set up ssh keys between my live and backup system, then effectively just do something like the following to backup -
zfs send -i last_snapshot pool/my/dataset@new_snapshot | ssh backup_host zfs recv backup/my/dataset
If you use ZFS on the backup machine if gives you several benefits -
- You know the data has been sent intact because the recv would fail otherwise.
- ZFS is now protecting the data on the backup system, and provides checksums so you know the backup data is 100% bit for bit correct.
- On the backup system you can not only access the data immediately, but can also look in any snapshot. (Obviously it's up to you to decide how many snapshots you keep on the live/backup system. You just need to make sure you keep the last "sent" snapshot on both systems so you can do the next incremental send
- You can immediately restore the entire dataset from any snapshot by just doing -
zfs send backup/my/dataset@snapshot_i_want_to_restore | wherever_you_want_to_send_to
Personally I would always suggest against storing ZFS backups on anything other than another pool. You don't know the state of the data and restoring is hours or days of hoping you don't get a recv error.
