Clarification on simple, incremental ZFS backup

[Paul Kraus]

On Jul 1, 2015, at 13:10, Brandon J. Wandersee <brandon.wandersee at gmail.com> wrote:

> Sure, but then what happens in the worst-case scenario, where the source pool
> needs to be clobbered and replaced? That's what I understood the purpose
> of replication to be--a means of completely restoring a system that's
> beyond recovery. If I were to create a snapshot and then send it to the
> backup drive, I couldn't later restore a complete filesystem from it,
> could I?  I'd need a complete replica to make that happen, correct?

You need to start with a FULL stream, then you apply incremental streams to it. On the replica you end up with all the snapshots from the source. So the process I follow is:

Take a snapshot
Send full stream -> recv to a designated backup pool (all it contains are copies of other datasets)
Take another snapshot
Send incremental stream of changes between two snapshots -> recv to same dataset
repeat ...

I take hourly snapshots on all my datasets and replicate nightly.

If you need a full recovery you zfs send a FULL stream in the reverse direction (and you can choose which snapshot to send, and even send them all once the first FULL is done using -I).

--
Paul Kraus
paul at kraus-haus.org