[OmniOS-discuss] snapshot and rollback of volume

[Eric Sproul]

On Fri, Jan 11, 2013 at 11:13 AM, Johan Kragsterman
<johan.kragsterman at capvert.se> wrote:
> Is there anybody that knows if there is any differencies between snapping a
> filesystem or a volume? According to oracle admin guide for zfs it seems not
> any diff.

That is correct.  ZFS filesystems and volumes are just different
representations of the underlying object store.

> What I done so far, is to make snapshots of an OS LUN volume, change the OS
> like putting a new file there, and then rollback the snapshot. But after I
> booted the server again, on what I thought was the "old" LUN/volume from the
> snapshot, it shows that the new file is there, so it hasn't been any
> change...

That seems the reverse of what I would expect.  I could understand the
snapshot *not* having a file that was thought to exist at the time of
the snapshot, due to buffering by the host OS, but to *have* a file
that was thought *not* to exist at the time of the snapshot sounds
odd.  Can you illustrate the sequence of events that led you to
conclude there was no apparent change?

In general, as with other SAN systems, getting a consistent snapshot
of the block storage device requires coordination with the client OS
that is using it, to ensure that in-flight state changes are committed
and further changes postponed until the snapshot operation completes.

Eric