> My guess is it would run 'sync' once or twice before beginning a
> snapshot, or maybe some more sophisticated equivalent. I think we'd know
> by now if it didn't.
I think we'd know if it did because `sync` can take an awfully long time
and the whole system should stall all new writes during that time: it
would be quite disruptive.
Instead when I read
This facility does require that the snapshot be made at a time when
the data on the logical volume is in a consistent state - the
VFS-lock patch for LVM1 makes sure that some filesystems do this
automatically when a snapshot is created, and many of the
filesystems in the 2.6 kernel do this automatically when a snapshot
is created without patching.
I understand it to mean "just as consistent as a real drive would be at the
time of a power loss". So there's no need to empty the journal or stall
new writes: just make sure the journal entries "up to now" have reached
the underlying device: the result is not necessarily "clean", but it
is consistent.
> Any database being used for an important purpose should be using
> atomic transactions as far as possible, for example.
IIUC the LVM facility described above is indeed meant to allow
preserving the atomicity properties across the LVM layer.
=== Stefan
