On Mon, Aug 15, 2022 at 3:20 PM J. Roeleveld <jo...@antarean.org> wrote:
>
> Actually, you can with the "-p / --pause" option.
> Also, as per the man-page, if you forget this, the process will simply inform
> you the target location is full and you can move slices away to a different
> location:
> "
> If the destination filesystem is too small to contain all the slices of the
> backup, the -p option (pausing before starting new slices) might be of
> interest. Else, in the case the filesystem is full, dar will suspend the
> operation, asking for the user to  make  free  space, then  continue its
> operation. To make free space, the only thing you cannot do is to touch the
> slice being written.
> "
>
> The pause-option will actually stop between slices and you can umount the
> target location and mount a different disk there.
>
> This option has been around for a while.

Hmm, sounds kind of non-ideal.

It sounds like you can either have it pause when full, or pause
between slices.  Neither is great.

If it pauses when full, and you can't touch the slice being written,
then you can't unmount the drive it is being written to.  So you end
up having to write to a scratch area and keep moving slices off of
that onto another drive.  At best that is extra IO which slows things
down, and of course you need scratch space.

If you pause between slices, then you have to have drives of equal
size to store to, otherwise you'll have to swap drives that aren't
completely full.

Ideally you'd want to write until the drive is almost full, then
finish a slice and pause.

However, it at least seems workable if slow.  Just have scratch space
for a bunch of slices, let it pause when full, then move slices off as
they are done, and accept that your backups will run at maybe 25% of
the speed of the scratch drive since it will be constantly seeking
between writing new slices and reading old ones.  Or if you have
enough RAM you could use a tmpfs for that but that seems really
cumbersome unless you use very small slices and have the shuffling
scripted.

-- 
Rich

Reply via email to