Gene Heskett said:

<snipped>

> The design decision is because amanda has absolutely no guarantee
> that
> someone will not eject the tape, or otherwise mess with the tapes
> position, from one days run to the next.  So amanda is designed to
> do
> 2 things:
>
> 1: rewind the tape and read the label to make sure its the right
> tape
> in the sequence to use next, doing this with each days invocation.
> This in itself destroys the location you get from an 'mt -f device
> tell' between last nights run and tonights.
>
> 2. Because not all tape drives can reliably do an 'mt -f device
> seof'
> to restore that location, and it would be a total loss to write over
> last nights backup with tonights, the chance of doing that is never
> taken.  Hence, amanda will never attempt to append to a tape.  It
> *must* know that it has the tapes undivided attention during the
> duration of a run.  To that end, if the drive has a door lock
> command
> or ioctl, it is used to prevent you from ejecting the tape in
> mid-run.
>
> Its inconvienient for some, but infinitly safer for all this way.

Couldn't "append to tape" be simulated? That is, rewind the tape and
check the tape label as normal. Then copy all contents of the tape
to the holding disk. Then use amdump to add to the holding disk.
Then flush the holding disk back to tape. The drawback to this
solution is that it would take quite a bit longer. However, there
would be no risk of accidentally overwriting portions of the tape.

I'm just asking in theory; I have no plans to implement anything
like this at my site.

-- 
Kurt Yoder
Sport & Health network administrator