--On Tuesday, February 15, 2005 17:44:32 +1100 Jamie Wilkinson <jaq AT 
spacepants DOT org> wrote:

> We've got one problem machine that has about 90GB in one DLE, and our tapes
> are 100GB LTO1s.
> 
> Of course, once a cycle, the server decides it can only write maybe 5 DLES
> out of 150+ to the tape, because one of them is 80GB+, and so all the other
> dumps end up being reported as
> 
>   FAILED [dumps way too big, 1 KB, must skip incremental dumps]
> 
> which afaict means that amanda doesn't even attempt to back up this DLE
> because it knows it's not going to fit it on the tape.
> 
> But I've got a ton of free space in the holding disk.  What I want to know
> is, is there an option to tell the server to do the dumps anyway and hold
> them, so later I can come along with another tape and flush them?

If you have plenty of holdingdisk space, I would think they would go there.
Even with the default reserve of 100%, you should still get the incremental.
You might have to look in the debug files to see why they were completely
skipped.

> 
> In general, we only need one LTO1 a day, this particular DLE is a freak
> statistic well outside the standard deviations of the other DLE sizes, so
> doing a flush once every cycle is more appealing than allocating two tapes
> per run.

Just because you set runtapes > 1 doesn't mean Amanda will use them all.
I keep runtapes set to a number higher than what normally gets used per
run, and have autoflush set true.  That way even if I have a tape error
and it all goes to disk, the next run can flush it all and do the backups
as well without running out of tape.
  If you normally only need one tape, then that's what it will normally
use, even if you have runtapes set to 2 or 3 or whatever.  The higher
number just allows it to use additional tape(s) if needed.

> 
> At this stage, splitting this DLE into smaller chunks is also not an option,
> in case you were wondering.

That would have been the next suggestion.

Frank