On Wed, Sep 14, 2005 at 10:30:16AM -0400, Guy Dallaire wrote:
> 2005/9/14, Jon LaBadie <jon AT jgcomp DOT com>:
> > 
> > I'm beginning to wonder if the reporting is faulty.  I.e. the actual
> > reason some dumpers is idle is not lack of holding space, but some
> > idle time is mistakenly assigned to the no-hold category.
> > 
> > I looked briefly at the source to driver.c.  That is the only place
> > I found that the string "no-hold" is used.  It seems to me that a
> > variable holding the reason for an idle dumper is not reset in a
> > loop in driver.c.  Thus, if some time there is a "no-hold" reason,
> > perhaps the reason stays at "no-hold" for other reasons.
> > 
> > There might be minor support for my theory if you have one or more
> > DLEs that specifically invoke the 'holdingdisk no' parameter.  For
> > example, when the partition that contains the holdingdisk is backed
> > up, often the use of the holdingdisk is prevented by the DLE.
> > 
> > I'll post a not to the hackers list describing my theory.  Maybe
> > someone who understands the code can confirm or reject it.
> > 
> 
> OK Thanks. I do have a single DLE that has the "holding disk no"
> parameter, and that is my holding disk. I think you MUST specify
> no-hold if you are backing up the holding disk.

Yeah, there could be thrashing of the disk, and I can envision a 
situation that used to exist in old version of cat where it could
continuously add to the end of a file:

        cat somefile >> somefile

That situation is specifically tested for in cat now.

A test whether that is the reason is to eliminate that parameter.
The simplest is to just comment out that DLE in the disklist for
a night and see the effect.  But maybe other data on that DLE is
too important for that.

An alternative, if you use gnutar rather than dump, is to exclude
the directory containing the holding disks space.  I have my
holding disk spread over 4 drives.  I named each consistantly
as dumps/amanda, relative to the filesystems root.  Thus in my
exclude file I simply have "./dumps/amanda" listed and all 4
holding disk spaces are skipped.  Perhaps you could comment
out the 'holdingdisk no' parameter and replace it with an
'exclude file append <relative path to holding disk>' for
a night and see the effect.  (check the syntax first)

jl
-- 
Jon H. LaBadie                  jon AT jgcomp DOT com
 JG Computing
 4455 Province Line Road        (609) 252-0159
 Princeton, NJ  08540-4322      (609) 683-7220 (fax)