To sum up the problems I have had:

I have now a properly working amanda working with my pictures
directory divided over 5 disks defined in the disklist, that all resides
on a Windows formatted separate disk.

I have done a number of tests with all kind of variations on my vfat
disk, but I couldn't get a handle on what or where the problem is.
Level 1 backups are totally wrong, because they are in all but 1 case
either full backups or very sizeable backups while they should be near
0.  And then they were also different in size (that is unpredictable) in
a next level 1 run and all this on a directory tree in which no changes
were made between the test runs. The culprit seems to be the vfat disk
which holds the directory, because when I copy the pictures directory to
a linux ext4 partition, everything works just fine, including proper
level 1 dumps.

As variations I have done (all with a vtape test setup):

- running on a read-only mounted directory and actually checked
  date/time settings with a find (for recently changed files) after a
  level 1 amdump: failed although there was no change in file times.
- running with program "GNUTAR" with an empty gnutar-list-dir
  definition or with one: failure
- tested each defined disk separately: failure for each with
  consistently 1 exception: de Pictures_M10 behaved as should.
- level 1 for Pictures_M10 backups are ok, both for a no changed
  directory or with some test files added.  I could not find any reason
  why this disk was different from the others.
- However when I amdumped another disk together with the Pictures_M10
  disk, both failed.
- actual precise amdump results can vary for the same disks over time.
- tested with tar 1.27 and tar 1.30: all fail
- with or without the property "CHECK-DEVICE" "NO" in amgtar: no
and then I have probably forgotten some variations.

It seems that amanda (or probably tar?) does not produce the correct
information to perform a level 1 backup. I have looked to find the
details about what type of info amanda (time info of files and/or
directories, size, etc?) is using to create the tar info, which I
suppose goes into the files in the gnutar-lists directory. These are
binary files, which makes investigation difficult (for me).  So
I have given up on the vfat format. But as it is claimed that amanda
also runs on Windows, how does it works in Windows on vfat format?

Anyhow, I then decided to reformat my picture disk to ntfs and see what
that would deliver.  In a first test, level 1 was a near 0 size dump, so
everything seemed ok.  However when I added a few large files, level 1
remained 0.  It seemed an inverse problem.  But then I noticed that only
accessing a file did not only change atime, but also ctime equally.  I
remounted the picture disk with a noatime option added to the mount and
this now solved all problems.  In hindsight, I should have tested that
option as well with the vfat disk, but when I tested file times after
an amdump on vfat , I noticed no time changes at all. So noatime
on vfat would probably not have solved the problems.

The ntfs disk is mounted as:
/dev/sdc1  /home/charles/pictures       ntfs-3g  \
    defaults,uid=1000,gid=100,umask=0022,nofail,noatime,rw  0 0

Cheers, Charles

Charles Stroom
email: charles at (remove the "no-spam.")

Reply via email to