[rdiff-backup-users] Techniques for trimming the sizes of backups

[Anthony Toole]

Hi All,

I've been running rdiff-backup successfully on a few servers for
several years now - but noticed recently that the disk space
requirements have got quite large (not a sudden change in disk usage,
just something I looked in to when it became an issue!).

I've got a few questions about how to reduce this - ideally without
using the "--remove-older-than" option over the whole backup, as
having the history has been useful in the past.  Really just after
making point changes to certain files in the backup, without affecting
the integrity of the backup as a whole.

1) How to safely remove a given file from the backup directory?
I've got a given file (Mail.tar.gz) that should never have been backed
up, but was - and because the file was regenerated daily and around
the time the rdiff-backup script ran the file would either appear
different every day or would appear/disappear depending on timings.

Looking in {backup}/rdiff-backup-data/increments/ I see three
different sets of files that reference Mail.tar.gz in their names:
- ~40Gb: rdiff-backup-data/increments/Mail.tar.gz.20*diff
- ~9.5Gb: rdiff-backup-data/increments/Mail.tar.gz.20*snapshot
- ~0: rdiff-backup-data/increments/Mail.tar.gz.20*missing

There are presumably also a bunch of meta-data files that reference Mail.tar.gz.

So, given the fact that I do NOT want this file in the backups at all,
but ideally want to keep all history going back for everything else in
the backup, what are my options for removing this?

Is it safe to just delete all the
rdiff-backup-data/increments/Mail.tar.gz* files?  I don't care that
this will leave some phantom trace of this file, as long as it doesn't
affect the rest of the backups..

2) What is the different between the mirror_metadata* and
file_statistics* files?  Are they both required for restoring for
backups?  file_statistics appears to be largely a subset of the
mirror_metadata information..

I only ask as for a given user they have ~2.5Gb of mirror_metadata
(which seems to contain all the important backup meta-data), but
~6.5Gb of file_statistics* files - so if the latter is not necessary
for restoring from backups then I can save some disk space.

TIA,
Anthony