there is a usability problem I'm keeping to face with xfsdump:
the lack of being able to exclude directory trees from dumps.
When managing a relatively large file server one may face the
problem of wasted backup space/time/bandwidth. Users used to
keep many non-important files (temporary files, software tarballs,
xxx...) on a fileserver which are not really necessary to back
up, are often large and already compressed thus not further
compressible on the backup media. But an average user cannot
be expected to flag some of his files from a shell (that he
may not even have access to) with SGI_XFSDUMP_SKIP_FILE and
anyway, even for a highly experienced UN*X user this would
be quite a big hassle to do.
I believe the right thing to resolve this situation would
be to split the storage space into two filesystems, one
having the important data that is to be backed up and
another one having the useless stuff that is not backed up.
This way it's even possible to keep a tighter control over
quotas on the "valuable data" filesystem while loosening
it on the "scrap" filesystem. But...
The only practical problem with this solution is that at an
institution or company the storage space is usually sized after
the requirements of the valuable data, so statically allocating
a large space to a "scrap" filesystem may pose a problem later
on when the same space is needed for valuable data. This means
that the administrator practically has to keep the large storage
space in one piece to be able to reallocate the space occupied
by scrap data to valuable data later on. As far as I can remember
only DEC AdvFS can master that elegantly but AdvFS is not among
what we have on Linux to choose from...
So at least with XFS the possible route seems to have a single
large filesystem holding all data and exclude non-important files
from dumps. This would be a lot easier to do if xfsdump would
support flagging a directory with an SGI_XFSDUMP_SKIP_DIR or
similar attribute in order to skip it's contents recursively
without having to flag every single file inside. Would this be
possible to implement in xfsdump?