[Bug-tar] question RE incremental backups

[Phil Schumm]

Hi,

I have a couple of questions that I'd be grateful if someone could  
help answer.  As a brief background, we manage several Apple Xserves  
running OS X Server 10.5, which provide services to a relatively small  
group of users based mainly on open source platforms (e.g., Apache,  
Subversion, Trac, TurboGears, etc.).  We have used a backup strategy  
developed around BRU in the past, but are now considering moving to  
GNU tar (for several reasons).  OS X 10.5 comes with version 1.15.1  
pre-installed, patched by Apple to handle HFS extended attributes and  
resource forks.

We have been running several tests using --listed-incremental to  
create (and restore) full (level 0) and differential (level 1)  
backups.  Everything has worked nicely, exactly according to the  
documentation.  The one thing we noticed, however, is that files are  
picked up in the incremental dumps even when only their atime has  
changed -- despite the fact that neither their ctime nor mtime has  
changed (i.e., the file has merely been accessed but not modified in  
any way).  This can make the differential archives pretty large,  
especially in our case since we have several large database files  
(e.g., ZODB files for Zope or SQLite database files) that are accessed  
regularly but changed much less frequently.  My question is: Is this  
the expected behavior, and, if so, is there any way to generate an  
incremental dump capable of restoring the filesystem *except* for the  
atimes that would be smaller in size?

A second (and less important) question concerns the implementation of  
the --exclude-from option when combined with --compare.   
Unfortunately, we are unable to use --verify when creating an archive  
because -- even with the patched version of GNU tar distributed with  
OS X -- the --verify option doesn't appear to handle HFS resource  
forks (i.e., GNU tar tries unsuccessfully to compare the ._* files  
representing the resource forks in the archive with the filesystem).   
For this reason, we have tried using --compare to verify an archive  
after creating it.  However, while this works fine with full (level 0)  
dumps, using it with differential (level 1) dumps requires manually  
excluding those files that haven't changed and are therefore not  
included in the archive; otherwise, GNU tar reports that the contents  
of the directories containing those files differ.  Problem is, when we  
pass in a long list of all the filenames excluded from the archive  
(generating by parsing the output from gnutar --list) via the -- 
exclude-from option, the time required for the comparison blows up.   
This suggests that GNU tar is iterating through all of the files  
listed in the --exclude-from file over and over, rather than storing  
them in a hash table and looking for matches that way.  Is this  
correct, and, if so, is there some other way to use --compare with a  
level 1 (or higher) dump?

Any information or pointers to references on these issues would be  
much appreciated (I have been unsuccessful Googling on them).

Thanks,


-- Phil