On Mon, Dec 17, 2007 at 08:49:18PM -0800, Joshua Penix wrote:

Also Bacula's incremental depends on files' timestamps, not their ctime, so files that get moved without having their timestamps manually updated won't get picked up during an incremental run.

It's harder than this.  If you rename a directory, the ctime of the files
contained in the directory don't get renamed.  The backup software either
has to detect the directory has been renamed, and backup everything under
it again (GNU tar does this), or detect the rename on restore (dump,
xfsdump, star).  The bugs I find are almost always in this detection (in
all of dump, xfsdump, and star).  There is a danger in doing the most
complicated part of a backup operation during the part that most people
never run (restore).

Given that nothing has replaced freeveracity/veracity, and tripwire and
friends don't quite allow you to check a directory that has been restored
to a different location, I suspect there are very few people who verify
their backups.

I honestly don't consider a backup to be valid unless I can demonstrate a
restore on a machine with a blank HD.

Dave


--
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list

Reply via email to