On Wednesday 25 April 2007 05:42, Brad Roberts wrote:
> I don't think this is a good idea.  The file changing _is_ a problem.  I
> definitely prefer that amanda not abort the entire backup because of it,
> but it's definitely a sign that the backed up file might not be a valid
> backup.  It's probably only valid if the file is an append only file,
> such as a log.

 Hi

 Well, this is pretty much what always happens if you are backing up some busy 
servers (like I am). Amanda will end up stuck in this situation, complaining 
about strange dumps every day.
 The way I see it, the file changing is not the kind of problem Amanda should 
be worrying about. The tar warning does not mean that there is a problem with 
the filesystem per se, but that the backup for that file might be in an 
inconsistent state.
 It is the administrator that should worry about that, and if you _really_ 
want to ensure strict filesystem consistency during a backup, you should take 
a snapshot of it (for example with LVM or equivalent) and then do a dump from 
it anyways.
 Because if you complain about the "file changed" warning in tar as being a 
problem you should also remember that Amanda dumps are only consistent as 
long as you don't have dependencies between files, and there is no way Tar or 
Amanda can know about that. The real way to deal with that is with snapshots 
or with a server in runlevel 1. That's why I won't dump /oracle 
or /var/lib/postgresql directly, but will make an SQL dump or maybe even stop 
the DB server before dumping.

 So, in most cases, I'd rather much more have a partial dump of the file that 
changed (normally a log file), than having to exclude it from the dump just 
to stop Amanda from complaining and then have no log file at all in case of a 
disaster.

 Anyway, if you don't think this is a good idea you should have complained 
sooner, in November 2006, when this change went into Amanda (present in 
amanda 2.5.1p3):

r130 | martinea | 2006-11-29 12:22:02 +0000 (Wed, 29 Nov 2006) | 2 lines

* client-src/sendbackup.c: Ignore tar exit with status 1.

 So my patch was only finishing what had already been started since, AFAIK, 
one does not make much sense without the other.

 Hope this helped.

 Best regards

Claudio Martins



-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]

Reply via email to