Bacula-users

Re: [Bacula-users] [Bacula-devel] feature suggestion: convert old full backups to? "reverse" incremental backups

2009-09-05 10:18:48
Subject: Re: [Bacula-users] [Bacula-devel] feature suggestion: convert old full backups to? "reverse" incremental backups
From: Kern Sibbald <kern AT sibbald DOT com>
To: bacula-devel AT lists.sourceforge DOT net, Gavin McCullagh <gavin.mccullagh AT gcd DOT ie>
Date: Sat, 5 Sep 2009 16:15:49 +0200
On Friday 04 September 2009 00:59:28 Gavin McCullagh wrote:
> Hi,
>
> On Thu, 03 Sep 2009, Kern Sibbald wrote:
> > I suspect that all that is already possible with Bacula but may require a
> > bit of scripting.  In short, you need support from a Bacula specialist,
> > and we don't do that on this list (bacula-devel).
>
> Fair enough.  I asked for suggested approaches on -users a few weeks ago
> but didn't get much back.  Perhaps we need to put our hands in our pockets
>
> :-)
> :
> > What you are asking for is a bit specialized, so I doubt that it is
> > something that would be directly implemented in Bacula, unless we saw
> > that it was something a lot of users want, which means it would need to
> > be a very clearly defined Feature Request, voted on, then someone would
> > need to volunteer to write it.
>
> I hadn't really imagined it to be specialized (long term snapshots without
> storing large numbers of full backups with lots of redundant files), but I
> guess it must be as there don't seem to be many people popping up who want
> it.  I did get one positive response on the -users list and I know a lot of
> people use the rsync + hard links approach (one drawback of which you
> mention below).  This proposal is the simplest way I can think of (so far)
> to get the same kind of storage efficiency with Bacula's volume storage.
> We mostly use disk storage rather than tapes just now, so perhaps it
> doesn't make sense for tapes.

It sounds to me like the easiest way for you to accomplish what you want is to 
buy a decent tape drive. Tapes are very good for long term storage. If you 
buy a drive that can hold a Full backup of everything on one cassette, you 
have an ideal way of storing data relatively inexpensively long term.

>
> I can write up a draft feature request (perhaps trying to state it more
> clearly) if that helps.  If you guys really don't like the idea you can
> always shoot it down anyway, there's no real need for a vote if it's really
> not good.   Should I use the descriptions in projects.gz as a template?

If you want to submit a Feature Request, you should see: 

www.bacula.org -> Feature Requests

>
> An alternative description which sets out the potential space savings is
> outlined at the end of this email.  Apologies to those for whom this is
> obvious.
>
> > As I say, the other solution is to ask on the bacula-users list or ask
> > for professional help.
> >
> > Thanks for using Bacula.
>
> Many more thanks for creating it, we're very appreciative :-)
>
> > PS: Be careful with programs like BackupPC that create a lot of hard
> > links. If you create enough of them, you will not be able to boot your
> > system because fsck must keep track of all the hard linked files in
> > memory, so if you have too many hard links and not enough memory, you may
> > one day be stuck. The same problem exists with restoring hard links with
> > Bacula.  On a normal system that has a few hundred or even a hundred
> > thousand, no problem, but if you start getting millions of them you may
> > be in trouble.
>
> That's good to be aware of, many thanks.  I know that attempting a "du" on
> that tree of the filesystem is something that takes an unreasonably long
> time right now, which may be an indicator that I need to look at this.

Yes, that would worry me a lot.

>
> Gavin
>
>
> If you suppose for the sake of argument that x Bytes of new data get
> permanently added to the data set each month, that they don't get removed
> (I'd guess this is pretty common on email servers) and that you keep
> indefinite monthly snapshots (again, lots of people like long memories on
> email servers), the total data consumed by the backups increases as
> follows:
>
>  total_data(month1) = x
>  total_data(month2) = 2x + x
>  total_data(month3) = 3x + 2x + x
>  total_data(month4) = 4x + 3x + 2x + x

I don't think you have looked into the possibility of doing separate "long 
term" backups to tape.  If you do it correctly, I am not convinced that the 
growth in data is as bad as you say.  If you do a Full once a year then 
incrementals once a month, I think you will have a reasonable solution.

>
> Using the system I've outlined, I think the required storage scales
> linearly.  Using monthly full backups each time, you get an arithmetic
> progression which gets progressively worse.  After 12 months, the
> arithmetic progression is at 78x, where the linear is at 12x.

I have never understood your "system".  From what I see you are proposing to 
make incrementals back in time -- that is something that I do not think is 
possible and with the information you supplied, I could not imagine any 
practical way to implement a working solution.  It could be that I am missing 
something.  

I would suggest that you consider implementing a solution that does a 
special "archival" or "long term" storage backup as I noted above. The total 
incremental data it backs up over one year is not really excessive, and lends 
itself to a tape or disk based solution.

>
> Obviously that's a very simplified picture, but I think it reasonably
> represents our situation.
>
Regards,

Kern

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with 
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
_______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users