Excellent discussion Jon!  Thank you so very much for your thoughts!!

I was aware of Amanda's philosophy of "I want one full dump every dumpcycle
which is n days" and then to let Amanda find the optimal combination of
full/incrementals, and that's one of the reasons that Amanda was so
attractive, other than it uses native compression tools (BIG plus for me).

Heck, I even learned about the Tower of Hanoi scheme!  But one piece of
advice stuck with me, and that is "when restoring data, you don't have time
to try to remember/decipher which files are on which tapes". In other
words, backup / on a regular schedule, even if it'll cost you some disk
space/time....it'll be worth it when it comes time to recover.  I suppose I
can make things more complex when I get really bored. ;)

As I'm working out the backup plan, it's starting to seem that I'll want to
have a dumpcycle of 7 days, and then keep those weekly full dumps
elsewhere.  At this time, we don't have a massive quantity of critical
data, and we do have enough backup storage to keep 52 full dumps/year.  One
the reasons that I'm going to keep so many is because of the way one of our
database has been (poorly) designed. There have been questions asked such
as:

"On which day did this record change?"

Unfortunately, even with the binlog enabled, some of the queries don't get
recorded (long story), so at least for the database config, it seems like
we'll be keeping many full dumps.


Another part of my backup plan is going to have two configs.  One for the
database instance ('hypercritical data') that will be backed up insanely
often, and another config for merely really important data (svn repos,
wikis, etc) which will have a less taxing schedule.

Now I suppose that I could combine these two schedules into one config, but
it seems that it would make more sense from an administrative view to have
two (more?) configs.

Do you/anyone else have any thoughts on that?



--
Andrius


On Sat, Jan 11, 2014 at 9:10 PM, Jon LaBadie <[email protected]> wrote:

> On Sat, Jan 11, 2014 at 12:31:52PM -0800, Andrius D. Ilgunas wrote:
> > Hmm...Jon's question got me to a'thnkin' about something that I haven't
> yet
> > read about.
> >
> > So let's say the dump cycle is 30 days and I want to keep a copy of the
> > previous month's dump before the tape1 gets overwritten so that I can
> keep
> > a record of the state of the database from two months ago.
> >
> > If I understand the terminology, this would be an archive?
>
> Andrius,
>
> You might want to read further about the way Amanda differs from most
> backup systems.  Typically they act as you are describing, a full dump
> of everything once in a while and incrementals in-between.  So you get
> a monster dump followed by a bunch of small dumps.
>
> Amanda tries to level out the size of the dumps.  You define objects to
> be backed up, called DiskList Entries (DLEs) in Amanda, and how you want
> them backed up (they need not all be the same).  Amanda then spreads the
> full back ups of the DLEs across the whole dump cycle.  At the beginning
> of each dump run Amanda spends time determining how big a dump it expects
> from each DLE as several levels, at least two, level 0 full and level 1
> incremental.  Then it determines how best to achieve balanced dumping
> while still keeping to your defined parameters.  For any specific DLE it
> may continue to do a level 1 incremental, switch it to a smaller level 2
> incremental, or promote it to an early level 0 full dump.
> >
> > So a strategy might be: on the 30th day, do a full dump, and copy it to
> > 'somewhere safe', whether on S3, or a DVD.    That seems like a pretty
> > straight-forward concept, but might this idea be improved upon somehow?
>
> A strong recommendation is to never have less than two full dumpcycles.
> I suspect most Amanda installations maintain well over 2 dumpcycles.  If
> you are reusing your storage media, and you only have one level 0,
> should it be overwritten by mistake or something bad happen to it, you
> only have less useful incrementals.
>
> Jon
> --
> Jon H. LaBadie                 [email protected]
>  11226 South Shore Rd.          (703) 787-0688 (H)
>  Reston, VA  20190              (609) 477-8330 (C)
>

Reply via email to