DB Recovery

[Lennart Sorensen]

On Wed, Sep 21, 2005 at 10:30:41PM -0400, Christopher Browne wrote:
> And extracting data from a binary database structure isn't likely to
> turn out terribly well either.
> 
> Databases tend to be "hives" of heavily interlinking systems of
> pointers, very difficult to recover data from in a systematic way. 
> They start "systematic," but break pretty badly :-(.

Yeah getting data off a broken filesystem is easy for jpegs and other
images that you can scan for, as long as they are not fragmented.
Fragmented data you don't want to try and get off it.

> Getting fragments oughn't be too difficult, but doing it
> systematically will be really difficult.
> 
> The one "saving grace" is that you generally have TWO sorts of data:
> 
>  1.  The data itself, which tends to be pretty "dense", where you have
>        files chock full of tuples;
> 
>  2.  Indexes, very likely to be completely destroyed by any kind of damage,
>       which are purely a re-encoding of the real data.  If an index breaks,
>       it is normally fairly easy to regenerate it.
> 
> But really, the only way you ever want to recover this stuff is from a backup.
> 
> Being forced to find what fragments you can reconstruct out of the
> "shrapnel" left after the explosion is a real expensive way to learn
> just how important backups are.

I would prefer to never have to try it. :)

> I'd thought people learned a lot of these lessons out of the
> cautionary results from:
>  a) The ice storm of 1999
>  b) The destruction of data centres that took place 2001-09-11
>  c) The TLUG presentation on how *little* is recoverable from busted
>       disk drives <http://gtalug.org/wiki/Meetings:2003-07>.  Which was a
>       VERY valuable presentation!
> 
> But I guess what we continue to learn from history is that people
> don't learn from history :-(.

What is it?

History always repeats itself
Those that don't learn from history are bound to repeat it?

Lennart Sorensen
--
The Toronto Linux Users Group.      Meetings: http://tlug.ss.org
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://tlug.ss.org/subscribe.shtml