[Gllug] [OT] Appreciation

Nix nix at esperi.org.uk
Wed Dec 20 23:33:19 UTC 2006


On 20 Dec 2006, Bruce Richardson stated:
> On Sun, Dec 17, 2006 at 01:12:05AM +0000, Nix wrote:
>> Likewise. But an even more important thing to keep printed out is a
>> bunch of magic numbers of important filesystems. That way, if a machine
>> goes tits-up and you don't have access to gpart or its partition table,
>> you can mock up something that finds magic numbers: from that, you
>> can determine the *start* of every partition, and then you're home free.
>> (If you have filesystem images in files on some of those partitions,
>> things can get a little more interesting, of course: but that's
>> quite rare.)
>
> It's more important never to be in a position where this kind of magic
> is necessary.  If you are responsible for crucial business data (or
> specially configured systems in a key role) and you find yourself
> resorting to this level to get it back, your employers have reason to
> question your suitability for the job.

The lady who cooked my meals would have lost her data. Ergo, this was
critical. (And when I pulled her eggs out of the fire, she was suitably
amazed. And, hey, it was exhilarating, and it wasn't *my* data so I
didn't actually have to panic.)

(Backups? We didn't have money for *backups*. We just barely had money
for *rent*. We backed up by replication onto a mass of dysfunctional
machines... and then three went down in different ways at once. Of
course it also turned out that some critical data wasn't replicated...
because I didn't know it was critical, or even that it existed.)

In my experience, critical business data is replicated using what might
be called a single-disk RAID array, backed up with scripts whose
standard error is redirected to /dev/null, and periodically smashed by
buggy proprietary software. (My own home systems are *far* more
redundant and robust than, say, work's primary source tree, or most of
our clients' oh-so-critical databases, and it's not as if I spent that
much time on getting it that way either. They have one backup, if you're
lucky, while I have staged backups up to four years ago (at declining
levels of temporal resolution), remote and replicated storage of
critical state, and a RAID array... of course everyone here proably
agrees that this is the minimal degree of security required for peace
of mind. Alas that my various workplaces have never agreed.)

>                                         Whether they are work systems or
> home machines, you are far, *far* more likely to see general disk
> failures or data corruption than problems with a partition table.  Cue

These days, yes. In the days when Disk Manager and its friends were
around, no. Those were *nasty* pieces of software.

> Martin's standard rant about backups, because good backups protect you
> from all three scenarios whereas printing off your parition table might
> rescue you from one.

I prefer defence in depth and not having to do a huge restore when ten
minutes of typing will do the job just as well :)

-- 
`He accused the FSF of being "something of a hypocrit", which
 shows that he neither understands hypocrisy nor can spell.'
   --- jimmybgood
-------------- next part --------------
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug


More information about the GLLUG mailing list