What constitutes a backup, was:F21 partitioning circus

[Chris Murphy]

On Mon, Feb 23, 2015 at 4:06 PM, Andrew R Paterson
<andy.paterson at ntlworld.com> wrote:
> Hang on there Chris, (new thread really)
> why do you think using a mirror as a backup is a bad idea?

I'm not saying it's a bad idea. I'm just denying it's a backup. What
you have is a degraded array on the shelf that's at best an incidental
archive because it cannot be kept up to date. Once it's replaced, it
can't be readded to the working array, and thus merely caught up.
Plus, by default there isn't an internal bitmap. So what you're doing
is either not keeping it updated, and thus not a backup; or it's being
wiped out and rebuilt each time, in which case it's a slow monolithic
backup that has no archive.

This means any kind of corruption, filesystem or data, silent or
otherwise, eventually replicates itself into all copies.

So no, it's not a backup and it's not even a particularly good archive
if it's subject to rotation.

RAID1+ is not ever a backup, it is about improving availability
(uptime) by mitigating a particular kind of device failure. Just
because you can willfully instigate a faux-failure and shelve that
actually OK member and call it a backup doesn't mean it's a backup.


> As opposed to taking your box offline, and doing a level 0 backup to another
> disk - you end up with a serial backup which must be parsed - I end up with a
> filesystem that I can mount?

Umm, why is an offline backup the only alternative? rsync can do
online archive updates with -au.


> To me this is one of the benefits of mirroring - I can mount one of my old
> detached mirrors somewhere else and get at my old data.

You can ro mount any rsync created backup and get at your old data
also while having a low probability of compromising it. You can make
this differential or incremental so they're fast. You can also
optionally use checksum verification at the source and destination
which would tend to expose silent data corruption at least in the data
itself.


> That's aside from the lower risk of losing the data in the first place.
> I don't particularily need to archive data - just preserve it.

What's the distinction?

> I think you will find this idea is becoming more common these days.

Yes so is skin cancer, what's your point?


> So please give some good reasons for archive (backup) better than checkpoint
> (detached mirror)??

For one, achive≠backup≠raid. For two, the backup needs to be on its
own filesystem, and on a separate device. And this is an example of
neither, because the fs is the same, and the logical block device is
actually the same too.


-- 
Chris Murphy