Superblock backup

Superblock backup

[Piergiorgio Sartor]

Hi,

again, following the RAID-5 story, I was wondering
if could make sense to have a new superblock type,
let's say 2.x, having backup copies (of persistent
or quasi persistent data) in different areas of the
drive components.
This would be similar to filesystem or gpt.

Is there anything planned for this?

In any case, it would be still interesting to have
a "backup" capability at hand, like a "dd" of the
superblock to some secure file.

Any idea or suggestions?

Thanks,

bye,

-- 

piergiorgio

Re: Superblock backup

[NeilBrown]

[-- Attachment #1: Type: text/plain, Size: 1269 bytes --]

On Thu, 10 May 2012 19:44:31 +0200 Piergiorgio Sartor
<piergiorgio.sartor@nexgo.de> wrote:

> Hi,
> 
> again, following the RAID-5 story, I was wondering
> if could make sense to have a new superblock type,
> let's say 2.x, having backup copies (of persistent
> or quasi persistent data) in different areas of the
> drive components.
> This would be similar to filesystem or gpt.
> 
> Is there anything planned for this?

This is one of the idea that has floated around in my head for a while,
possibly a v1.3 metadata.

However my thought there was always to protect against a bad block on the
device.  Protecting against data corruption due to bugs or sys-admin mistakes
is always best handled by backups.
e.g. take a 'dump' (either binary or textual) of the metadata every day, or
every reboot or something. (And store it on a different host).

Both are probably useful, but for the recently problem, a regular backup
would be the most useful.

> 
> In any case, it would be still interesting to have
> a "backup" capability at hand, like a "dd" of the
> superblock to some secure file.

"mdadm -E" would probably be enough - just write a cron job :-)

NeilBrown


> 
> Any idea or suggestions?
> 
> Thanks,
> 
> bye,
> 


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]

Re: Superblock backup

[Piergiorgio Sartor]

iHi Neil,

> > In any case, it would be still interesting to have
> > a "backup" capability at hand, like a "dd" of the
> > superblock to some secure file.
> 
> "mdadm -E" would probably be enough - just write a cron job :-)

actually, I was thinking "dd" because a restore
operation could be another "dd" (from file to device).

Does this make sense? Or old superblocks should
never replace newer ones?

I mean, let's say each day the first few MB of
each component device is stored somewhere.
As soon as some irreparable damage happens, the
files can be restored in the proper devices.
Practically rolling back to the previous day
the superblock status.

Would this work?

Thanks,

bye,

-- 

piergiorgio