redundant-servers question

[Lennart Sorensen]

On Tue, Jan 22, 2008 at 03:34:36PM -0500, teddy mills wrote:
> 
> I have servers I need to make redundant.
> Live PC and a backup PC, both exactly the same hardware.
> Servers are Gentoo with RAID1.
> 
> If I have any problems with any of the live servers then I pull the 
> network plug, and drop that network cable into its exact-copy backup server.
> 
> I thought about using rsync, to keep the backup servers the same as the 
> live servers.
> The weak point is the rsync complexity. If I miss even one important 
> file for rsync, then it renders the whole solution almost useless.
> (ie. with 1000% confidence, I should be able to remove the live server 
> at any time, and replace it with the backup.)
> As well, I would have to do a worst-case scenario test..ie..replace a 
> live server with the backup.
> 
> I would much prefer to use mdadm's rebuilding feature and hot-swapping 
> drives. (once per 24 hours, or as needed)
> 
> LIVE PC.
> RAID1 with hotswap 2 drives.DRIVE-A and DRIVE-B
> 
> BACKUP PC with 1 hotswap drive DRIVE-C.
> 
> Everyday, hotswap exchange DRIVE-B and DRIVE-C.
> fdisk /dev/DRIVE-B, and let mdadm rebuild it.
> Leave the working backup in the backup PC.
> 
> Using this method, I am virtually guaranteed that the backup servers are 
> exactly the same as the live servers.
> Only problem with this solution, is that every day I am breaking and 
> rebuilding the raid.
> With no room for errors. A LOT more risky than just letting the RAID1 
> run alone.

Well you do realize raid1 isn't limited to just 2 drives right?

You could leave the 2 disks in the raid alone, and add a 3rd mirror (at
least as far as I know) that you are swapping.

I think I would personally prefer to just make sure my rsync works.  You
could do the rsync, then take the disk out of the backup server and
connect it on the live server and do a diff between the drives to see
that all the files are the same (of course if the live server is doing
work a few files might still be changing).

Replication is a very hard problem to solve.

--
Len Sorensen
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists