Bruno,
Specifically what I ran into with a Fedora Core 2 system was
the following. The machine was delivered with two SATA drives
attach to a promise raid controller that was being run in Ultra mode.
I set up /boot, /, swap, /var and /home as software RAID partitions
that were mirrored. Over a few months, I noticed that sometimes when
I rebooted the drive providing the boot partition would change.
During one of these episodes the machine decided that the two drives
weren't in sync any longer. While trying to resync them I discovered
that the drive, which was being sync'd from, had developed bad blocks
and could never complete the resynchronization. Fortunately I was
able to obtain instructions for removing the drive with the bad
blocks from the RAID-1 set and marking the second drive as being
properly synchronized so that I could add a second drive and
sync to that. Luckily, I didn't have any data loss.
What worries me is if such a scenario would become even more
complicated if LVM were involved. FYI, it seems to me that
bad blocks are the weak point of RAID-1. It can cope fine with
drive failure but can become quite confused if one of the two
drives develops bad blocks.
Jack
On Fri, Feb 23, 2007 at 12:32:20PM -0600, Bruno Wolff III wrote:
>
> As long as you can get the mirrors back in sync, LVM isn't going to be an
> obstacle. If you can't, then it is going to complicate things. Though at
> that point you are likely going to your backups anyway and it may not
> matter that much.
