Re: raid1 issue, somewhat related to recent "debian on big machines"

[Goswin von Brederlow <goswin-v-b@web.de>]

Francesco Pietra <chiendarret@gmail.com> writes:

The important bit is:

> Disk failure on sda1, disabling device

So one of your disks failed. The kernel marked it as such and
continious on the other disk alone. This is invisible to the
application (apart from the short hickup) and no data is corrupted or
lost. That is why you use raid1 after all.

> Personalities : [raid1]
> md6 : active raid1 sda8[2](F) sdb8[1]
>       102341952 blocks [2/1] [_U]
>
> md5 : active raid1 sda7[2](F) sdb7[1]
>       1951744 blocks [2/1] [_U]
>
> md4 : active raid1 sda6[2](F) sdb6[1]
>       2931712 blocks [2/1] [_U]
>
> md3 : active raid1 sda5[2](F) sdb5[1]
>       14651136 blocks [2/1] [_U]
>
> md1 : active raid1 sda2[2](F) sdb2[1]
>       6835584 blocks [2/1] [_U]
>
> md0 : active raid1 sda1[2](F) sdb1[1]
>       2931712 blocks [2/1] [_U]
>
> md2 : active raid1 sda3[2](F) sdb3[1]
>       14651200 blocks [2/1] [_U]

That is a lot of raids. Have you ever thought about using LVM? The
different raid1 will mess up each others assumption about the head
positioning of the component devices. On read the linux kernel tries
to use the disk with the shorter seek and assumes the head is where it
left it on the last access. But if one of the other raids used that
disk the head will be way off.

I would suggest the following scheme:

sda1 / sdb1 : 100Mb raid1 for /boot (or 1GB for / + /boot)
sda2 / sdb2 : rest raid1 with lvm

MfG
        Goswin