Re: raidframe query

[David Brownlee <abs%netbsd.org@localhost>]

On 21 September 2011 11:39, Patrick Welche <prlw1%cam.ac.uk@localhost> wrote:
> I just tried setting up a raid 1 mirror on two "identical" disks.
>
> raidctl -v -C raid0.conf raid0 (BTW the -v doesn't seem to be verbose anymore)
> raidctl -I 2134123413 raid0
> raidctl -iv raid0
>
> created a raid0 where /dev/wd0a is happy but /dev/wd1a has failed.
>
> SeaTools thinks that these seagate barracuda IDE disks are OK.
> atactl wd1 smart status
> thinks that wd1 has 82 uncorrectable errors whereas wd0 has none.
> Would that be why wd1a is marked as failed?
>
> Given that I'm starting from scratch, and the disks are empty, is there
> any way of making a happy raid set?
>
> BTW if the mirror stripe size is 128 blocks, that means that all the
> partitions sizes of raid0 should be multiples of 128? It doesn't matter
> what the sizes/offsets of wd0/wd1 are?

That raidframe stripe value is "sectors per stripe unit", which is per
device, so for two devices your effective stripe size is going to be
twice that. Your filesystem block size should be equal to or a
multiple of the total stripe size, but currently no more than 64K.

As a general rule try to keep the offsets a multiple of the stripe
size, at at least 4K (to avoid issues on 4K sector drives).

You should be able to raidctl -R the failed unit to try to tell it to
rebuild, though I would be very suspicious of that error rate.

You may also want to enable smart via atactl, and check
/var/log/messages for disk error messages.