zfs discuss - May 2009 - ZFS + USB media + mirror/raidz = weird

Hello, folks.
kind of problem with mirrors and raidz.

System & config:
? SunOS 5.11, snv_111a
? Service system/filesystem/rmvolmgr is disabled.

Hardware:
? Asus EeePC Box B202
? Two USB 3.5" inches boxes.

If I reboot: mirror or raidz works fine. But if I connect physical USB
cables in different USB port, then I get my mirror (or raidz)
completely screwed up. If I just shutdown the box, restore physical
configuration ? everything fubar and no errors reported.

I took an output of zdb before and after cable swap, so that diff
shows me only different txg on rpool (not related to my own pool of
USB drives). Everything else is just identical. When zpool with USB
drives are corrupted due to cables swap, "zdb -uuu <pool-name>
<ENTER>" won''t report anything, yelling that there are no
such pool at
all (although just "zdb <ENTER>" reports it exists).

I understand, that there is a lot of steps, before ZFS see the drives. But...
how come that USB cable swapping can affect ZFS corrupted to be like this?
It somewhat does not sounds right to me in general: I assume ZFS must
find disks by a labels, right?

Anyone help me to understand what''s going on here, please?
If here is any hack/cure/idea/wish or at least necrology, please
share as well. :-)

Thanks a lot.

--
Kind regards, bm

Bogdan M. Maryniuk wrote:
> Hello, folks.
> kind of problem with mirrors and raidz.
>
> System & config:
> ? SunOS 5.11, snv_111a
> ? Service system/filesystem/rmvolmgr is disabled.
>
> Hardware:
> ? Asus EeePC Box B202
> ? Two USB 3.5" inches boxes.
>
> If I reboot: mirror or raidz works fine. But if I connect physical USB
> cables in different USB port, then I get my mirror (or raidz)
> completely screwed up. If I just shutdown the box, restore physical
> configuration ? everything fubar and no errors reported.
>   
The current imported configuration is stored in the /etc/zfs/zpool.cache
file.  You can take a look at it using "zdb -C"  Upon reboot, the
imported
pools (and devices) shown in the zpool.cache will be automatically
imported.  Others won''t.

The solution is to clear the pool entries from the zpool.cache.  This is
easily accomplished by doing "zpool export poolname" even if the
pool is not imported.  If you know you will be juggling devices around,
you can also export prior to reboot.
 -- richard

> I took an output of zdb before and after cable swap, so that diff
> shows me only different txg on rpool (not related to my own pool of
> USB drives). Everything else is just identical. When zpool with USB
> drives are corrupted due to cables swap, "zdb -uuu <pool-name>
> <ENTER>" won''t report anything, yelling that there are
no such pool at
> all (although just "zdb <ENTER>" reports it exists).
>
> I understand, that there is a lot of steps, before ZFS see the drives.
But...
> how come that USB cable swapping can affect ZFS corrupted to be like this?
> It somewhat does not sounds right to me in general: I assume ZFS must
> find disks by a labels, right?
>
> Anyone help me to understand what''s going on here, please?
> If here is any hack/cure/idea/wish or at least necrology, please
> share as well. :-)
>
> Thanks a lot.
>
> --
> Kind regards, bm
> _______________________________________________
> zfs-discuss mailing list
> zfs-discuss at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
>