> Because the vdev tree is calling them 'disk', zfs is attempting to open
> them using disk i/o instead of file i/o.
This was correct, thank you. lofiadm was useful to loopback mount
the image files to provide disk i/o.
> ZFS has much more opportunity to recover from device failure when it has a
> redundant config. Splitting the 9T & 2T LUNs into a few separate LUNs and
> then using raidz would be highly desirable.
Yes, I completely agree. This configuration is something I inherited
from a previous colleague, and did not make good use of ZFS. The
underlying RAID system suffered a catastrophic failure, but even prior
to that, we would consistently have the pools become DEGRADED when ZFS
would report corruption that the underlying RAID system could not
detect and repair.
Thank you to all responders, and also thank you to Drive Savers who
recovered the image for us. If you find yourself in the regrettable
situation of having a dozen terabytes go missing one day without
proper backups, I highly recommend their services.
zfs-discuss mailing list