zfs discuss - Oct 2008 - ZFS Pool can''t be imported on fresh install with one broken vdev

Hi,

I recently had to reinstall OpenSolaris on my home server, after I managed to
break the install while fiddling around with the new package system.
Anyway, the day I wanted to do the reinstall one of my hard disks broke down
with a head crash, leaving the pool in the following state:

-----------------------------------------------------------------------------
# zpool import
# pwd
/systpool/ROOT/snv_91/etc/zfs
nathan# ls
zpool.cache
nathan# zpool import
  pool: datapool
    id: 4576033949025518067
 state: DEGRADED
status: The pool was last accessed by another system.
action: The pool can be imported despite missing or damaged devices.  The
        fault tolerance of the pool may be compromised if imported.
   see: http://www.sun.com/msg/ZFS-8000-EY
config:

        datapool    DEGRADED
          raidz1    DEGRADED
            c1d0    ONLINE
            c1d1    ONLINE
            c2d0    FAULTED  corrupted data
            c0d1    ONLINE
-----------------------------------------------------------------------------

The problem is, that the import doesn''t work, even with -f:

-----------------------------------------------------------------------------
# zpool import datapool
cannot import ''datapool'': pool may be in use from other
system, it was last accessed by nathan.alashan.nongo (hostid: 0x119c53df) on Tue
Sep 30 21:10:12 2008
use ''-f'' to import anyway
# zpool import -f datapool
cannot import ''datapool'': one or more devices is currently
unavailable
-----------------------------------------------------------------------------

I discussed this issue with my colleagues and we''re sure that this
should work.
Am I missing something here, or does the combination of a fresh install and a
broken pool trigger a bug in ZFS?

Regards
Christian Walther
--
This message posted from opensolaris.org

Nobody?
Should I open a bug report?

Maybe some more info: The port the broken disk was attached to is no used by the
new boot device containing a fresh zpool. I didn''t want to loose all
information from the old root pool, so I decided to use another disk
temporarily.
Is it possible that the zpool import doesn''t fail because
there''s a device missing, but because the port of this missing device
is in use by another one belonging to a different poo?
-- 
This message posted from opensolaris.org

Okay, I found out what the problem was:

As I expected in my last post ZFS didn''t like the idea of having
another disk containing a running zpool on a location that was previously
occupied by a disk that died. Last weekend I created a few snapshots to be moved
to another disk, so today I was able to remove this disk. A normal
# zpool import datapool
afterwards did the trick.

For the record: My configuration is based on 4 PATA Disks and 1 SATA drive. The
SATA drive is supposed to be a boot disk (I''m about to get another one
to setup a proper mirror).
Now, one of my PATA disks died after I managed to ruin the boot archives, so I
had to reinstall. Since I wanted to keep the configuration of my first install
intact I had to use another disk -- and there was a spare 20GB PATA drive lying
around. I attached it to the port the broken disk was attached to before,
because it was the only free PATA port.
During the reinstall the 20GB PATA drive attached to the port of the previously
failed disk became the new rootpool. Something ZFS doesn''t seem to
like.

Conclusion: Never attach a disk that is not supposed to be a replacement for a
faulted drive to a port that is used in a zpool configuration.

The question remains wether or not this is supposed to be standard behaviour, or
a bug. Might be a philosophical issue to be discussed. But at least I expect ZFS
to be more precise in this regard. A message like "Error: Pool
can''t be exported because at least one device has been exchanged with a
device belonging to a different pool that is already imported on this
system" would be fine.
-- 
This message posted from opensolaris.org