Corrupted zpool import -f FAILS state FAULTED

[Ronald Klop]

On Fri, 10 May 2013 16:07:36 +0200, Outback Dingo <outbackdingo at gmail.com>  
wrote:

> On Fri, May 10, 2013 at 9:45 AM, Volodymyr Kostyrko  
> <c.kworr at gmail.com>wrote:
>
>> 09.05.2013 15:31, Outback Dingo:
>>
>>  ok zfsgurus, FreeBSD 9.1-STABLE box zpool import -f reports pool status
>>> Faulted, one of more devices contains corrupted data, however its  
>>> showing
>>> the guid as faulted in the poll, and not the actual disk device  
>>> /dev/daX,
>>> the pool is a single vdev 24 disk raidz3. Essentially the hardward
>>> platform
>>> is a dual node system, with 8 enclosures connected to 24 SAS drives  
>>> via 4
>>> LSI cards. I am not currently using geom_multipath, but the box is  
>>> zoned
>>> so
>>> that each node can see 50% of the drives,
>>> in case of Failure, carp kicks in and migrates "zpool import -af" the
>>> pools
>>> onto the other node. it seems as though somehow the pool is now seeing
>>> guid
>>> and not devices, not sure if they have switched devices ids due to a
>>> reboot.
>>>
>>
>> Am not a zfs guru, but I'll try to help.
>>
>> Any console log snippets are welcome. What does "showing the guid as
>> faulted in the pool" looks like.
>>
>> What are the guids for all partitions? Do they interlap for different
>> nodes?
>>
>> ZFS recognizes devices by tasting they vdev labels and not by their
>> logical location and naming. It can safely report any vdev location -  
>> but
>> it requires the same set vdevs to bring pool online.
>
>
> zdb shows valid data on the drives, no drives have been removed from the  
> box
> whats confusing is why its using guids and not devices daX is what  
> puzzles
> me
>  camcontrol devlist and dmesg clearly show the devices are there. The SAS
> bus is shared
> so both nodes with 2 LSI controllers each see all drives. We were  
> utilizing
> a failover script
> if nodeA dies, carp would kick the script to import the pool to nodeB,  
> both
> nodes are in the
> same chassis and see all the enclosures and all the drives

Are the machines configured the same? As in _exactly_ the same. Glabel  
modules, hint files, sysctls, etc.

Ronald.