zfs discuss - May 2012 - Unexpected error adding a cache device to existing pool

On a Solaris 11 system I have a pool that was originally built with a 
log a cache device on a single SSD.  The SSD died and I realised I 
should have a mirror log, so I''ve just tried to replace the log a cache
with a pair of SSDs.

Adding the log was OK:

zpool add -f export log mirror c10t3d0s0 c10t4d0s0

But adding the cache fails:

zpool add -f export cache c10t3d0s1 c10t4d0s1
invalid vdev specification
the following errors must be manually repaired:
/dev/dsk/c10t3d0s2 is part of active ZFS pool export. Please see zpool(1M).
/dev/dsk/c10t3d0s1 overlaps with /dev/dsk/c10t3d0s2

Now that looks impossible to repair, s2 can''t be removed.  The SSD 
partition table is:

Total disk cylinders available: 19932 + 2 (reserved cylinders)

Part      Tag    Flag     Cylinders         Size            Blocks
   0 unassigned    wm       0 -  2674       16.00GB    (2675/0/0)   33555200
   1 unassigned    wm    2675 - 19931      103.22GB    (17257/0/0) 216471808
   2     backup    wu       0 - 19931      119.22GB    (19932/0/0) 250027008

Is there a solution?

-- 
Ian.

On 5/14/12 2:02 AM, Ian Collins wrote:
> Adding the log was OK:
>
> zpool add -f export log mirror c10t3d0s0 c10t4d0s0
>
> But adding the cache fails:
>
> zpool add -f export cache c10t3d0s1 c10t4d0s1
> invalid vdev specification
> the following errors must be manually repaired:
> /dev/dsk/c10t3d0s2 is part of active ZFS pool export. Please see zpool(1M).
> /dev/dsk/c10t3d0s1 overlaps with /dev/dsk/c10t3d0s2
The only solution I know of is to get rid of the whole-disk slice s2 
from the disk label. I ended up using prtvtoc to dump the table, editing 
it by hand, and feeding it to fmthard.

You could also try making s0 start at cylinder 1 instead of zero, so 
zpool doesn''t see a magic number on s2, but I don''t know if
that will be
enough.

-- 
Carson

On 05/14/12 10:32 PM, Carson Gaspar wrote:
> On 5/14/12 2:02 AM, Ian Collins wrote:
>> Adding the log was OK:
>>
>> zpool add -f export log mirror c10t3d0s0 c10t4d0s0
>>
>> But adding the cache fails:
>>
>> zpool add -f export cache c10t3d0s1 c10t4d0s1
>> invalid vdev specification
>> the following errors must be manually repaired:
>> /dev/dsk/c10t3d0s2 is part of active ZFS pool export. Please see
zpool(1M).
>> /dev/dsk/c10t3d0s1 overlaps with /dev/dsk/c10t3d0s2
> The only solution I know of is to get rid of the whole-disk slice s2
> from the disk label. I ended up using prtvtoc to dump the table, editing
> it by hand, and feeding it to fmthard.
>
> You could also try making s0 start at cylinder 1 instead of zero, so
> zpool doesn''t see a magic number on s2, but I don''t know
if that will be
> enough.
>
Thank you for the suggestions Carson.

Making s0 start at cylinder 1 did the trick.  I''m sure I
didn''t have to
do that when I originally built the pool, but that was back on Solaris 
11 Express.

-- 
Ian.