zfs discuss - Sep 2010 - zpool create using whole disk - do I add "p0"? E.g. c4t2d0 or c42d0p0

I have seen conflicting examples on how to create zpools using full disks.  The
zpool(1M) page uses "c0t0d0" but OpenSolaris Bible and others show
"c0t0d0p0".  E.g.:

zpool create tank raidz c0t0d0 c0t1d0 c0t2d0 c0t3d0 c0t4d0 c0t5d0
zpool create tank raidz c0t0d0p0 c0t1d0p0 c0t2d0p0 c0t3d0p0 c0t4d0p0 c0t5d0p0

I have not been able to find any discussion on whether (or when) to add the
"p0" to the disk name while creating whole disk zpools.

In my  case, I will have 9 disks in Raid-Z3 (home file server; requirements are
adequate performance, maximum long-term storage reliability and minimal
maintenance).  I want to setup the zpools correct the first time to avoid any
future issues.

Thanks for you help and insight.

-- Craig
-- 
This message posted from opensolaris.org

Hi Craig,

D''oh. I kept wondering where those p0 examples were coming from.

Don''t use the p* devices for your storage pools. They represent
the larger fdisk partition.

Use the d* devices instead, like this example below:

zpool create tank raidz c0t0d0 c0t1d0 c0t2d0 c0t3d0 c0t4d0 c0t5d0

Thanks,

Cindy

On 09/07/10 09:59, Craig Stevenson wrote:
> I have seen conflicting examples on how to create zpools using full disks. 
The zpool(1M) page uses "c0t0d0" but OpenSolaris Bible and others show
"c0t0d0p0".  E.g.:
> 
> zpool create tank raidz c0t0d0 c0t1d0 c0t2d0 c0t3d0 c0t4d0 c0t5d0
> zpool create tank raidz c0t0d0p0 c0t1d0p0 c0t2d0p0 c0t3d0p0 c0t4d0p0
c0t5d0p0
> 
> I have not been able to find any discussion on whether (or when) to add the
"p0" to the disk name while creating whole disk zpools.
> 
> In my  case, I will have 9 disks in Raid-Z3 (home file server; requirements
are adequate performance, maximum long-term storage reliability and minimal
maintenance).  I want to setup the zpools correct the first time to avoid any
future issues.
> 
> Thanks for you help and insight.
> 
> -- Craig

> Hi Craig,
> Don''t use the p* devices for your storage pools. They
> represent the larger fdisk partition.
> 
> Use the d* devices instead, like this example below:
Good advice, something I wondered about too. 

However, aside from my having guessed right once (I think...) I have no clue why
this should be. Can you expound a bit on the reasoning behind this advice?
-- 
This message posted from opensolaris.org

Hi--

It might help to review the disk component terminology description:

c#t#d#p# = represents the the fdisk partition on x86 systems, where
you can have up to 4 fdisk partitions, such as one for the Solaris
OS or a Windows OS. An fdisk partition is the larger container of the
disk or disk slices.

c#t#d# = represents the whole disk.

c#t#d#s# = represents the disk slice, used for the root pool because
the current boot limitation that says we must boot from a slice.

The issue is that if you don''t understand that the c#t#d#p# device
contains the c#t#d# or c#t#d#s# devices, you might create a pool
that contains p#, d#, and s# components, in an overlapping kind of
way (we''ve seen it). A bug exists to prevent pool creation with p#
devices.

You are probably okay if you use c0t0d0p0 and c0t1d0p0 and never
overlap the fdisk components but we don''t test this configuration
and its not supported.

Thanks,

Cindy

On 09/08/10 23:07, R.G. Keen wrote:
>> Hi Craig,
>> Don''t use the p* devices for your storage pools. They
>> represent the larger fdisk partition.
>>
>> Use the d* devices instead, like this example below:
> 
> Good advice, something I wondered about too. 
> 
> However, aside from my having guessed right once (I think...) I have no
clue why this should be. Can you expound a bit on the reasoning behind this
advice?