Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Questions about ZFS pool dataset naming conventions.

635 views
Skip to first unread message

Grant Taylor

unread,
May 12, 2020, 8:10:47 PM5/12/20
to
Since I got such a quick response to my last question, I figured I'd ask
another one that's been bugging me.

1) Why do ZFS pools use <pool name>/ROOT/solaris as their install
point? Why not <pool name>/solaris?

I get wanting something to make it easy to separate other datasets in
the pool.

2) Why capital ROOT instead of lowercase root?

Is it possibly caps to make it more annoying to type and thus less
likely to be accidentally typed?

3) Do the same answers apply to <pool name>/BOOT/solaris?

4) Is there any historical significance to "rpool" / "bpool" / "tank"
(case insensitive)?

Thank you for helping this n00b learn another thing.



--
Grant. . . .
unix || die

cindy.sw...@gmail.com

unread,
May 13, 2020, 10:10:56 AM5/13/20
to
Hi Grant,

An important point is that the rpool, boot, OS components should be left alone and any non-rpool, boot, OS data should go into a data pool and not the rpool. I wouldn't spend too much time on the rpool components and how its structured. By keeping rpool and non-rpool data separate, you can apply file system properties to more easily match your data and rpool stays small and mostly static and also easier to recover. Adding non-rpool data to an rpool is not supported.

A few comments below.

Solaris 11.4 ZFS Admin Guide is a good reference:

https://docs.oracle.com/cd/E37838_01/html/E61017/index.html

Thanks, Cindy



On Tuesday, May 12, 2020 at 6:10:47 PM UTC-6, Grant Taylor wrote:
> Since I got such a quick response to my last question, I figured I'd ask
> another one that's been bugging me.
>
> 1) Why do ZFS pools use <pool name>/ROOT/solaris as their install
> point? Why not <pool name>/solaris?
ROOT is just a mount-point structure. The lowercase root is the default home directory for the root (superuser) user.
>
> I get wanting something to make it easy to separate other datasets in
> the pool.

See above. Create a separate pool for your data pool.
>
> 2) Why capital ROOT instead of lowercase root?

Again, just a mount point structure and also Solaris 11 has a lowercase root directory.
>
> Is it possibly caps to make it more annoying to type and thus less
> likely to be accidentally typed?

I agree using caps is not very UNIX like.
>
> 3) Do the same answers apply to <pool name>/BOOT/solaris?

I see a lowercase boot/solaris directory.


>
> 4) Is there any historical significance to "rpool" / "bpool" / "tank"
> (case insensitive)?

rpool is for root pool, bpool looks like boot pool. The use of tank for a data pool name comes from the Matrix movie series. The ZFS eng team was keen on this movie series when ZFS was developed.

John D Groenveld

unread,
May 13, 2020, 3:56:56 PM5/13/20
to
In article <54d26a19-c02f-4312...@googlegroups.com>,
<cindy.sw...@gmail.com> wrote:
>ROOT is just a mount-point structure. The lowercase root is the default
>home directory for the root (superuser) user.

# beadm create foo
# zfs list -r rpool

>I agree using caps is not very UNIX like.

IMO as effective as using leading underscore to mark a private
interface or variable.

>rpool is for root pool, bpool looks like boot pool. The use of tank for
>a data pool name comes from the Matrix movie series. The ZFS eng team
>was keen on this movie series when ZFS was developed.

Jeff Bonwick explains the Matrix references in the docs in the
Birth of ZFS talk::
<URL:https://openzfs.org/wiki/OpenZFS_Developer_Summit_2015>

John
groe...@acm.org

Grant Taylor

unread,
May 14, 2020, 1:08:21 AM5/14/20
to
On 5/13/20 8:10 AM, cindy.sw...@gmail.com wrote:
> Hi Grant,

Hi Cindy,

> An important point is that the rpool, boot, OS components should be
> left alone and any non-rpool, boot, OS data should go into a data
> pool and not the rpool.

I hear you. I think I understand what you're saying, and why you say
it. (I've said quite similar things about volume groups for AIX and
Linux for 15+ years.)

But, I think ZFS is starting to make it into a space where doing that is
not an option.

I.e. I have a system (glorified bastion host) that is a 1U box with 4 ×
3½" drives that I'm rebuilding. The hardware RAID-6 lost 3 drives
during the current WFH climate. I've rebuilt it with new drives as a
temporary measure. But, I'd like to rebuild it again in the near future
as a 4 way ZFS /mirror/. (I really don't want to have this problem
again.) I don't need space. I need stability and longevity. I'd like
the pool to outlast the OS that's installed on the machine.

I'd really like to build a pair of them and have each machine do a zfs
send to each other's pools. That way, even if I loose four hard drives,
I would still have the files on the other machine.

Aside: The drives are multi TB for an OS install that's (considerably)
less than 10 GB.

I could partition the drives and have bpool and an rpool on them. (I
had originally thought about a 4 way mirror for bpool and RAIDZ2 for
rpool. But decided that everything could easily fit in the 4 way
mirror.) The reason for having rpool separate was that GRUB doesn't
support booting from any form of Z-RAID more complex than mirroring.

So, I guess that I could still partition the drives and have rpool dpool
on the same drives. But that seems a little silly to me for my
particular use case.

So, I'm in a situation that really lends itself to having OS specific
files and DATA (non-OS specific) file in the same pool. The hardware
doesn't lend itself to attaching more drives. Nor do I think I would do
so if it did. I simply don't see the need. Not when each system's
rpool can hold the necessary data from the other system.

Please correct me if I'm wrong about any of this.

> I wouldn't spend too much time on the rpool components and how its
> structured.

I have a nasty habit of asking that pesky "why" question, particularly
in the context of why something was done the way that it was.

> By keeping rpool and non-rpool data separate, you can apply file system
> properties to more easily match your data and rpool stays small and
> mostly static and also easier to recover.

Having rpool and dpool isn't really an option for this use case.

> Adding non-rpool data to an rpool is not supported.

I'm taking that to mean that Oracle (et al.) won't support me as in hold
my hand if I have non-OS specific files in the rpool.

Something that's *SIGNIFICANTLY* different than it won't work.

All the crazy things that you could do with old Cisco gear vs what TAC
would(n't) ""support you do comes to mind.

> A few comments below.
>
> Solaris 11.4 ZFS Admin Guide is a good reference:
>
> https://docs.oracle.com/cd/E37838_01/html/E61017/index.html

ACK

> Thanks, Cindy

Thank you Cindy.

> ROOT is just a mount-point structure. The lowercase root is the
> default home directory for the root (superuser) user.

I get the difference between "ROOT" as the base for the system and
"root" as the root user's home directory. To me, those are two
different things.

That doesn't give any indication why "ROOT" as the base for the system
is capital.

Or are you saying that the capital was done to differentiate the two?

> See above. Create a separate pool for your data pool.

See above. That's not always a viable option.

> Again, just a mount point structure and also Solaris 11 has a lowercase
> root directory.

Understood.

I'm trying to understand the history and motivation behind it being
named "ROOT", and specifically why it's capitals.

> I agree using caps is not very UNIX like.

There are plenty of commands that have options that use capital letters.
I believe there are even some commands that have capitals in their
name. (X11 related things come to mind.)

> I see a lowercase boot/solaris directory.

Okay.

I wonder if the case difference is version related.

> rpool is for root pool, bpool looks like boot pool. The use of tank
> for a data pool name comes from the Matrix movie series. The ZFS eng
> team was keen on this movie series when ZFS was developed.

ACK

Thank you for your reply Cindy.

Grant Taylor

unread,
May 14, 2020, 1:10:27 AM5/14/20
to
On 5/13/20 1:56 PM, John D Groenveld wrote:
> IMO as effective as using leading underscore to mark a private
> interface or variable.

Agreed.

About the only thing it will likely do is keep someone from typing it
accidentally. They will have to want to type it.

> Jeff Bonwick explains the Matrix references in the docs in the Birth of
> ZFS talk:: <URL:https://openzfs.org/wiki/OpenZFS_Developer_Summit_2015>

Thank you for the link. I watched Jeff's talk and a few others. I'll
watch more as time permits.

Thank you for your reply John.

Scott

unread,
May 14, 2020, 8:24:18 AM5/14/20
to
On Wednesday, May 13, 2020 at 10:08:21 PM UTC-7, Grant Taylor wrote:
> I.e. I have a system (glorified bastion host) that is a 1U box with 4 ×
> 3½" drives that I'm rebuilding. The hardware RAID-6 lost 3 drives
> during the current WFH climate. I've rebuilt it with new drives as a
> temporary measure. But, I'd like to rebuild it again in the near future
> as a 4 way ZFS /mirror/. (I really don't want to have this problem
> again.) I don't need space. I need stability and longevity. I'd like
> the pool to outlast the OS that's installed on the machine.

IMO you don't want to use zpools on a host that has a hardware RAID-anything host bus adapter. My experience is with HPE hardware.

If you want to use ZFS use it with a simple HBA that will present the raw HDD to you without any interference, translation, or buffering.

Regards, Scott

Grant Taylor

unread,
May 14, 2020, 2:02:35 PM5/14/20
to
On 5/14/20 6:24 AM, Scott wrote:
> IMO you don't want to use zpools on a host that has a hardware
> RAID-anything host bus adapter. My experience is with HPE hardware.

I agree in an ideal world, that it is best to have dumb HBAs between ZFS
and the disks. But that's not always an option.

Nor does hardware RAID give me nearly all the options that ZFS does.

> If you want to use ZFS use it with a simple HBA that will present
> the raw HDD to you without any interference, translation, or buffering.

That is not an option in all cases. Nor do I think that sub-optimal
hardware options should exclude me from using ZFS.

cindy.sw...@gmail.com

unread,
May 14, 2020, 2:14:39 PM5/14/20
to
Hi Grant,

See my comments below. I agree with Scott's comments not to use any kind of hardware RAID with ZFS root pool. Its a royal pain to replace a failed device and undoes a lot of goodness of ZFS root pool brings to device management.

Thanks, Cindy
You are not incorrect and if like me, will learn by experience.

I think you are saying that a 4-way mirrored root pool is safer than a 2-way rpool and 2-way data pool. Just be clear, the root pool only supports a 4-way mirror and not a 2x2 mirror.

If you are going to create a file system in rpool then do not use existing ROOT or VARSHARE components. Create something like rpool/data and then run tests like create a BE, update it and roll back.

MHO is that disks are cheap and if continued operation is your number one priority, follow the supported path because this is untested.

Also review the best practices section in the ZFS Admin Guide.
Yes, that is what I remember but memories from 2008, when ZFS boot was developed, are dim.

Gary R. Schmidt

unread,
May 14, 2020, 10:54:08 PM5/14/20
to
Many RAID adapters can be configured as having each disk as a RAID0
volume, which you can then use as disks for a ZFS setup.

I know the LSI MegaRAID controller can be set up that way, I've done to
a few RX500's and their relatives over the years. (Not all to run
Solaris or ZFS, but to get them to do what we wanted.)

Cheers,
Gary B-)

--
Waiting for a new signature to suggest itself...

Grant Taylor

unread,
May 19, 2020, 12:41:59 AM5/19/20
to
On 5/14/20 8:51 PM, Gary R. Schmidt wrote:
> Many RAID adapters can be configured as having each disk as a RAID0
> volume, which you can then use as disks for a ZFS setup.

Yep.

> I know the LSI MegaRAID controller can be set up that way, I've done to
> a few RX500's and their relatives over the years.  (Not all to run
> Solaris or ZFS, but to get them to do what we wanted.)

I'm using Dell PowerEdge RAID Controllers because that's what's in the
re-purposed systems and I can't (for many reasons) change the hardware.

At least with some LSI cards, you can change them from Integrated RAID
(a.k.a. IR) mode to Initiator / Target (a.k.a. IT) mode. That's my
preference if I can do so.

Sadly the systems that I'm working with at the moment don't support that.

Grant Taylor

unread,
May 19, 2020, 12:47:56 AM5/19/20
to
On 5/14/20 12:14 PM, cindy.sw...@gmail.com wrote:
> Hi Grant,

Hi Cindy,

> See my comments below. I agree with Scott's comments not to use any
> kind of hardware RAID with ZFS root pool. Its a royal pain to replace
> a failed device and undoes a lot of goodness of ZFS root pool brings
> to device management.

Understood. (See my reply to Gary for more details.)

> I think you are saying that a 4-way mirrored root pool is safer than
> a 2-way rpool and 2-way data pool.

Yes. In many different ways.

> Just be clear, the root pool only supports a 4-way mirror and not a
> 2x2 mirror.

Yep, the zpool would be the typical:

<pool>
<mirror>
<disk 1>
<disk 2>
<disk 3>
<disk 4>

> If you are going to create a file system in rpool then do not use
> existing ROOT or VARSHARE components. Create something like rpool/data
> and then run tests like create a BE, update it and roll back.

I was planing on staying out of <pool>/ROOT and any other system data
sets. I'd create a new one, likely based on the name of the other
system, possibly under "backup" or something like that.

<pool>/backups/<other system name>

I /might/ put this under data.

<pool>/data/backups/<other system name>

> MHO is that disks are cheap and if continued operation is your number
> one priority, follow the supported path because this is untested.

Supported path per say isn't really an option for this. (See my reply
to Gary.) I'm making the best of what I have while trying to follow the
Solaris & ZFS spirit.

> Also review the best practices section in the ZFS Admin Guide.

Understood. Will do.

> Yes, that is what I remember but memories from 2008, when ZFS boot
> was developed, are dim.

ACK

Ian Collins

unread,
May 19, 2020, 12:51:26 AM5/19/20
to
On 19/05/2020 16:41, Grant Taylor wrote:
> On 5/14/20 8:51 PM, Gary R. Schmidt wrote:
>> Many RAID adapters can be configured as having each disk as a RAID0
>> volume, which you can then use as disks for a ZFS setup.
>
> Yep.
>
>> I know the LSI MegaRAID controller can be set up that way, I've done to
>> a few RX500's and their relatives over the years.  (Not all to run
>> Solaris or ZFS, but to get them to do what we wanted.)
>
> I'm using Dell PowerEdge RAID Controllers because that's what's in the
> re-purposed systems and I can't (for many reasons) change the hardware.

Which Dell controller model are you stuck with? Most do have the option
of having each disk as a RAID0 volume or simply as a raw drive.

--
Ian.

Grant Taylor

unread,
May 19, 2020, 1:35:26 AM5/19/20
to
On 5/18/20 10:51 PM, Ian Collins wrote:
> Which Dell controller model are you stuck with?  Most do have the option
> of having each disk as a RAID0 volume or simply as a raw drive.

It looks like a PERC H710P.

Yes, I can create a bunch of independent single disk RAID 0 volumes. I
don't think I've been able to geet pass-through working in a while.
I'll take another swing at that when I'm doing the physical work.

At the moment, I'm trying to learn some history about ZFS and why things
were done the way that they were.

Ian Collins

unread,
May 19, 2020, 1:48:36 AM5/19/20
to
On 19/05/2020 17:35, Grant Taylor wrote:
> On 5/18/20 10:51 PM, Ian Collins wrote:
>> Which Dell controller model are you stuck with?  Most do have the option
>> of having each disk as a RAID0 volume or simply as a raw drive.
>
> It looks like a PERC H710P.
>
> Yes, I can create a bunch of independent single disk RAID 0 volumes. I
> don't think I've been able to geet pass-through working in a while.
> I'll take another swing at that when I'm doing the physical work.

https://www.dell.com/community/PowerEdge-HDD-SCSI-RAID/Enable-passthrough-on-H710P/td-p/4365545

> At the moment, I'm trying to learn some history about ZFS and why things
> were done the way that they were.

A worthy quest :)

--
Ian.

Gary R. Schmidt

unread,
May 19, 2020, 1:54:07 AM5/19/20
to
Most of the PowerEdge RAID adaptors that have crossed my path can be
flashed between IR and IT modes.

It's not a job for the faint-hearted (IIRC it requires a Windows box),
but usually it can be done.

Casper H.S. Dik

unread,
May 19, 2020, 6:10:02 AM5/19/20
to
Grant Taylor <gta...@tnetconsulting.net> writes:

>Since I got such a quick response to my last question, I figured I'd ask
>another one that's been bugging me.

>1) Why do ZFS pools use <pool name>/ROOT/solaris as their install
>point? Why not <pool name>/solaris?

Under root, all the other boot environment exist.

So instead of having:

rpool/11.3-sru3
rpool/11.3-sru5
rpool/11.4-FCS

we gather them under root:

rpool/ROOT/11.3-sru3
rpool/ROOT/11.3-sru5
rpool/ROOT/11.4-FCS

So we use only one reserved name (well, two: ROOT and VARSHARE) and
under ROOT we have an own name space for boot environments.

>I get wanting something to make it easy to separate other datasets in
>the pool.

>2) Why capital ROOT instead of lowercase root?

Because most customers use lowercase file ssytems; we use ROOT and
VARSHARE so they where much less likely to conflict with customers.

>Is it possibly caps to make it more annoying to type and thus less
>likely to be accidentally typed?

No. In ZFS you required to give names to file systems; in UFS you'd
have "/" and it has no name.

>3) Do the same answers apply to <pool name>/BOOT/solaris?

What thype of system is that?

>4) Is there any historical significance to "rpool" / "bpool" / "tank"
>(case insensitive)?


rpool == root pool
bpool == boot pool (I think this is specific to specific larger systems
which have no internal storage, if I remember correctly)

"tank" is what is generally used in examples but in reality it is used a
lot; it is NOT reserved and has no specific meaning for Solaris.

Casper

John D Groenveld

unread,
May 19, 2020, 7:27:30 AM5/19/20
to
In article <5k8cpg-...@paranoia.mcleod-schmidt.id.au>,
Gary R. Schmidt <grsc...@acm.org> wrote:
>It's not a job for the faint-hearted (IIRC it requires a Windows box),
>but usually it can be done.

It can be done from FreeDOS and probably Unbreakable Linux.
But I would just tell Dell to swap in the HBA part for the PERC.
John
groe...@acm.org

Grant Taylor

unread,
May 20, 2020, 11:02:31 AM5/20/20
to
On 5/18/20 11:48 PM, Ian Collins wrote:
> https://www.dell.com/community/PowerEdge-HDD-SCSI-RAID/Enable-passthrough-on-H710P/td-p/4365545

Hum. I'll check that out.

Not that I'm going to make any more changes to this system remotely,
during extended WFH, without functional remote console.

But that's good information for future.

> A worthy quest :)

I thought so.

Thankfully I can benefit from the knowledgeable people here in
comp.unix.solaris. :-)

Grant Taylor

unread,
May 20, 2020, 11:04:31 AM5/20/20
to
On 5/18/20 11:49 PM, Gary R. Schmidt wrote:
> Most of the PowerEdge RAID adaptors that have crossed my path can be
> flashed between IR and IT modes.

That's good to know.

> It's not a job for the faint-hearted (IIRC it requires a Windows box),
> but usually it can be done.

I've done it on a few non-PERC controllers. Things either worked, or
reverted back to I.R. mode. I've not bricked anything yet. Thankfully.

However, this is a future endeavor.

Grant Taylor

unread,
May 20, 2020, 11:05:29 AM5/20/20
to
On 5/19/20 5:27 AM, John D Groenveld wrote:
> It can be done from FreeDOS and probably Unbreakable Linux.

Good to know.

> But I would just tell Dell to swap in the HBA part for the PERC.

As previously stated, this is re-use of existing hardware. Engaging
Dell to alter it or send me parts for me to alter it, is not an option
this time.

Grant Taylor

unread,
May 20, 2020, 11:14:35 AM5/20/20
to
On 5/19/20 4:09 AM, Casper H.S. Dik wrote:
> Under root, all the other boot environment exist.
>
> So instead of having:
>
> rpool/11.3-sru3
> rpool/11.3-sru5
> rpool/11.4-FCS
>
> we gather them under root:
>
> rpool/ROOT/11.3-sru3
> rpool/ROOT/11.3-sru5
> rpool/ROOT/11.4-FCS
>
> So we use only one reserved name (well, two: ROOT and VARSHARE) and
> under ROOT we have an own name space for boot environments.

Thank you for clarification.

> Because most customers use lowercase file ssytems; we use ROOT and
> VARSHARE so they where much less likely to conflict with customers.

Understood.

> No. In ZFS you required to give names to file systems; in UFS you'd
> have "/" and it has no name.

I get the requirement to give the file system a name. But I don't see
what difference that has to do with "ROOT" vs "root".

> What thype of system is that?

I ran across it in multiple documents, many of which on Oracle's
website. I don't have specifics at hand.

> rpool == root pool
> bpool == boot pool (I think this is specific to specific larger systems
> which have no internal storage, if I remember correctly)

My understanding is that bpool comes into play when the rpool can't be
booted from for one reason or another. rpool's structure comes to mind,
as in RAIDZ{1,2,3} / stripe.

> "tank" is what is generally used in examples but in reality it is used a
> lot; it is NOT reserved and has no specific meaning for Solaris.

Thank you for clarifying that tank / TANK is not reserved.
0 new messages