zfs discuss - Sep 2007 - "zoneadm clone" doesn''t support ZFS snapshots in s10u4?

according to the zoneadm(1m) man page on s10u4:

     clone [-m copy] [-s zfs_snapshot] source_zone

         Install a zone by copying an  existing  installed  zone.
         This  subcommand  is  an  alternative way to install the
         zone.

         -m copy

             Force the clone to be a copy, even if a "ZFS  clone"
             is possible.


         -s zfs_snapshot

             Specify the name of a ZFS snapshot  to  use  as  the
             source of the clone. The snapshot must be a snapshot
             of the source zone taken from  a  previous  "zoneadm
             clone" installation.

         The source zone must be halted  before  this  subcommand
         can be used.


but this doesn''t work:

# zoneadm -z myzone clone -s mysnap blah
-s: illegal option -- s
usage:  clone [-m method] zonename
        Clone the installation of another zone.

without "-s mysnap", zoneadm simply copies the contents of the source 
zone, and doesn''t do a "zfs clone" as it should. should I
file a bug report?

I suspect that somehow the man page is out of sync with reality. most of 
the zones I want to create will be blank, and therefore will consume 
lots of disk space without the use of "zfs clone". how do I take 
advantage of "zfs clone" for zones?

and how does one sensibly back up the clones without duplication? "zfs 
send" on all of the ZFS filesystems will give me about 20 times as much 
data in the backup as I really need.

grant.

On 9/19/07, grant beattie <grant at grunta.com>
wrote:
> according to the zoneadm(1m) man page on s10u4:
>
>      clone [-m copy] [-s zfs_snapshot] source_zone
>
>          Install a zone by copying an  existing  installed  zone.
>          This  subcommand  is  an  alternative way to install the
>          zone.
That''s interesting... I reported this as a bug during the S10U2
(11/06) beta and it got fixed for 11/06.  My bug report was closed as
"6480274 as duplicate of 6383119."  This was wrong - I was reporting a
man page bug and not the feature request (CR 6383119).  The feature
request from me was the previous bug that I filed in that beta
program.  :)

Someone really wants this man page to be out of sync with the command.

The rather consistent answer is that zoneadm clone will not do zfs
until live upgrade does zfs.  Since there is a new project in the
works (Snap Upgrade) that is very much targeted at environments that
use zfs, I would be surprised to see zfs support come into live
upgrade.

-- 
Mike Gerdts
http://mgerdts.blogspot.com/

On Wed, 19 Sep 2007, Mike Gerdts wrote:
> The rather consistent answer is that zoneadm clone will not do zfs until
> live upgrade does zfs.  Since there is a new project in the works (Snap
> Upgrade) that is very much targeted at environments that use zfs, I
> would be surprised to see zfs support come into live upgrade.
I for one would like to see live upgrade support ZFS.  Even with Snap
Upgrade on the horizon (the page on the OpenSolaris site says
''March'' but
the current scedule is a sea of TBDs [see
http://opensolaris.org/os/project/caiman/Snap_Upgrade/schedule/ for
more]), I think there are enough systems and configs out there to warrant
putting ZFS support into Live Upgrade.

LU is familiar and proven, and deployed widely.  My opinion only, but I
think it would be short-sighted to stop supporting it just because another
technology is being developed.


Regards,
markm

On 9/20/07, Mark J Musante <mmusante at east.sun.com>
wrote:
> I for one would like to see live upgrade support ZFS.  Even with Snap
> Upgrade on the horizon (the page on the OpenSolaris site says
''March'' but
> the current scedule is a sea of TBDs [see
> http://opensolaris.org/os/project/caiman/Snap_Upgrade/schedule/ for
> more]), I think there are enough systems and configs out there to warrant
> putting ZFS support into Live Upgrade.
>
> LU is familiar and proven, and deployed widely.  My opinion only, but I
> think it would be short-sighted to stop supporting it just because another
> technology is being developed.
I''m with you.  Snap Upgrade has a March deliverable for OpenSolaris,
which means that the widely deployed base (Solaris 10 and earlier)
won''t see it for some time after that, at best.

If people find the use of zfs clones more important than live upgrade
(or upgrade, for that matter) search the zfs or zones lists from late
last year or early this year - I documented a manual procedure that is
essentially the following - there may be more details in the older
message.

zoneadm -z master detach
zfs snapshot
zfs clone
zoneadm -z master attach
zonecfg -z newzone create -t master
# change IP''s et. al.
zoneadm -z newzone attach
zoneadm -z newzone boot -s
zlogin newzone sys-unconfig
zoneadm -z newzone boot
zlogin -C newzone

-- 
Mike Gerdts
http://mgerdts.blogspot.com/

Mike,

I followed your procedure for cloning zones and it worked well up until
yesterday when I tried applying the S10U4 kernel patch 12001-14 and it
wouldn''t apply because I had my zones on zfs :(

I''m still figuring out how to fix this other than moving all of my
zones onto UFS.

Anyone got any tips?

matthew
 
 
This message posted from opensolaris.org

Matthew Flanagan wrote:
> Mike,
>
> I followed your procedure for cloning zones and it worked well up until
yesterday when I tried applying the S10U4 kernel patch 12001-14 and it
wouldn''t apply because I had my zones on zfs :(
>
> I''m still figuring out how to fix this other than moving all of my
zones onto UFS.
>   
I don''t have any advice, unfortunately, but I do know that in my case 
putting zones on UFS is simply not an option. there must be a way 
considering there is nothing in the documentation to suggest that zones 
on ZFS are not supported.

one question though, why does patchadd care about filesystems in the 
first place? what if I put my zones on VxFS, or QFS? I don''t see why it
should make any difference to patchadd. live upgrade is obviously 
another kettle of fish entirely, though.

grant.

Mike, Grant,

I reported the zoneadm.1m man page problem to the man page group.

I also added some stronger wording to the ZFS Admin Guide and the
ZFS FAQ about not using ZFS for zone root paths for the Solaris 10
release and that upgrading or patching is not supported for either
Solaris 10 or Solaris Express release.

It is true that none of the current install/patch tools recognize
ZFS yet and this has been discussed many times before. You can follow
the ZFS boot/root project here:

http://opensolaris.org/os/community/zfs/boot/

I''ve tried to add all the things you *can''t* do with ZFS in
the ZFS
FAQ and the next topic to add is around the inability to split
mirrors.

If you are considering doing something with ZFS where recovering
would be difficut, then please ask here first. :-)

Cindy

Mike Gerdts wrote:
> On 9/19/07, grant beattie <grant at grunta.com> wrote:
> 
>>according to the zoneadm(1m) man page on s10u4:
>>
>>     clone [-m copy] [-s zfs_snapshot] source_zone
>>
>>         Install a zone by copying an  existing  installed  zone.
>>         This  subcommand  is  an  alternative way to install the
>>         zone.
> 
> 
> That''s interesting... I reported this as a bug during the S10U2
> (11/06) beta and it got fixed for 11/06.  My bug report was closed as
> "6480274 as duplicate of 6383119."  This was wrong - I was
reporting a
> man page bug and not the feature request (CR 6383119).  The feature
> request from me was the previous bug that I filed in that beta
> program.  :)
> 
> Someone really wants this man page to be out of sync with the command.
> 
> The rather consistent answer is that zoneadm clone will not do zfs
> until live upgrade does zfs.  Since there is a new project in the
> works (Snap Upgrade) that is very much targeted at environments that
> use zfs, I would be surprised to see zfs support come into live
> upgrade.
>

grant beattie wrote:
> I don''t have any advice, unfortunately, but I do know that in my
case
> putting zones on UFS is simply not an option. there must be a way 
> considering there is nothing in the documentation to suggest that zones 
> on ZFS are not supported.
> 
There''s a very explicit "Do not place the zonepath on ZFS for this
release" in this doc: 
http://docs.sun.com/app/docs/doc/817-1592/z.conf.start-5?a=view

> one question though, why does patchadd care about filesystems in the 
> first place? what if I put my zones on VxFS, or QFS? I don''t see
why it
> should make any difference to patchadd. live upgrade is obviously 
> another kettle of fish entirely, though.
patch and install tools can''t figure out pools yet.  If you have a 1GB 
pool and 10 filesystems on it, du reports each having 1GB, do you have 
10GB capacity?  The tools can''t tell.  Please check the archives, this 
subject has been extensively discussed.

CT

On 9/20/07, Matthew Flanagan <mattimustang at gmail.com>
wrote:
> Mike,
>
> I followed your procedure for cloning zones and it worked
> well up until yesterday when I tried applying the S10U4
> kernel patch 12001-14 and it wouldn''t apply because I had
> my zones on zfs :(
Thanks for sharing.  That sucks.
> I''m still figuring out how to fix this other than moving all of my
zones onto UFS.
How about a dtrace script that changes the fstyp in statvfs() returns
to say that it is ufs?  :)

I bet someone comes along and says that isn''t supported either...

-- 
Mike Gerdts
http://mgerdts.blogspot.com/

On 9/21/07, Christine Tran <Christine.Tran at sun.com>
wrote:
> patch and install tools can''t figure out pools yet.  If you have a
1GB
> pool and 10 filesystems on it, du reports each having 1GB, do you have
> 10GB capacity?  The tools can''t tell.  Please check the archives,
this
> subject has been extensively discussed.
Two responses come immediately to mind...

1) Thanks for protecting stupid/careless people from doing bad things.
2) UNIX has a longstanding tradition of adding a -f flag for cases
when the sysadmin realizes there is additional risk but feels that
appropriate precautions have been taken.

I would really like to ask Sun for a roadmap as to when this is going
to be supported.  Since this is the zfs list (not zones or install
list) and it is OpenSolaris (not Solaris) I guess I should probably
find a more appropriate forum.

So, for now I will use OpenSolaris where I can and wait patiently for
the new installer + snap upgrade basket and wait for it to find its
way into Solaris in about a year or two.  In the meantime, I''ll
probably end up putting most zones on a particular competitor''s NAS
devices and looking into how well their file system cloning
capabilities play in coordination with iSCSI.

<irony>
Oh, wait!  What if the NAS device runs out of space while I''m
patching?  Better rule out the thin provisioning capabilities of the
HDS storage that Sun sells as well.
</irony>

-- 
Mike Gerdts
http://mgerdts.blogspot.com/

Bah, wrong list.

A timeline would be really nice for when this is likely to be sorted
out - higher
priority than ZFS root IMO.

---------- Forwarded message ----------
From: Dick Davies <rasputnik at gmail.com>
Date: 22 Sep 2007 23:21
Subject: Re: [zfs-discuss] "zoneadm clone" doesn''t support
ZFS snapshots in
To: zones-discuss at opensolaris.org


On 21/09/2007, Mike Gerdts <mgerdts at gmail.com> wrote:
>
> I would really like to ask Sun for a roadmap as to when this is going
> to be supported.
The best way around this I can think of is to have a
''template'' zone
for cloning on UFS that you use to build your other (ZFS-backed) zones.
Then delegate a dataset into each zone to hold the important stuff.

Come upgrade time, you drop all the ''child'' zones, patch the
template
and use it to re-provision the other zones. Then drop the dataset back in.

Course, it''ll take a while to clone the template since it''s
UFS-backed...



--
Rasputin :: Jack of All Trades - Master of Nuns
http://number9.hellooperator.net/


-- 
Rasputin :: Jack of All Trades - Master of Nuns
http://number9.hellooperator.net/

Mike Gerdts wrote:
> <irony>
> Oh, wait!  What if the NAS device runs out of space while I''m
> patching?  Better rule out the thin provisioning capabilities of the
> HDS storage that Sun sells as well.
> </irony>
That failure mode exists anyway.  There is no real way to reserve
space prior to patching.

Some might remember the days when patches didn''t check for available
space... :-(
  -- richard