zfs discuss - Sep 2010 - zpool upgrade and zfs upgrade behavior on b145

Not sure what the best list to send this to is right now, so I have selected
a few, apologies in advance.

A couple questions.  First I have a physical host (call him bob) that was
just installed with b134 a few days ago.  I upgraded to b145 using the
instructions on the Illumos wiki yesterday.  The pool has been upgraded (27)
and the zfs file systems have been upgraded (5).

chris at bob:~# zpool upgrade rpool
This system is currently running ZFS pool version 27.
Pool ''rpool'' is already formatted using the current version.

chris at bob:~# zfs upgrade rpool
7 file systems upgraded

The file systems have been upgraded according to "zfs get version
rpool"

Looks ok to me.

However, I now get an error when I run zdb -D.  I can''t remember
exactly
when I turned dedup on, but I moved some data on rpool, and "zpool
list"
shows 1.74x ratio.

chris at bob:~# zdb -D rpool
zdb: can''t open ''rpool'': No such file or directory

Also, running zdb by itself, returns expected output, but still says my
rpool is version 22.  Is that expected?

I never ran zdb before the upgrade, since it was a clean install from the
b134 iso to go straight to b145.  One thing I will mention is that the
hostname of the machine was changed too (using these
instructions<http://wiki.genunix.org/wiki/index.php/Change_hostname_HOWTO>).
bob used to be eric.  I don''t know if that matters, but I
can''t open up the
"Users and Groups" from Gnome anymore, *"unable to su"* so
something is
still not right there.

Moving on, I have another fresh install of b134 from iso inside a virtualbox
virtual machine, on a total different physical machine.  This machine is
named weston and was upgraded to b145 using the same Illumos wiki
instructions.  His name has never changed.  When I run the same zdb -D
command I get the expected output.

chris at weston:~# zdb -D rpool
DDT-sha256-zap-unique: 11 entries, size 558 on disk, 744 in core
dedup = 1.00, compress = 7.51, copies = 1.00, dedup * compress / copies 7.51

However, after zpool and zfs upgrades *on both machines*, they still say the
rpool is version 22.  Is that expected/correct?  I added a new virtual disk
to the vm weston to see what would happen if I made a new pool on the new
disk.

chris at weston:~# zpool create test c5t1d0

Well, the new "test" pool shows version 27, but rpool is still listed
at 22
by zdb.  Is this expected /correct behavior?  See the output below to see
the rpool and test pool version numbers according to zdb on the host weston.


Can anyone provide any insight into what I''m seeing?  Do I need to
delete my
b134 boot environments for rpool to show as version 27 in zdb?  Why does zdb
-D rpool give me can''t open on the host bob?

Thank you in advance,

-Chris

chris at weston:~# zdb
rpool:
    version: 22
    name: ''rpool''
    state: 0
    txg: 7254
    pool_guid: 17616386148370290153
    hostid: 8413798
    hostname: ''weston''
    vdev_children: 1
    vdev_tree:
        type: ''root''
        id: 0
        guid: 17616386148370290153
        create_txg: 4
        children[0]:
            type: ''disk''
            id: 0
            guid: 14826633751084073618
            path: ''/dev/dsk/c5t0d0s0''
            devid: ''id1,sd at
SATA_____VBOX_HARDDISK____VBf6ff53d9-49330fdb/a''
            phys_path: ''/pci at 0,0/pci8086,2829 at d/disk at
0,0:a''
            whole_disk: 0
            metaslab_array: 23
            metaslab_shift: 28
            ashift: 9
            asize: 32172408832
            is_log: 0
            create_txg: 4
test:
    version: 27
    name: ''test''
    state: 0
    txg: 26
    pool_guid: 13455895622924169480
    hostid: 8413798
    hostname: ''weston''
    vdev_children: 1
    vdev_tree:
        type: ''root''
        id: 0
        guid: 13455895622924169480
        create_txg: 4
        children[0]:
            type: ''disk''
            id: 0
            guid: 7436238939623596891
            path: ''/dev/dsk/c5t1d0s0''
            devid: ''id1,sd at
SATA_____VBOX_HARDDISK____VBa371da65-169e72ea/a''
            phys_path: ''/pci at 0,0/pci8086,2829 at d/disk at
1,0:a''
            whole_disk: 1
            metaslab_array: 30
            metaslab_shift: 24
            ashift: 9
            asize: 3207856128
            is_log: 0
            create_txg: 4
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20100909/2a57debb/attachment-0001.html>

Hi Cindy,

I did see your first email pointing to that
bug<http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6538600>.
Apologies for not addressing it earlier.  It is my opinion that the behavior
Mike, and I <http://illumos.org/issues/217> (or anyone else upgrading
pools
right now) is seeing is a entirely new and different bug.  The bug you point
to, originally submitted in 2007 says it manifests itself before a reboot.
Also you say exporting and importing clear the problem.  After several
reboots, zdb still shows the older pool version, which means that this is a
new bug or perhaps the bug you are referencing is not listing clearly and
accurately what it should be and is incomplete.

Suppose an export and import can update the pool label config on a large
storage pool, great.  How would someone go about exporting the rpool the
operating system is on??  As far as I know, It''s impossible to export
the
zpool the operating system is running on.  I don''t think it can be
done, but
I''m new so maybe I''m missing something.

One option I have not explored that might work:  Booting to a live CD that
has the same or higher pool version present and then doing:   zpool import
&& zpool import -f rpool && zpool export rpool  and then
rebooting into the
operating system.  Perhaps this might be an option that "works" to
update
the label config / zdb for rpool but I think fixing the root problem would
be much more beneficial for everyone in the long run.  Being that zdb is a
troubleshooting/debugging tool, I would think that it''s necessary for
it to
be aware of the proper pool version to work properly and so admins know
what''s really going on with their pools.  The bottom line here is that
if
zdb is going to be part of zfs, it needs to display what is currently on
disk, including the label config.  If I were an admin thinking about
trusting hundreds of GB''s of data to zfs I would want the debugger to
show
me whats really on the disks.

Additionally, even though zpool and zfs "get version" display the true
and
updated versions, I''m not convinced that the problem is zdb, as the
label
config is almost certainly set by the zpool and/or zfs commands.  Somewhere,
something is not happening that is supposed to when initiating a zpool
upgrade, but since I know virtually nothing of the internals of zfs, I do
not know where.

Sincerely,

-Chris
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20100929/35fa47b1/attachment.html>

>Additionally, even though zpool and zfs "get version" display the
true and
>updated versions, I''m not convinced that the problem is zdb, as the
label
>config is almost certainly set by the zpool and/or zfs commands.  Somewhere,
>something is not happening that is supposed to when initiating a zpool
>upgrade, but since I know virtually nothing of the internals of zfs, I do
The problem is likely in the boot block or in grub.

The development version did not update the boot block; newer versions of 
beadm do fix boot blocks.

For now, I''d recommend you upgrade the boot block on all halves of a 
bootable mirror before you upgrade the zpool version or the zfs version.
export/import won''t help.

Casper

Well strangely enough, I just logged into a OS b145 machine.  It''s
rpool is
not mirrored, just a single disk.  I know that zdb reported zpool version 22
after at least the first 3 reboots after rpool upgrade, so I stopped
checking.  zdb now reports version 27.  This machine has probably been
rebooted about five or six times since the pool version upgrade.  One should
not have to reboot six times!  More mystery to this pool upgrade behavior!!

-Chris
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20100929/713fee3d/attachment.html>