zfs discuss - Jul 2007 - ZFS receive issue running multiple receives and rollbacks

Hi, all,

Environment: S10U3 running as VMWare Workstation 6 guest; Fedora 7 is
the VMWare host, 1 GB RAM

I''m creating a solution in which I need to be able to save off state on
one host, then restore it on another. I''m using ZFS snapshots with ZFS
receive and it''s all working fine, except for some strange behavior
when
I perform multiple rollbacks and receives.

Here''s what I''m seeing:

- On the first host (a VMWare virtual machine, actually), I create an
initial snapshot
- I modify the state of the ZFS file system
- I create a second snapshot
- I perform a zfs send with the -i argument between the two snapshots.
Size of incremental diffs file is around 1.3 GB.
- On the second host (also a VMWare virtual machine, which is a copy of
the first host), I perform zfs receive
- The receive performs correctly in around 3 to 3 1/2 minutes, and at
the end of it, the state of the second host is identical to the state of
the first host

Now, when I perform the following steps:

- Restore the state of the second host to the initial state (using zfs
rollback -r snapshotname)
- Run zfs receive for a second time on the second host

Now the second host appears to lock up. I wait half an hour and the zfs
receive command has not completed. I try to terminate the command with
cntl-c and I get no response.

But if I take the following action:

- Open a second terminal window on the second host
- Power off the second host (reboot might be enough here but since it''s
VMWare, poweroff is easy enough)
- Restart the second host
- Restore the state of the second host to the initial state (using zfs
rollback -r snapshotname)
- Run zfs receive on the second host

Now the state of the second host is restored correctly. The zfs receive
takes around 3 to 3/12 minutes.

So, is there something I need to do/run on S10 that will let me run zfs
receive for the second time without having to restart the OS?

Thanks,

David

-- 
David Goldsmith
Course Developer
Sun Identity Management Suite
Sun Learning Services
Voice: (415) 375-8236 (inside Sun: x81217)
E-mail: David.Goldsmith at sun.com
Blog: http://blogs.sun.com/openroad

David Goldsmith wrote:
> - Restore the state of the second host to the initial state (using zfs
> rollback -r snapshotname)
> - Run zfs receive for a second time on the second host
> 
> Now the second host appears to lock up. I wait half an hour and the zfs
> receive command has not completed. I try to terminate the command with
> cntl-c and I get no response.
Just a guess here, but perhaps you''re running 32-bit and running out of
address space.  Or 1GB of memory is not enough.  Is that 1GB for each solaris 
virtual machine, or 1GB total for linux + 2x solaris?

--matt

Hi, Matt,

1. My VMWare host has 4 GB. The VMWare guest (Solaris 10) has 1 GB. I
think that at one point I reset the guest to have 2 GB and ran into the
same problem, but I''m not 100% sure. If you think it''s worth
trying, I
will.

2. I''m running S10U3 as 32-bit. I don''t know if I can run
64-bit Solaris
10 with 32-bit Linux as the host OS. Does anyone know if that will work?
If so, I''ll give it a shot.

Still, for the machine to just stall like that strikes me as
problematic. Is there anywhere I could be looking for diagnostics, or
any zfs receive command options I could use to get an idea of what''s
going on?

Thanks,

David

Matthew Ahrens wrote:
> David Goldsmith wrote:
>> - Restore the state of the second host to the initial state (using zfs
>> rollback -r snapshotname)
>> - Run zfs receive for a second time on the second host
>>
>> Now the second host appears to lock up. I wait half an hour and the zfs
>> receive command has not completed. I try to terminate the command with
>> cntl-c and I get no response.
>
> Just a guess here, but perhaps you''re running 32-bit and running
out
> of address space.  Or 1GB of memory is not enough.  Is that 1GB for
> each solaris virtual machine, or 1GB total for linux + 2x solaris?
>
> --matt

-- 
David Goldsmith
Course Developer
Sun Identity Management Suite
Sun Learning Services
Voice: (415) 375-8236 (inside Sun: x81217)
E-mail: David.Goldsmith at sun.com
Blog: http://blogs.sun.com/openroad

On 7/5/07, David Goldsmith <David.Goldsmith at sun.com>
wrote:
> 2. I''m running S10U3 as 32-bit. I don''t know if I can run
64-bit Solaris
> 10 with 32-bit Linux as the host OS. Does anyone know if that will work?
> If so, I''ll give it a shot.
ISTR that if you have hardware virtualization (Intel VT, on Core 2 Duo
6*** chips, or AMD''s equivalent technology) you can indeed do that.
See http://www.vmware.com/pdf/processor_check.pdf .

Will

Thanks, Will, but for the solution I''m building, I can''t
predict the
hardware the VM will run on, and I don''t want to restrict it to the
limited list in the Processor Check document. So unfortunately I think
I''m stuck running 32-bit Solaris for this one.

So it would be nice to have zfs not hang (or give the appearance of
hanging) in the middle of a receive. (Seems like a bug to me that it
does that but I still have much to learn about ZFS.)  Even a message
that said "You need to reboot your system before you can do this"
would
be preferable. Heck, it would at least make the Windows user running
this solution feel right at home. :-)

Thanks,

David

Will Murnane wrote:
> On 7/5/07, David Goldsmith <David.Goldsmith at sun.com> wrote:
>> 2. I''m running S10U3 as 32-bit. I don''t know if I can
run 64-bit Solaris
>> 10 with 32-bit Linux as the host OS. Does anyone know if that will
work?
>> If so, I''ll give it a shot.
> ISTR that if you have hardware virtualization (Intel VT, on Core 2 Duo
> 6*** chips, or AMD''s equivalent technology) you can indeed do
that.
> See http://www.vmware.com/pdf/processor_check.pdf .
>
> Will

-- 
David Goldsmith
Course Developer
Sun Identity Management Suite
Sun Learning Services
Voice: (415) 375-8236 (inside Sun: x81217)
E-mail: David.Goldsmith at sun.com
Blog: http://blogs.sun.com/openroad