[Qemu-devel] Occasional clockjump in Win2012 after Live Migration

[Qemu-devel] Occasional clockjump in Win2012 after Live Migration

[Peter Lieven]

Hi,

is anyone aware of a problem with a clock jump in Windows (observed in Server 2012)
where after a successful live migration the clock jumps roughly 2 days into the future?

Maybe this is already fixed we observed this with qemu-kvm-1.2.0. I have not yet managed
to reproduce this, but it definetly happens.

Where is Windows getting the system clock from? RTC or addtionally internal clocksources
like HPET, PM_TIMER etc?

Thanks,
Peter

Re: [Qemu-devel] Occasional clockjump in Win2012 after Live Migration

[Vadim Rozenfeld]

Does your VM belong to domain or workgroup? 
Best regards,
Vadim.

----- Original Message -----
From: "Peter Lieven" <pl@kamp.de>
To: qemu-devel@nongnu.org, vrozenfe@redhat.com
Sent: Friday, December 13, 2013 3:03:45 AM
Subject: Occasional clockjump in Win2012 after Live Migration

Hi,

is anyone aware of a problem with a clock jump in Windows (observed in Server 2012)
where after a successful live migration the clock jumps roughly 2 days into the future?

Maybe this is already fixed we observed this with qemu-kvm-1.2.0. I have not yet managed
to reproduce this, but it definetly happens.

Where is Windows getting the system clock from? RTC or addtionally internal clocksources
like HPET, PM_TIMER etc?

Thanks,
Peter

Re: [Qemu-devel] Occasional clockjump in Win2012 after Live Migration

[Peter Lieven]

Am 13.12.2013 05:12, schrieb Vadim Rozenfeld:
> Does your VM belong to domain or workgroup? 

We had 2 vServers where this happened. One was a Domain Controller and the second was an independent Workgroup Server.

Do you have evidence how the DateTime Clock is driven in Windows 2012?

Peter

Re: [Qemu-devel] Occasional clockjump in Win2012 after Live Migration

[Vadim Rozenfeld]

On Fri, 2013-12-13 at 09:27 +0100, Peter Lieven wrote:
> Am 13.12.2013 05:12, schrieb Vadim Rozenfeld:
> > Does your VM belong to domain or workgroup? 
> 
> We had 2 vServers where this happened. One was a Domain Controller and the second was an independent Workgroup Server.
> 
> Do you have evidence how the DateTime Clock is driven in Windows 2012?

Should be CMOS periodic timer.

> 
> Peter
> 

Re: [Qemu-devel] Occasional clockjump in Win2012 after Live Migration

[Peter Lieven]

Am 13.12.2013 11:10, schrieb Vadim Rozenfeld:
> On Fri, 2013-12-13 at 09:27 +0100, Peter Lieven wrote:
>> Am 13.12.2013 05:12, schrieb Vadim Rozenfeld:
>>> Does your VM belong to domain or workgroup? 
>> We had 2 vServers where this happened. One was a Domain Controller and the second was an independent Workgroup Server.
>>
>> Do you have evidence how the DateTime Clock is driven in Windows 2012?
> Should be CMOS periodic timer.
you mean the RTC?
Ok, so if there is a huge clock jump this one must have got messed up?

Peter

Re: [Qemu-devel] Occasional clockjump in Win2012 after Live Migration

[Vadim Rozenfeld]

On Fri, 2013-12-13 at 12:50 +0100, Peter Lieven wrote:
> Am 13.12.2013 11:10, schrieb Vadim Rozenfeld:
> > On Fri, 2013-12-13 at 09:27 +0100, Peter Lieven wrote:
> >> Am 13.12.2013 05:12, schrieb Vadim Rozenfeld:
> >>> Does your VM belong to domain or workgroup? 
> >> We had 2 vServers where this happened. One was a Domain Controller and the second was an independent Workgroup Server.
> >>
> >> Do you have evidence how the DateTime Clock is driven in Windows 2012?
> > Should be CMOS periodic timer.
> you mean the RTC?
Right.
> Ok, so if there is a huge clock jump this one must have got messed up?
It might be. We have seen some similar problem during WHQL testing some
time ago. The problem happened because time on the host was not set
correct. So guest was taking this wrong time on boot and then it was a
big jump after resync with domain controller.

best regards,
Vadim.   
> 
> Peter
> 

Re: [Qemu-devel] Occasional clockjump in Win2012 after Live Migration

[Peter Lieven]

Am 13.12.2013 13:53, schrieb Vadim Rozenfeld:
> On Fri, 2013-12-13 at 12:50 +0100, Peter Lieven wrote:
>> Am 13.12.2013 11:10, schrieb Vadim Rozenfeld:
>>> On Fri, 2013-12-13 at 09:27 +0100, Peter Lieven wrote:
>>>> Am 13.12.2013 05:12, schrieb Vadim Rozenfeld:
>>>>> Does your VM belong to domain or workgroup? 
>>>> We had 2 vServers where this happened. One was a Domain Controller and the second was an independent Workgroup Server.
>>>>
>>>> Do you have evidence how the DateTime Clock is driven in Windows 2012?
>>> Should be CMOS periodic timer.
>> you mean the RTC?
> Right.
>> Ok, so if there is a huge clock jump this one must have got messed up?
> It might be. We have seen some similar problem during WHQL testing some
> time ago. The problem happened because time on the host was not set
> correct. So guest was taking this wrong time on boot and then it was a
> big jump after resync with domain controller.
our system forbids starting of vservers on a host where the time is not synched. ;-)

the big jump i see seems to be introduced by the live migration. i will investigate further.

Peter