Hello,
I have clock synchronization problems with Fedora 18/KVM host and windows server guests. Different 64bit windows servers, all has some time synchronization problems.
Some of them will slow down time until time difference is 60 seconds and then clock is synchronizes from internet. But 60 seconds is a long time. Kerberos authentication sometimes files due to time skew.
Some machines will run OK day or two, but then clock stops. After some hours clock is synchronized again and continues to work properly. Curious, that during time is stopped, server works properly (except kerberos authentications, which is very time sensitive.
For example, attaching 2 graphs from 2 windows servers. They show time difference measured by check_mk nagios plugin. 10k means 10000 seconds.
How I should set clock/time/NTP on these guests?
I have absolutelly no problems with Linux guests, they work perfectly.
Thank you. SAL
On 06/24/2013 02:18 PM, Ján ONDREJ (SAL) wrote:
Hello,
I have clock synchronization problems with Fedora 18/KVM host and windows server guests. Different 64bit windows servers, all has some time synchronization problems.
Some of them will slow down time until time difference is 60 seconds and then clock is synchronizes from internet. But 60 seconds is a long time. Kerberos authentication sometimes files due to time skew.
Some machines will run OK day or two, but then clock stops. After some hours clock is synchronized again and continues to work properly. Curious, that during time is stopped, server works properly (except kerberos authentications, which is very time sensitive.
For example, attaching 2 graphs from 2 windows servers. They show time difference measured by check_mk nagios plugin. 10k means 10000 seconds.
How I should set clock/time/NTP on these guests?
I have absolutelly no problems with Linux guests, they work perfectly.
Maybe try altering the guest config with:
<clock offset='utc'> <timer name='pit' tickpolicy='delay'/> <timer name='rtc' tickpolicy='catchup'/> </clock>
https://bugs.launchpad.net/nova/+bug/1011848 https://bugzilla.redhat.com/show_bug.cgi?id=828332
virt-manager/virt-install should be doing this by default.
On Tue, Jun 25, 2013 at 3:01 PM, Cole Robinson crobinso@redhat.com wrote:
On 06/24/2013 02:18 PM, Ján ONDREJ (SAL) wrote:
Hello,
I have clock synchronization problems with Fedora 18/KVM host and
windows
server guests. Different 64bit windows servers, all has some time synchronization problems.
Some of them will slow down time until time difference is 60 seconds
and
then clock is synchronizes from internet. But 60 seconds is a long time. Kerberos authentication sometimes files due to time skew.
Some machines will run OK day or two, but then clock stops. After some hours clock is synchronized again and continues to work properly. Curious, that during time is stopped, server works properly (except
kerberos
authentications, which is very time sensitive.
For example, attaching 2 graphs from 2 windows servers. They show time difference measured by check_mk nagios plugin. 10k means 10000 seconds.
How I should set clock/time/NTP on these guests?
I have absolutelly no problems with Linux guests, they work perfectly.
Maybe try altering the guest config with:
<clock offset='utc'> <timer name='pit' tickpolicy='delay'/> <timer name='rtc' tickpolicy='catchup'/> </clock>https://bugs.launchpad.net/nova/+bug/1011848 https://bugzilla.redhat.com/show_bug.cgi?id=828332
virt-manager/virt-install should be doing this by default.
virt mailing list virt@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/virt
Have you checked the clock of the hypervisor the guest is running on?
Hello,
On Tue, Jun 25, 2013 at 03:17:06PM -0400, Jamie Fargen wrote:
On Tue, Jun 25, 2013 at 3:01 PM, Cole Robinson crobinso@redhat.com wrote:
On 06/24/2013 02:18 PM, Ján ONDREJ (SAL) wrote: Maybe try altering the guest config with:
<clock offset='utc'> <timer name='pit' tickpolicy='delay'/> <timer name='rtc' tickpolicy='catchup'/> </clock>
I have to test it longer time on multiple machines, but I still see clock skew between -3s to +0.5s in my guest. I have no time guest time synchronization now (NTPd and timeserver are turned off in windows).
Can I have exact time in windows? Testing on windows 2008R2.
https://bugs.launchpad.net/nova/+bug/1011848 https://bugzilla.redhat.com/show_bug.cgi?id=828332
virt-manager/virt-install should be doing this by default.
This is still not default in Fedora 18. My packages: virt-manager-0.9.5-1.fc18.noarch libvirt-0.10.2.6-1.fc18.x86_64
Byt, my settings are deprecated. In qemu log I see: Warning: option deprecated, use lost_tick_policy property of kvm-pit instead.
Have you checked the clock of the hypervisor the guest is running on?
Sure, host is synchronized using ntpd.
SAL
On 06/26/2013 05:27 AM, Ján ONDREJ (SAL) wrote:
Hello,
On Tue, Jun 25, 2013 at 03:17:06PM -0400, Jamie Fargen wrote:
On Tue, Jun 25, 2013 at 3:01 PM, Cole Robinson crobinso@redhat.com wrote:
On 06/24/2013 02:18 PM, Ján ONDREJ (SAL) wrote: Maybe try altering the guest config with:
<clock offset='utc'> <timer name='pit' tickpolicy='delay'/> <timer name='rtc' tickpolicy='catchup'/> </clock>I have to test it longer time on multiple machines, but I still see clock skew between -3s to +0.5s in my guest. I have no time guest time synchronization now (NTPd and timeserver are turned off in windows).
Can I have exact time in windows? Testing on windows 2008R2.
https://bugs.launchpad.net/nova/+bug/1011848 https://bugzilla.redhat.com/show_bug.cgi?id=828332
virt-manager/virt-install should be doing this by default.
This is still not default in Fedora 18. My packages: virt-manager-0.9.5-1.fc18.noarch libvirt-0.10.2.6-1.fc18.x86_64
Sorry, that sentence was meant to read like: 'virt-manager/virt-install does not currently set that XML, but it should be.'
Byt, my settings are deprecated. In qemu log I see: Warning: option deprecated, use lost_tick_policy property of kvm-pit instead.
Hmm. Can you file a fedora libvirt bug with your guest XML, and the contents of /var/log/libvirt/qemu/$vmname.log ?
Thanks, Cole
Hello,
On Tue, Jun 25, 2013 at 03:01:15PM -0400, Cole Robinson wrote:
On 06/24/2013 02:18 PM, Ján ONDREJ (SAL) wrote:
Hello,
I have clock synchronization problems with Fedora 18/KVM host and windows server guests. Different 64bit windows servers, all has some time synchronization problems.
Some of them will slow down time until time difference is 60 seconds and then clock is synchronizes from internet. But 60 seconds is a long time. Kerberos authentication sometimes files due to time skew.
Some machines will run OK day or two, but then clock stops. After some hours clock is synchronized again and continues to work properly. Curious, that during time is stopped, server works properly (except kerberos authentications, which is very time sensitive.
For example, attaching 2 graphs from 2 windows servers. They show time difference measured by check_mk nagios plugin. 10k means 10000 seconds.
How I should set clock/time/NTP on these guests?
I have absolutelly no problems with Linux guests, they work perfectly.
Maybe try altering the guest config with:
<clock offset='utc'> <timer name='pit' tickpolicy='delay'/> <timer name='rtc' tickpolicy='catchup'/> </clock>https://bugs.launchpad.net/nova/+bug/1011848 https://bugzilla.redhat.com/show_bug.cgi?id=828332
virt-manager/virt-install should be doing this by default.
These didn't helped me to fix my problem. Also trying combinations of these without success:
<clock offset='variable' adjustment='2' basis='utc'> <timer name='rtc' tickpolicy='catchup'/> <timer name='pit' tickpolicy='delay'/> <timer name='hpet' present='yes'/> </clock> qemu:commandline <qemu:arg value='-rtc-td-hack'/> </qemu:commandline>
Any ideas? Is this a bug?
SAL
On 07/03/2013 01:54 AM, "Ján ONDREJ (SAL)" wrote:
Hello,
On Tue, Jun 25, 2013 at 03:01:15PM -0400, Cole Robinson wrote:
On 06/24/2013 02:18 PM, Ján ONDREJ (SAL) wrote:
Hello,
I have clock synchronization problems with Fedora 18/KVM host and windows server guests. Different 64bit windows servers, all has some time synchronization problems.
Some of them will slow down time until time difference is 60 seconds and then clock is synchronizes from internet. But 60 seconds is a long time. Kerberos authentication sometimes files due to time skew.
Some machines will run OK day or two, but then clock stops. After some hours clock is synchronized again and continues to work properly. Curious, that during time is stopped, server works properly (except kerberos authentications, which is very time sensitive.
For example, attaching 2 graphs from 2 windows servers. They show time difference measured by check_mk nagios plugin. 10k means 10000 seconds.
How I should set clock/time/NTP on these guests?
I have absolutelly no problems with Linux guests, they work perfectly.
Maybe try altering the guest config with:
<clock offset='utc'> <timer name='pit' tickpolicy='delay'/> <timer name='rtc' tickpolicy='catchup'/> </clock>https://bugs.launchpad.net/nova/+bug/1011848 https://bugzilla.redhat.com/show_bug.cgi?id=828332
virt-manager/virt-install should be doing this by default.
These didn't helped me to fix my problem. Also trying combinations of these without success:
<clock offset='variable' adjustment='2' basis='utc'> <timer name='rtc' tickpolicy='catchup'/> <timer name='pit' tickpolicy='delay'/> <timer name='hpet' present='yes'/> </clock> <qemu:commandline> <qemu:arg value='-rtc-td-hack'/> </qemu:commandline>
Any ideas? Is this a bug?
Possibly, please file a Fedora qemu bug and list all the info you've given here.
Thanks, Cole
On Wed, 03 Jul 2013 11:08:48 -0400 Cole Robinson wrote:
Any ideas? Is this a bug?
Possibly, please file a Fedora qemu bug and list all the info you've given here.
I certainly know that no virtual machine I've ever installed has been able to keep time worth spit using any kind of emulated hardware. (Some older ones are so bad networking goes down because the time is so far off they forget to renew their DHCP lease).
Time never made sense till newer versions of linux got virtualized clocks that actually knew they ought to get time from the host.
I don't suppose anyone is working on a special clock driver for windows?
The vmware web site has lots of info about getting time in virt machine to work better, and a lot of the advice worked for me to make time a bit better under KVM as well.
On Wed, Jul 03, 2013 at 11:39:14AM -0400, Tom Horsley wrote:
On Wed, 03 Jul 2013 11:08:48 -0400 Cole Robinson wrote:
Any ideas? Is this a bug?
Possibly, please file a Fedora qemu bug and list all the info you've given here.
I certainly know that no virtual machine I've ever installed has been able to keep time worth spit using any kind of emulated hardware. (Some older ones are so bad networking goes down because the time is so far off they forget to renew their DHCP lease).
Time never made sense till newer versions of linux got virtualized clocks that actually knew they ought to get time from the host.
I don't suppose anyone is working on a special clock driver for windows?
The vmware web site has lots of info about getting time in virt machine to work better, and a lot of the advice worked for me to make time a bit better under KVM as well.
What does this mean? Can I use KVM as host for windows server guests?
If yes, I need to make clock at least a bit in sync. My current deltas are from 0 seconds to 30000 seconds. More tha one hour skew is not acceptable for servers. Also I am unable to synchronize windows guest time using NTP. It's still not in sync.
Why there are so many of clock parameters and their suggestions, if there is nothing I can do to at least be aproximatelly in sync (<3 seconds skew).
Thank you for reply.
SAL
Hello,
On Wed, Jul 03, 2013 at 11:08:48AM -0400, Cole Robinson wrote:
On 07/03/2013 01:54 AM, "Ján ONDREJ (SAL)" wrote:
Hello,
On Tue, Jun 25, 2013 at 03:01:15PM -0400, Cole Robinson wrote:
On 06/24/2013 02:18 PM, Ján ONDREJ (SAL) wrote:
Hello,
I have clock synchronization problems with Fedora 18/KVM host and windows server guests. Different 64bit windows servers, all has some time synchronization problems.
Some of them will slow down time until time difference is 60 seconds and then clock is synchronizes from internet. But 60 seconds is a long time. Kerberos authentication sometimes files due to time skew.
Some machines will run OK day or two, but then clock stops. After some hours clock is synchronized again and continues to work properly. Curious, that during time is stopped, server works properly (except kerberos authentications, which is very time sensitive.
For example, attaching 2 graphs from 2 windows servers. They show time difference measured by check_mk nagios plugin. 10k means 10000 seconds.
How I should set clock/time/NTP on these guests?
I have absolutelly no problems with Linux guests, they work perfectly.
Maybe try altering the guest config with:
<clock offset='utc'> <timer name='pit' tickpolicy='delay'/> <timer name='rtc' tickpolicy='catchup'/> </clock>https://bugs.launchpad.net/nova/+bug/1011848 https://bugzilla.redhat.com/show_bug.cgi?id=828332
virt-manager/virt-install should be doing this by default.
These didn't helped me to fix my problem. Also trying combinations of these without success:
<clock offset='variable' adjustment='2' basis='utc'> <timer name='rtc' tickpolicy='catchup'/> <timer name='pit' tickpolicy='delay'/> <timer name='hpet' present='yes'/> </clock> <qemu:commandline> <qemu:arg value='-rtc-td-hack'/> </qemu:commandline>
Any ideas? Is this a bug?
Possibly, please file a Fedora qemu bug and list all the info you've given here.
Looks like there are 2 different clock problems and looks like I can identify, when they appear.
I have 2 types of machines, 1st with virtio drivers and 2nd without virtio drivers (default hardware configurations).
1st type - virtio drivers installed, version virtio-win-0.1-59.iso (or 52) - clock sometimes stops working completelly. I see more hours same time on machine. - screenshot image-virtio.png attached
2nd type - IDE disk and default network (e1000?) - clock skew is 0-80 seconds. It's better like for 1st type. - screenshot image-novirtio.png attached
Looks like there are some problems with virtio drivers, which stops clock. Now I will try, if I can avoid problems with only one virtio driver (net or disk).
Are ther commercial customers using these drivers? Also have problems?
SAL
On 07/04/2013 02:03 AM, "Ján ONDREJ (SAL)" wrote:
Hello,
On Wed, Jul 03, 2013 at 11:08:48AM -0400, Cole Robinson wrote:
On 07/03/2013 01:54 AM, "Ján ONDREJ (SAL)" wrote:
Hello,
On Tue, Jun 25, 2013 at 03:01:15PM -0400, Cole Robinson wrote:
On 06/24/2013 02:18 PM, Ján ONDREJ (SAL) wrote:
Hello,
I have clock synchronization problems with Fedora 18/KVM host and windows server guests. Different 64bit windows servers, all has some time synchronization problems.
Some of them will slow down time until time difference is 60 seconds and then clock is synchronizes from internet. But 60 seconds is a long time. Kerberos authentication sometimes files due to time skew.
Some machines will run OK day or two, but then clock stops. After some hours clock is synchronized again and continues to work properly. Curious, that during time is stopped, server works properly (except kerberos authentications, which is very time sensitive.
For example, attaching 2 graphs from 2 windows servers. They show time difference measured by check_mk nagios plugin. 10k means 10000 seconds.
How I should set clock/time/NTP on these guests?
I have absolutelly no problems with Linux guests, they work perfectly.
Maybe try altering the guest config with:
<clock offset='utc'> <timer name='pit' tickpolicy='delay'/> <timer name='rtc' tickpolicy='catchup'/> </clock>https://bugs.launchpad.net/nova/+bug/1011848 https://bugzilla.redhat.com/show_bug.cgi?id=828332
virt-manager/virt-install should be doing this by default.
These didn't helped me to fix my problem. Also trying combinations of these without success:
<clock offset='variable' adjustment='2' basis='utc'> <timer name='rtc' tickpolicy='catchup'/> <timer name='pit' tickpolicy='delay'/> <timer name='hpet' present='yes'/> </clock> <qemu:commandline> <qemu:arg value='-rtc-td-hack'/> </qemu:commandline>
Any ideas? Is this a bug?
Possibly, please file a Fedora qemu bug and list all the info you've given here.
Looks like there are 2 different clock problems and looks like I can identify, when they appear.
I have 2 types of machines, 1st with virtio drivers and 2nd without virtio drivers (default hardware configurations).
1st type - virtio drivers installed, version virtio-win-0.1-59.iso (or 52)
- clock sometimes stops working completelly. I see more hours same time on machine.
- screenshot image-virtio.png attached
2nd type - IDE disk and default network (e1000?)
- clock skew is 0-80 seconds. It's better like for 1st type.
- screenshot image-novirtio.png attached
Looks like there are some problems with virtio drivers, which stops clock. Now I will try, if I can avoid problems with only one virtio driver (net or disk).
Are ther commercial customers using these drivers? Also have problems?
Best way to get some response for these questions is to file a separate bug against the virtio-win component in Fedora, where all the virtio-win guys are watching.
Thanks, Cole
On Mon, Jul 08, 2013 at 01:37:01PM -0400, Cole Robinson wrote:
On 07/04/2013 02:03 AM, "Ján ONDREJ (SAL)" wrote:
Hello,
On Wed, Jul 03, 2013 at 11:08:48AM -0400, Cole Robinson wrote:
On 07/03/2013 01:54 AM, "Ján ONDREJ (SAL)" wrote:
Hello,
On Tue, Jun 25, 2013 at 03:01:15PM -0400, Cole Robinson wrote:
On 06/24/2013 02:18 PM, Ján ONDREJ (SAL) wrote:
Hello,
I have clock synchronization problems with Fedora 18/KVM host and windows server guests. Different 64bit windows servers, all has some time synchronization problems.
Some of them will slow down time until time difference is 60 seconds and then clock is synchronizes from internet. But 60 seconds is a long time. Kerberos authentication sometimes files due to time skew.
Some machines will run OK day or two, but then clock stops. After some hours clock is synchronized again and continues to work properly. Curious, that during time is stopped, server works properly (except kerberos authentications, which is very time sensitive.
For example, attaching 2 graphs from 2 windows servers. They show time difference measured by check_mk nagios plugin. 10k means 10000 seconds.
How I should set clock/time/NTP on these guests?
I have absolutelly no problems with Linux guests, they work perfectly.
Maybe try altering the guest config with:
<clock offset='utc'> <timer name='pit' tickpolicy='delay'/> <timer name='rtc' tickpolicy='catchup'/> </clock>https://bugs.launchpad.net/nova/+bug/1011848 https://bugzilla.redhat.com/show_bug.cgi?id=828332
virt-manager/virt-install should be doing this by default.
These didn't helped me to fix my problem. Also trying combinations of these without success:
<clock offset='variable' adjustment='2' basis='utc'> <timer name='rtc' tickpolicy='catchup'/> <timer name='pit' tickpolicy='delay'/> <timer name='hpet' present='yes'/> </clock> <qemu:commandline> <qemu:arg value='-rtc-td-hack'/> </qemu:commandline>
Any ideas? Is this a bug?
Possibly, please file a Fedora qemu bug and list all the info you've given here.
Looks like there are 2 different clock problems and looks like I can identify, when they appear.
I have 2 types of machines, 1st with virtio drivers and 2nd without virtio drivers (default hardware configurations).
1st type - virtio drivers installed, version virtio-win-0.1-59.iso (or 52)
- clock sometimes stops working completelly. I see more hours same time on machine.
- screenshot image-virtio.png attached
2nd type - IDE disk and default network (e1000?)
- clock skew is 0-80 seconds. It's better like for 1st type.
- screenshot image-novirtio.png attached
Looks like there are some problems with virtio drivers, which stops clock. Now I will try, if I can avoid problems with only one virtio driver (net or disk).
Are ther commercial customers using these drivers? Also have problems?
Best way to get some response for these questions is to file a separate bug against the virtio-win component in Fedora, where all the virtio-win guys are watching.
Thanks, done.
https://bugzilla.redhat.com/show_bug.cgi?id=982459
Btw, it's a network driver, which causes this. With disk driver I have no problems.
Now I am switching my guests to e1000 driver to workaround this problem.
SAL
-
Cole Robinson -
Jamie Fargen -
Ján ONDREJ (SAL) -
Tom Horsley