Re: [chrony-users] Validating PHC operation on Microsoft VM

2025-03-04 Thread Mingye Wang
The hypervisor is fine -- in the sense that it's allowing guest
programs to run, and the frequency isn't too variable for chrony to
keep its own time from NTP. It's just the host time is a bit off, but
as you can see chrony is correctly not using it by marking it as '?'
(unusable, check sources -v), so you end up getting the right time
from Google.

PS: Your image is well received.
--
Mingye Wang (Artoria2e5)

On Wed, Feb 26, 2025 at 12:40 AM Steven Sommars
 wrote:
>
>
>> If you keep the refclock enabled with the nearby NTP sources, will it
>> be marked as a falseticker?
>
>
> PHC (noselect) plus 4 of the Google anycast
> chronyc -n sources;chronyc -n sourcestats
> MS Name/IP address Stratum Poll Reach LastRx Last sample
> ===
> #? PHC0  0   3   377 5+29ms[  +29ms] +/-   
> 17us
> ^+ 216.239.35.0  1   6   37752   -512us[ -512us] +/- 
> 5576us
> ^* 216.239.35.12 1   6   37753-80us[  -75us] +/- 
> 5068us
> ^- 216.239.35.4  1   6   37750   +566us[ +566us] +/-   
> 18ms
> ^- 216.239.35.8  1   6   37759   +672us[ +677us] +/-   
> 18ms
> Name/IP AddressNP  NR  Span  Frequency  Freq Skew  Offset  Std Dev
> ==
> PHC06   340 -8.207 10.836+29ms39us
> 216.239.35.0   19  12  1162 -0.074  1.127-60us   415us
> 216.239.35.12  64  38   67m +0.000  0.101-46us   281us
> 216.239.35.4   64  35   67m +0.020  0.199+46us   552us
> 216.239.35.8   64  30   68m +0.053  0.172  -8862ns   510us
>
> Looks like a hypervisor problem.
>

--
To unsubscribe email [email protected]
with "unsubscribe" in the subject.
For help email [email protected]
with "help" in the subject.
Trouble?  Email [email protected].



Re: [chrony-users] Validating PHC operation on Microsoft VM

2025-02-26 Thread Steven Sommars
I synchronized the host using chrony and nearby stratum 1's and then
compared the PHC time to the host time.
The results are bizarre.   I have only oneVM, so this result might not hold
elsewhere.
Miroslav suggests this might be a hyper-v issue.

Hope this image gets through the mailing list.  If not, I'll post it
elsewhere.
[image: image.png]

On Wed, Feb 26, 2025 at 2:22 AM Miroslav Lichvar 
wrote:

> On Wed, Feb 26, 2025 at 12:59:20PM +0800, Mingye Wang wrote:
> > I can see why Microsoft Azure picked that option: they don't want to
> > overload the NTP pool (like I've complained about recently)[1], but
> > their approach is a bit off. Specifically, I believe they should have
> > specified a nonzero stratum so that chrony does not place too much
> > trust in the host time. Is there anyone from Azure that I can contact
> > on this issue?
>
> If there is only one source, it doesn't matter what stratum it has.
> Its weight can change only relative to other sources.
>
> If you want to propose changes to the default configuration, you will
> need to find out where it is coming from. There could be a base image
> provided by the OS vendor and other changes layered on top maintained
> elsewhere. Not easy to follow. I think most of the Azure stuff is on
> github.
>
> --
> Miroslav Lichvar
>
>
> --
> To unsubscribe email [email protected]
> with "unsubscribe" in the subject.
> For help email [email protected]
> with "help" in the subject.
> Trouble?  Email [email protected].
>
>


Re: [chrony-users] Validating PHC operation on Microsoft VM

2025-02-26 Thread Miroslav Lichvar
On Wed, Feb 26, 2025 at 12:59:20PM +0800, Mingye Wang wrote:
> I can see why Microsoft Azure picked that option: they don't want to
> overload the NTP pool (like I've complained about recently)[1], but
> their approach is a bit off. Specifically, I believe they should have
> specified a nonzero stratum so that chrony does not place too much
> trust in the host time. Is there anyone from Azure that I can contact
> on this issue?

If there is only one source, it doesn't matter what stratum it has.
Its weight can change only relative to other sources.

If you want to propose changes to the default configuration, you will
need to find out where it is coming from. There could be a base image
provided by the OS vendor and other changes layered on top maintained
elsewhere. Not easy to follow. I think most of the Azure stuff is on
github.

-- 
Miroslav Lichvar


-- 
To unsubscribe email [email protected] 
with "unsubscribe" in the subject.
For help email [email protected] 
with "help" in the subject.
Trouble?  Email [email protected].



Re: [chrony-users] Validating PHC operation on Microsoft VM

2025-02-25 Thread Mingye Wang
On Tue, Feb 25, 2025 at 11:57 AM Steven Sommars
 wrote:
> refclock PHC /dev/ptp_hyperv poll 3 dpoll -2 offset 0

I can see why Microsoft Azure picked that option: they don't want to
overload the NTP pool (like I've complained about recently)[1], but
their approach is a bit off. Specifically, I believe they should have
specified a nonzero stratum so that chrony does not place too much
trust in the host time. Is there anyone from Azure that I can contact
on this issue?

[1]: 
https://community.ntppool.org/t/rant-some-of-our-clients-shouldnt-even-exist-vm-farms/3774
--
Mingye Wang (Artoria2e5)

--
To unsubscribe email [email protected]
with "unsubscribe" in the subject.
For help email [email protected]
with "help" in the subject.
Trouble?  Email [email protected].



Re: [chrony-users] Validating PHC operation on Microsoft VM

2025-02-25 Thread Steven Sommars
> If you keep the refclock enabled with the nearby NTP sources, will it
> be marked as a falseticker?
>

PHC (noselect) plus 4 of the Google anycast
chronyc -n sources;chronyc -n sourcestats
MS Name/IP address Stratum Poll Reach LastRx Last sample
===
#? PHC0  0   3   377 5+29ms[  +29ms] +/-
17us
^+ 216.239.35.0  1   6   37752   -512us[ -512us] +/-
5576us
^* 216.239.35.12 1   6   37753-80us[  -75us] +/-
5068us
^- 216.239.35.4  1   6   37750   +566us[ +566us] +/-
18ms
^- 216.239.35.8  1   6   37759   +672us[ +677us] +/-
18ms
Name/IP AddressNP  NR  Span  Frequency  Freq Skew  Offset  Std
Dev
==
PHC06   340 -8.207 10.836+29ms
 39us
216.239.35.0   19  12  1162 -0.074  1.127-60us
415us
216.239.35.12  64  38   67m +0.000  0.101-46us
281us
216.239.35.4   64  35   67m +0.020  0.199+46us
552us
216.239.35.8   64  30   68m +0.053  0.172  -8862ns
510us

Looks like a hypervisor problem.


Re: [chrony-users] Validating PHC operation on Microsoft VM

2025-02-24 Thread Miroslav Lichvar
On Mon, Feb 24, 2025 at 09:57:21PM -0600, Steven Sommars wrote:
> AFAICT chrony is happy, but is providing inaccurate time: Time is in error
> by 30-50 msec with substantial jitter.
> 
> Dropping the PHC refclock and using nearby stratum 1's give much better
> results.

If you keep the refclock enabled with the nearby NTP sources, will it
be marked as a falseticker?

> Any suggestions for further debugging?

To me that sounds like the problem is in the synchronization of the
VM host's clock. An issue with the hyperv clocksource would impact
both PHC and NTP sources.

-- 
Miroslav Lichvar


-- 
To unsubscribe email [email protected] 
with "unsubscribe" in the subject.
For help email [email protected] 
with "help" in the subject.
Trouble?  Email [email protected].



[chrony-users] Validating PHC operation on Microsoft VM

2025-02-24 Thread Steven Sommars
I'm running chrony 4.5 (which is the current default) on a commercial
Microsoft Azure VM (Linux myVm 6.8.0-1021-azure).  By default a PTP
hardware clock is used as a time source:
chrony.conf has no servers defined

refclock PHC /dev/ptp_hyperv poll 3 dpoll -2 offset 0

No external servers are defined.  This is the default configuration.

Current time source:  hyperv_clocksource_tsc_page
Available time sources: hyperv_clocksource_tsc_page hyperv_clocksource_msr
acpi_pm
This is the default setting.

AFAICT chrony is happy, but is providing inaccurate time: Time is in error
by 30-50 msec with substantial jitter.

Dropping the PHC refclock and using nearby stratum 1's give much better
results.

Any suggestions for further debugging?