[PVE-User] Old, 32bit, debian and clock drift.

[PVE-User] Old, 32bit, debian and clock drift.

[Marco Gaiarin]

PVE 6.4, updated to the latest patch.


I've P2V an old debian squeeze box, 32bit, that have an impressive clock
drift, even if they had 'ntpd' up&running.

I've tried to remove '/etc/adjtime' and reboot, but nothing changed.

The strange thing is i got in logs:

	Feb 10 15:15:37 sdinny kernel: [ 9481.004112] Clocksource tsc unstable (delta = 63065327 ns)

but:
	sdinny:~# cat /sys/devices/system/clocksource/clocksource0/available_clocksource 
	kvm-clock hpet acpi_pm 
	sdinny:~# cat /sys/devices/system/clocksource/clocksource0/current_clocksource 
	kvm-clock

So 'tsc' is even not listed on clocksources, nor are used.


What i'm missing?! Thanks.

-- 
  ``... La memoria conta veramente solo se tiene insieme l'impronta del
  presente e il progetto del futuro, se permette di fare senza dimenticare
  quel che si voleva fare, di diventare senza smettere di essere,
  di essere senza smettere di diventare...''		(Italo Calvino)

Re: [PVE-User] Old, 32bit, debian and clock drift.

[Marco Gaiarin]

Mandi! Arjen via pve-user
  In chel di` si favelave...

> Possibly both the para-virtualized clock source and the ntp daemon are adjusting for drift and therefore overcompensating and making everything worse.
> Try disabling ntpd, stop and start the VM and see if that works better? It works well for my VMs, but they are not running 24/7.

Ahem, the VM in question IS the NTP server for that site... but i've a
dozens of site like that, and this seems the only one with a clock drift...

-- 
   Di questa cavolo di pianura, di questa gente senza misura, che gia`
   confonde la notte e il giorno, e la partenza con il ritorno, e la
   richezza con i rumore, ed il diritto con il favore	(F. De Gregori)

Re: [PVE-User] Old, 32bit, debian and clock drift.

[Marco Gaiarin]

Mandi! Arjen via pve-user
  In chel di` si favelave...

> Sorry, I must have missed that. Other people also encountered your problem:
>  https://v13.gr/2016/02/15/running-an-ntp-server-in-a-vm-using-kvm/
> Instead of disabling ntpd, do not use the kvm-clock source on the NTP server VM.
> Either way, make sure not to combine the two.

OK, i've give it a try, for now, manually changing the clocksource with:

	echo 'hpet' > /sys/devices/system/clocksource/clocksource0/current_clocksource

But some things seems strange to me...

1) this happen on *TWO* VMs in two different installation; i've dozen of
similar installation where:

	root@vdmsv1:~# cat /sys/devices/system/clocksource/clocksource0/current_clocksource
	kvm-clock

and clock are perfectly in sync; true that these problematic VMs are pretty old debian,
but not older than others that works as expected.


2) other docs, like:
	https://docs.oracle.com/en/database/oracle/oracle-database/21/ladbi/setting-clock-source-vm.html

confirm that 'tsc' is the preferred clock source for VMs; but in these box
that depicted the trouble, 'TSC' clock source get 'kicked off' because
'unreilable'.


So, doing a little 'survey' on my VMs, seems that *all* VMs use by default
'kvm-clock' as the default clock source, but on those two VMs 'tsc' clock
source is unreilable, and get kicked off, and 'kvm-clock' is unreilable too,
but get not kicked off.


Really, really strange...

-- 
  Chi parla male, pensa male e vive male.
  Bisogna trovare le parole giuste: le parole sono importanti!
						Nanni Moretti in Palombella Rossa

Re: [PVE-User] Old, 32bit, debian and clock drift.

[Marco Gaiarin]

Mandi! Marco Gaiarin
  In chel di` si favelave...

> OK, i've give it a try, for now, manually changing the clocksource with:
>         echo 'hpet' > /sys/devices/system/clocksource/clocksource0/current_clocksource

No, also 'hpet' clocksource drift...

-- 
  Se non trovi nessuno vuol dire che siamo scappati alle sei-shell (bash,
  tcsh,csh...)							(Possi)