Lima VM clock appears to drift

[spacedub]

Description

I have a Debian VM that has been up for 6 days.

Reported date is:
Wed Jan 18 21:30:48 UTC 2023

macos reported date is:
Wed Jan 18 13:48:26 PST 2023

So, 18 minutes apart.

limactl version 0.11.3-27-g7a74ca9

I am using https://github.com/lima-vm/lima/blob/master/examples/debian.yaml slightly modified for my VM.

I assume installing ntp would solve this, but then (assuming this is right), should this not be by default?

[spacedub]

FWIW, sudo hwclock --hctosys is not helping.

Let me know if you need more info about my current rig.

[spacedub]

Also (obviously) confirming that installing ntp does help...

Installing ntp does not help...

dmp@lima-runner:~$ sudo apt-get install ntp
# Wrong date
dmp@lima-runner:~$ date
Wed Jan 18 21:39:57 UTC 2023
# Correct date, seems to be ntp date
dmp@lima-runner:~$ date
Wed Jan 18 21:57:36 UTC 2023
dmp@lima-runner:~$ date
Wed Jan 18 21:57:37 UTC 2023
# Reverting to the previous time after a minute...
dmp@lima-runner:~$ date
Wed Jan 18 21:40:53 UTC 2023

[afbjorklund]

You can check the source of the hwclock, from inside the VM (with lima):

cat /sys/devices/system/clocksource/clocksource0/current_clocksource

anders@lima-default:/home/anders$ cat /sys/devices/system/clocksource/clocksource0/current_clocksource
kvm-clock

[jandubois]

This is kind of a duplicate of #850 (note that I put a link to a blogpost about how Docker Desktop deals with this into a comment over there).

Also maybe related #939.

You didn't mention your host OS; are you on macOS or Linux?

[spacedub]

@jandubois macOS Ventura (M1 max)

@afbjorklund

dmp@lima-runner:~$ cat /sys/devices/system/clocksource/clocksource0/current_clocksource
arch_sys_counter

[spacedub]

@jandubois @afbjorklund I am completely out of my depth on this topic.
Is there a short-term workaround (even something really hack-ish) that I can use?

[afbjorklund]

I was surprised to hear that NTP wouldn't work (although coarsely), but it probably needs a local server (like DHCP uses)

Would be nice to have a HVF version of the KVM clocks - but it would probably require some modules on the Linux side...

So yeah, it sounds like the same issue as:

• Clockdrift in the created guest #850