Here is an update on the situation:
I installed -current on this VM, clean install, and the ntpd error does not
happen anymore. But the clock issues remain, even with ntpd running.
The ntpd starts without complaints now, and seems to be running with its
regular processes:
_ntp 70093 0.0 0.5 920 2540 ?? S<sp 7:04PM 0:00.02 ntpd: ntp
engine (ntpd)
_ntp 51912 0.0 0.5 736 2464 ?? Isp 7:04PM 0:00.01 ntpd: dns
engine (ntpd)
root 46674 0.0 0.3 792 1640 ?? S<sp 7:04PM 0:00.00
/usr/sbin/ntpd -s
I have set kern.timecounter.hardware to tsc:
# systctl kern.timecounter
kern.timecounter.tick=1
kern.timecounter.timestepwarnings=0
kern.timecounter.hardware=tsc
kern.timecounter.choice=i8254(0) tsc(-1000) dummy(-1000000)
trondd asked for the output of ntpctl -sa, which shows me the following:
# ntpctl -sa
0/4 peers valid, 1/1 sensors valid, constraint offset -3s, clock unsynced,
clock offset is 12771.710ms
peer
wt tl st next poll offset delay jitter
213.154.236.182 from pool pool.ntp.org
1 2 - 152s 300s ---- peer not valid ----
83.98.201.134 from pool pool.ntp.org
1 2 - 152s 300s ---- peer not valid ----
217.23.3.234 from pool pool.ntp.org
1 2 - 152s 300s ---- peer not valid ----
185.92.220.131 from pool pool.ntp.org
1 2 - 152s 300s ---- peer not valid ----
sensor
wt gd st next poll offset correction
vmmci0
1 1 0 15s 15s 18001.915ms 0.000ms
I am not sure how to interpret these numbers. I also don't understand the
"peer not valid" messages here. I have another OpenBSD VM which has the
exact same ntpd.conf and it does not complain about any of the peers.
I think my conclusion is that this is not something that can be solved at
the VM level.
S.
On Sat, Nov 3, 2018 at 8:10 PM Stefan Arentz <stefan.are...@gmail.com>
wrote:
> Hi everyone,
>
> I am having an issue where an OpenBSD VM running on vmd is having
> serious clock skew issues.
>
> I am relatively new to OpenBSD, so I am not sure how to properly debug
> this. What I hope is that I can provide a good amount of data and folks
> here can give me some hints and ask me for additional information to
> get to the root cause of this.
>
> So first some facts and symptoms:
>
> - Both Host and Guest are running OpenBSD 6.4. The host runs GENERIC.MP
> and the guest GENERIC.
> - The host runs 50 guests, all OpenBSD (openbsd.amsterdam)
> - Only this VM is having this clock issue (is this correct, or were
> there others?)
>
> - The guest has kern.timecounter.hardware=tsc
> - The time on the VM was set with rdate a couple of days ago, and as of
> now the VM is running about 4 hours behind.
> - ntpd is running (main process, dns engine, ntp engine)
> - when started or restarted, ntpd complains about "pipe write error
> (from main): No such file or directory" but does seem to start
>
> - I just ran rdate nl.pool.ntp.org and the date was properly updated
> - One minute after running rdate, the clock is already 7 seconds slow
>
> - The guest also has some severe networking issues. often I cannot type
> more than a few characters before a ~15 second delays happens.
> Interactive typing is difficult.
> - I can SSH into the Host and have none of these issues, ruling out
> connectivity issues between me (Toronto) and the Host (Amsterdam)
>
> It would be easy to blame this on NTPd, which does have an unexplained
> error message. However, I think even without running NTPd, the clock
> skew should not be this extreme.
>
> Somehow I have a gut feeling that the clock issues and the networking
> issues are related.
>
> I am root on the VM but I am not on the host. I do have vmctl access.
> However, the host admin is friendly (Hi Mischa) and is happy to help to
> debug this issue.
>
> I tried to ktrace ntpd to get more insight in the "pipe write error
> (from main): No such file or directory" error but I did not get useful
> info out of it. This may be because of my unfamiliarity with those
> tools.
>
> Help appreciated :-)
>
> S.
>
>
