Proxmox 7 install clobbers timesyncd conf file

[kyriazis]

Hello,

I am in the process of upgrading our cluster to Proxmox 7, and came across the following:

We have locally edited /etc/systemd/timesyncd.conf to use a local NTP server, since our firewall does not allow external NTP connections.
After upgrading to Proxmox 7, the above conf file edits are clobbered and the setting for our local NTP server has disappeared.

While upgrading, I've answered "no" to the question of whether the update should automatically install updated configuration files for various services, and in fact it does ask me for several conf files, but not timesyncd.conf.

Thanks!

George

 

[Stoiko Ivanov]

We have locally edited /etc/systemd/timesyncd.conf to use a local NTP server, since our firewall does not allow external NTP connections.
After upgrading to Proxmox 7, the above conf file edits are clobbered and the setting for our local NTP server has disappeared.

Sounds really odd - debian puts quite a bit of effort into not overwriting your changes...


is the conffile still there /etc/systemd/timesyncd.conf ? - what's it's content?

In any case - Proxmox VE changed it's suggestions away from systemd-timesyncd to a regular NTP server (chrony, ntpd, openntpd) - see:
https://pve.proxmox.com/wiki/Roadmap#Proxmox_VE_7.0 (search for chrony)

Is `systemd-timesyncd` still installed? (`dpkg -l |grep timesync`)

You might want to consider installing chrony (and configuring that with your local NTP-Server)

I hope this helps!

 

[kyriazis]

the conf file is still there, except it's been overwritten. A backup file is created (timesyncd.conf.dpkg-bak) but no user feedback that it occurred. The only way I found out is because ceph complained about some of my nodes having a drifting timestamp.

timesyncd is still installed and running.

Thanks for the pointer on the Proxmox 7 guidelines. I must've missed in while reading the 7.0 changes.

I would like to suggest that the installer somehow picks up the fact that timesyncd is still enabled and/or has a modified conf file and suggests alternatives. Since both timesyncd and chrony are Debian packages, maybe provide a local pve package that "does the right thing"? Looks like the decision to move to chrony is based on the server-based nature of Proxmox and hence "overrides" any decision made by Debian folks. Since base Proxmox features (like ceph) depend highly on time synchronization, it feels like the decision on what time sync daemon should run should really be a pve decision (since ceph is a pve package).

I am not sure what happens with new installs, though. Things could be easier there.

Thanks,

george

 

[fabian]

new installs made with the PVE iso/installer already default to chrony - we just don't want to force this on all existing users.

if you have a backup config file created by dpkg, you either said yes when prompted to use the config as shipped by the package, or you ran the upgrade with settings that did that automatically

 

[kyriazis]

I am sure I said "no", since the upgrade(s) asked about other configuration, just not that one.

Are there other settings that control config file prompting? I didn't think so..

 

[fabian]

yeah, you can pass dpkg options to force keeping or force overwriting config files, running as non-interactive / without a TTY/stdin available also affects the logic obviously.

you can check /var/log/apt/term.log if it is still around

 

[orion]

[...]
We have locally edited /etc/systemd/timesyncd.conf to use a local NTP server, since our firewall does not allow external NTP connections.
After upgrading to Proxmox 7, the above conf file edits are clobbered and the setting for our local NTP server has disappeared.

While upgrading, I've answered "no" to the question of whether the update should automatically install updated configuration files for various services, and in fact it does ask me for several conf files, but not timesyncd.conf.

The same happened at mine after upgrading to Proxmox 7. Fortunately, this is not a big problem.

 

[GoZippy]

so how did you resolve?

 

[kyriazis]

I re-edited timesyncd.conf after the upgrade. :-(

Not an ideal solution when having a big cluster, but scripting helps. Ansible helps even more.