Bug#1110145: systemd: Timer's elapse time is gradually drifting further into the future

[Universal Wombat]

On Wed, 30 Jul 2025 18:43:17 +0100 Luca Boccassi <bluca at debian.org> wrote:
 > Control: tags -1 moreinfo
 >
 > On Wed, 30 Jul 2025 19:15:13 +0200 loomylabs <universal.wombat at web.de>
 > wrote:
 > > Package: systemd
 > > Version: 252.38-1~deb12u1
 > > Severity: normal
 > > X-Debbugs-Cc: universal.wombat at web.de
 > >
 > > Dear Maintainer,
 > >
 > > I have a server configured to automatically suspend using "systemctl
 > suspend" after X minutes of inactivity on specific monitored ports.
 > When traffic is detected on those ports, the gateway wakes the server
 > via Wake-on-LAN.
 > >
 > > I have a systemd timer set to reboot the server sometime between 4
 > and 5 a.m.
 > >
 > > The timer looks like this:
 > >
 > > [Unit]
 > > Description=Daily reboot
 > >
 > > [Timer]
 > > OnCalendar=*-*-* 04:00:00
 > > RandomizedDelaySec=3600
 > > WakeSystem=true
 > > Unit=reboot.target
 > >
 > > [Install]
 > > WantedBy=timers.target
 > >
 > > Each time the server wakes from suspend, the elapsed time is
 > recalculated and shifts progressively further into the future,
 > significantly surpassing the 4 to 5 o'clock window.
 >
 > Can you reproduce on testing or unstable?
 >
 >


Sure, I can set up a Debian box on testing and try to reproduce the 
behavior, but that will take a few days.

Aside from that, I took another look at the systemd timer documentation 
and noticed this part in the "RandomizedDelaySec=" section: "The delay 
is added on top of the next determined elapsing time or the service 
manager's startup time, whichever is later."

Could it be that the drift is intentional? I'm not sure if a wake-up 
from suspend is considered a "service manager startup," but the time 
drift does seem to match the "whichever is later" condition. Looking at 
the logs, I notice that the timer elapses at 04:48:58, and the wake-up 
completes a few seconds later. So, systemd might interpret that as 
having missed the elapse time and therefore reschedules it slightly into 
the future. This cycle could then repeat a few times, resulting in 
significant shifts in the elapse time.

What do you think?