Fincer
commented
3 years ago @solsticedhiver
Yeah, 2 years after your initial report, this is still an issue and not fixed. The current systemd unit file implementation of alarm-fake-hwclock tend to break multiple critical systemd service initializations, causing unexpected loop cycles for them, causing chain of failure events for multiple critical systemd services.
For instance, in my case, because of fake-hwclock.service, one of failing services is systemd-tmpfiles-setup.service. Due to this failure many critical system daemons, such as MariaDB database server daemon, end up into failure state, too.
I understand the interest of starting fake-hwclock as early as possible but the current process is not clean, and the current systemd implementation can't be recommended for any critical Linux systems. To put it simple: fix systemd service files, and please, do proper stress tests for them.
I ended up replacing the systemd files as follows:
fake-hwclock.service
[Unit]
Description=Update hardware clock time
Documentation=man:fake-hwclock(8)
[Service]
Type=oneshot
ExecStart=/usr/bin/fake-hwclock
ExecStop=/usr/bin/fake-hwclockfake-hwclock.timer
[Unit]
Description=Update hardware clock time
[Timer]
OnBootSec=0s
OnUnitActiveSec=1min
[Install]
WantedBy=timers.targetI have changed the timer to run in 1 minute intervals, since the process itself does not take CPU processing time at all.
Not sure, but one might want to take a look at time-set.target and time-sync.target, too.
The above changes as a patch:
--- a/systemd/fake-hwclock.service 2021-01-22 22:36:03.958161099 +0200
+++ b/systemd/fake-hwclock.service 2021-01-22 22:38:22.498863654 +0200
@@ -1,15 +1,8 @@
[Unit]
-Description=Fake Hardware Clock
+Description=Update hardware clock time
Documentation=man:fake-hwclock(8)
-DefaultDependencies=no
-Before=systemd-journald.service
-Conflicts=shutdown.target
-Wants=fake-hwclock.timer
[Service]
-Type=simple
+Type=oneshot
ExecStart=/usr/bin/fake-hwclock
ExecStop=/usr/bin/fake-hwclock
-
-[Install]
-WantedBy=local-fs-pre.target
--- a/systemd/fake-hwclock.timer 2021-01-22 22:36:07.275806325 +0200
+++ b/systemd/fake-hwclock.timer 2021-01-22 22:37:49.834665426 +0200
@@ -1,6 +1,9 @@
[Unit]
-Description=Update fake hardware clock time
-PartOf=fake-hwclock.service
+Description=Update hardware clock time
[Timer]
-OnUnitActiveSec=15m
+OnBootSec=0s
+OnUnitActiveSec=1min
+
+[Install]
+WantedBy=timers.targetFeel free to create a PR.
Marietto2008
pulkomandy
When I run and enable fake-hwclock.service, I am seeing in my log a lot of dependany cycle found by auditd an then failed start of many systemd service.
This leads to many criticial failure, the worse being systemd-journald not logging anything.
Here an execprt of what I am seeing in my logs (volatile log in /run because nothing is written to disk, ever)
and possibly others
my guess is that it IS NOT a good idea to include a .timer service inside the .serivce file at all.
On a rpi3 running archlinux-arm, with root on an external SSD via usb.