My systemd rant
I'm not a big fan of systemd, but it does look like the transition
So I'm not going to wail and gnash my teeth about the change, but I just
wanted to list issues that bite me just so when a systemd supporter tries
to make some sort of optimistic claim that they broke nothing in the
transition I can point out my list of annoyances.
- Dropping to single-user mode if partition is missing
I had the /home partition die on a server machine (in a somewhat
dramatic incident involving a SATA power connector melting and
smoke pouring out of the system).
The system is the network gateway to my entire lab, so it
was important to get it back up quickly.
I popped out the drive and expected the system to boot normally
(though of course giving a warning on login to users that /home
I rebooted, and waited, and waited, and waited.
This machine is headless so I eventually had to drag a monitor
and keyboard over... just to find that 20 years of Linux behavior
had been changed, and instead of continuing boot with a warning
it waited 5 minutes for the partition to appear then dropped
to single-user mode. *without* starting packet forwarding.
The Debian people claim that this was listed in the Jessie release
notes (it was) but this is the kind of corner case change you
weren't expecting to have to worry about.
Hassle level: pretty high.
- Disabling GUI login
I was trying to setup a Raspbian Jessie system to boot headless
(for cluster nodes). Despite using
service disable lightdm it still would pop up on reboot.
*Obviously* I should have known that what I needed to do was
systemctl set-default multi-user.target
Hassle level: thank goodness for google or I never would have
figured this out.
- shutdown dropped some command line arguments
In Debian unstable around October 2015 shutdown command
moved to the one provided by systemd.
My old standby of shutdown -h -t1 now now errors out;
the -t option is no longer supported.
Hassle level: need to re-train 20 years of muscle memory when
rebooting a system.
- max number of FDs raised to 65536
On Debian Jessie the max number of open file descriptors
was raised from 4096 to 65536. ulimit -Hn
will show you this. This causes some interesting bugs in
programs, including some overflow bugs in the perf_event
interface. I thought this was a Debian policy decision,
but the kernel was not changed, nor was
/etc/security/limits.conf. It turns out this
is some sort of bug in systemd, although
Poettering himself denies that such a bug is possible
in the comments of
Update: it turns out this is actually a PAM bug that is triggered
by systemd being so fancy. PAM reads the RLIMIT_NOFILE value
from /proc/1/limits which is a bad idea if your init is using
a high value.
this debian bug and for reference the
original systemd change.
Hassle level: had to report various bugs to linux-kernel
for the perf_event interface.
Also wasted time verifying the source of the issue
because I mildy believed Poettering when he claimed it wasn't
Back to the VMW Software Page