Cannot run systemctl disable --now on a template unit

[ondrejbudai]

systemd version the issue has been seen with

245

Used distribution

Fedora 32

Expected behaviour you didn't see

systemctl disable --now osbuild-worker@.service disables and stops all services spawned from a template unit.

Unexpected behaviour you saw

When I run systemctl disable --now osbuild-worker@.service, I got:
Failed to stop osbuild-worker@.service: Unit name osbuild-worker@.service is missing the instance name. See system logs and 'systemctl status osbuild-worker@.service' for details

I tried running systemctl disable --now "osbuild-worker@*.service" and got:
Glob pattern passed to enable, but globs are not supported for this. Invalid unit name "osbuild-worker@*.service" escaped as "osbuild-worker@\x2a.service".

Steps to reproduce the problem

1. Create a template unit:

$ cat /usr/lib/systemd/system/osbuild-worker@.service
[Unit]
Description=OSBuild Composer Worker (%i)
Requires=osbuild-composer.socket

[Service]
Type=simple
ExecStart=/usr/libexec/osbuild-composer/osbuild-worker

2. Try disabling it and stopping in one step:

systemctl disable --now osbuild-worker@.service

I originally discovered this issue when trying to get the systemd_preun RPM macro to get working with a template unit. The systemd_preun macro uses systemctl disable --now which makes it unusable for a template unit.

The workaround I found is to use systemctl disable and systemctl stop separately like this:

systemctl --no-reload disable osbuild-worker@.service
systemctl stop "osbuild-worker@*.service"

This issue was originally reported in #9862 but imho it got accidentally closed.

[AlexAegis]

Just to let everyone know, this issue is still around in version 250

systemctl disable --user --now 'barrier@*'
Glob pattern passed to enable, but globs are not supported for this.
Invalid unit name "barrier@*" escaped as "barrier@\x2a".

[scheuref]

that's annoying...
how can we disable a pattern of services with a simple command?

systemctl status ceph-osd@*.service # => works
systemctl stop ceph-osd@*.service # => works
systemctl disable ceph-osd@*.service # => does not work... why no consistent behavior?

edit 1:

systemctl stop ceph-osd.target # => works
systemctl disable ceph-osd.target ceph.target # => no error, but all ceph-osd@* are still started after rebooting (because of their symlinks in /etc/systemd/system/ceph.target.wants/)

edit 2:

solution 1: using a text file to list all instances
systemctl list-units -t service --state active --no-pager --full --plain --no-legend "ceph-*" | awk '/ceph-/ {print $1}' | tee /etc/opt/ceph-service.txt
cat /etc/opt/ceph-service.txt | while read i; do echo $i; systemctl disable --now $i; sleep 10; done
#reboot
cat /etc/opt/ceph-service.txt | while read i; do echo $i; systemctl enable --now $i; sleep 10; done

solution 2: masking all instances
systemctl stop ceph-osd.target
systemctl mask $(systemctl list-units -t service --all --no-pager --full --plain --no-legend "ceph-osd*" | awk '/ceph-/ {print $1}')
#reboot
systemctl unmask $(systemctl list-units -t service --all --no-pager --full --plain --no-legend "ceph-osd*" | awk '/ceph-/ {print $1}')
#nok: systemctl start ceph-osd.target
#https://serverfault.com/questions/734123/systemd-restart-all-instances-of-an-instantiated-service-at-once
systemctl start --all "ceph-osd@*"

solution 3 (best): masking the template
systemctl stop ceph-osd.target
#nok: systemctl mask ceph-osd@.service # Failed to mask unit: File /etc/systemd/system/ceph-osd@.service already exists.
#https://github.com/systemd/systemd/issues/12084#issuecomment-802468029
mkdir -vp /etc/systemd/system.control/; ln -vs /dev/null /etc/systemd/system.control/ceph-osd@.service; systemctl daemon-reload
#reboot
rm -v /etc/systemd/system.control/ceph-osd@.service; systemctl daemon-reload
systemctl start --all "ceph-osd@*"

[ferdnyc]

@scheuref

systemctl disable unit@.service actually works — it's only when the --now is added in, that it produces an error. (In either case, it does also disable all instances of unit@*.service, although it doesn't mass stop them even if --now is present on the command.)

It's the mass-stop, more than the mass-disable, functionality that's really the problem. Mass-disabling instantiated templates works fine with unit@.service, at least in systemd 253.4:

$ sudo systemctl enable --now akmods@6.2.15-300.fc38.x86_64.service
Created symlink /etc/systemd/system/multi-user.target.wants/akmods@6.2.15-300.fc38.x86_64.service → /usr/lib/systemd/system/akmods@.service.

$ sudo systemctl list-units akmods@\*.service                      
  UNIT                                  LOAD   ACTIVE SUB    DESCRIPTION       >
  akmods@6.2.15-300.fc38.x86_64.service loaded active exited Builds and install>

$ sudo systemctl disable akmods@.service
Removed "/etc/systemd/system/multi-user.target.wants/akmods@6.2.15-300.fc38.x86_64.service".

$ sudo systemctl list-units akmods@\*.service                      
  UNIT                                  LOAD   ACTIVE SUB    DESCRIPTION       >
  akmods@6.2.15-300.fc38.x86_64.service loaded active exited Builds and install>

$ sudo systemctl stop akmods@6.2.15-300.fc38.x86_64.service

$ sudo systemctl list-units akmods@\*.service              
  UNIT LOAD ACTIVE SUB DESCRIPTION
0 loaded units listed. Pass --all to see loaded but inactive units, too.

$ sudo systemctl enable --now akmods@6.2.15-300.fc38.x86_64.service
Created symlink /etc/systemd/system/multi-user.target.wants/akmods@6.2.15-300.fc38.x86_64.service → /usr/lib/systemd/system/akmods@.service.

$ sudo systemctl disable --now akmods@.service
Removed "/etc/systemd/system/multi-user.target.wants/akmods@6.2.15-300.fc38.x86_64.service".
Failed to stop akmods@.service: Unit name akmods@.service is missing the instance name.
See system logs and 'systemctl status akmods@.service' for details.

$ sudo systemctl list-units akmods@\*.service                      
  UNIT                                  LOAD   ACTIVE SUB    DESCRIPTION       >
  akmods@6.2.15-300.fc38.x86_64.service loaded active exited Builds and install>

[ferdnyc]

This is also unfortunate, because /usr/lib/systemd/systemd-update-helper attempts to call systemctl disable --now on its arguments whenever its remove-system-units subcommand is used. Which means that /usr/lib/systemd/systemd-update-helper remove-system-units akmods@.service will return an error, and fail to stop any instantiated units of the template (though it will disable them).

[GrabbenD]

Isn't there a better way of masking / disabling services with wildcard in 2024?

• --all doesn't work for mask/unmask

  $ systemctl mask --all 'getty@tty*.service'
  Glob pattern passed to enable, but globs are not supported for this.

• --all wont work for disable/enable unless --now is specified (which is a deal breaker in a system which isn't booted/when bootstrapping in chroot)

  System has not been booted with systemd as init system (PID 1). Can't operate.

• Currently this is the only workaround that I've come across:

  $ cd /etc/systemd/system/getty.target.wants && \
    systemctl mask getty@tty*.service

• (Sequences are another alternative in this instance. However, wildcards would be cleaner)

  $ systemctl mask getty@tty{1..9}.service

[ferdnyc]

@GrabbenD You're commenting on a closed issue about something that was already fixed. If you think there are issues with how --all or wildcards are handled, I'd suggest opening a new issue documenting the problems. Commenting here is basically just shouting into the void.