Ghostty-like linux-cgroup = always feature via systemd-run. The benefit
is overall improved desktop interactivity by preventing resource hogging,
enhancing isolation, and ensuring fair resource distribution.
This works with { alacritty, kitty, konsole, qterminal, screen,
tmux: server } and interactive shells { bash, fish, zsh }.
Refer to the end of the readme for Ghostty and GNOME Terminal setup.
Niceness support refers to calling the nice command across multiple terminal
windows or panes, made possible by attaching to the same cgroup. This requires
the libcgroup package and have the cgconfig.service enabled.
Steps for Arch-based distributions:
source ~/.config/user-dirs.dirs
cd "$XDG_DOWNLOAD_DIR"
git clone --depth=1 https://aur.archlinux.org/libcgroup.git
cd libcgroup
gpg --keyserver keyserver.ubuntu.com --recv-keys E050B79D5B290D03
makepkg -si
sudo systemctl enable --now cgconfig.serviceInstallation
The installer adds 10 entries to /etc/cgconfig.conf, ending in {0..9}.
$ ./cgconfig.install
Adding pool of cgroup entries to /etc/cgconfig.conf
group user.slice/user-ID.slice/term-0 {
perm {
admin {
uid = [USER];
gid = [USER];
}
task {
tuid = [USER];
tgid = [USER];
}
}
}
...
Adding and enabling set-owner-cgroup-procs-ID.service
Created symlink '/etc/systemd/system/sysinit.target.wants/set-owner...
Please reboot the machine for the changes to take effect
Preferably, source the environment file near the top of your shell startup script for early activation.
# ~/.bashrc
source /path/to/env-systemd/attach_shell_to_unique_cgroup.bash
# ~/.config/fish/config.fish
source /path/to/env-systemd/attach_shell_to_unique_cgroup.fish
# ~/.zshrc
source /path/to/env-systemd/attach_shell_to_unique_cgroup.zshOr copy the env file to your shell function folder. Comment out the
line calling attach_shell_to_unique_cgroup. Instead, call it near
the top of your shell startup script.
Launch the terminal application and check the cgroup membership.
$ cat /proc/self/cgroup
---------------------------------------------------------------
0::/user.slice/.../user@1000.service/app.slice/shell-3608.scopeThe systemd-cgtop command can be used to monitor resource usage.
Being deeply nested, cgtop shows one service entry where the
app.slice task groups reside.
$ systemd-cgtop
----------------------------------------------------------------------
CGroup Tasks %CPU Memory Input/s Output/s
/ 986 1600.9 5G - -
user.slice 470 1599.8 2G - -
user.slice/user-1000.slice 470 1599.8 1.9G - -
user.slice/.../user@1000.service 155 1598.1 580.7M - -
user.slice/.../session-2.scope 306 0.2 1.1G - -
system.slice 66 0.1 444.9M - -The INVOCATION_ID environment variable is set by systemd for all
processes started as part of a service unit, including those launched
via systemd-run. If starting screen, tmux, or another terminal
emulator from inside the shell, clear the INVOCATION_ID variable.
$ env INVOCATION_ID= konsole -e zsh
$ env INVOCATION_ID= qterminal -e zsh
$ env INVOCATION_ID= screen -S my_project
$ env INVOCATION_ID= tmuxcgterm_attach
A process/thread's nice value has an effect for scheduling decisions only
relative to other process/threads in the same task group. For nice support
across multiple terminal windows or panes, call helper function with [0-9]
to attach multiple emulators to the same base cgroup.
# terminal one
$ cgterm_attach 1
0::/user.slice/user-1000.slice/term-1
$ nice -n 0 primesieve 1e12
Seconds: 6.144
# terminal two
$ cgterm_attach 1
0::/user.slice/user-1000.slice/term-1
$ nice -n 9 primesieve 1e12
Seconds: 10.478cgterm_detach
Detach reverts back to the originating cgroup. The result is the Linux kernel scheduler equalizes the distribution of CPU cycles across the task groups.
# terminal one
$ cgterm_detach
0::/user.slice/.../user@1000.service/app.slice/shell-3608.scope
$ nice -n 0 primesieve 1e12
Seconds: 10.434
# terminal two
$ cgterm_detach
0::/user.slice/.../user@1000.service/app.slice/shell-2715.scope
$ nice -n 9 primesieve 1e12
Seconds: 10.510cgterm_cpus
The cgterm_cpus [list,range,all] function can be used to get/set the
CPU affinity, which restricts the tasks to a specific set of CPUs.
$ cat /proc/self/cgroup
0::/user.slice/.../user@1000.service/app.slice/shell-3608.scope
$ cgterm_cpus
0-31
$ cgterm_cpus 0-7,16,20
$ cgterm_cpus
0-7,16,20
$ cgterm_cpus all
$ cgterm_cpus
0-31Two user-defined environment variables can be used to set performance or powersave CPUs. Set the IDs to your CPU specification or anything you like e.g. leave out the SMT siblings.
$ export CGTERM_PERFORMANCE_CPUS="0-15"
$ export CGTERM_PERFORMANCE_CPUS="0,2,4,6,8,10,12,14" # exclude SMT
$ cgterm_cpus performance
$ cgterm_cpus
0-15
$ export CGTERM_POWERSAVE_CPUS="16-31"
$ cgterm_cpus powersave
$ cgterm_cpus
16-31cgterm_memnodes
The cgterm_memnodes [list,range,all] function can be used to get/set the
CPU affinity, which restricts the tasks to a specific set of memory nodes.
Calling cgterm_cpus with all, performance, or powersave resets
the control group cpuset.mems to default.
$ cat /proc/self/cgroup
0::/user.slice/.../user@1000.service/app.slice/shell-3608.scope
$ cgterm_memnodes
0-1
$ cgterm_cpus
0-31
$ cgterm_memnodes 0
$ cgterm_cpus
0-7,16-23
$ cgterm_memnodes all
$ cgterm_cpus
0-31cgterm_quota
The cgterm_quota [1..100] function can be used to get/set the max CPU
quota percentage. The default percent is 100 with a range of 1 to 100.
Applying a change to a base cgroup emits an error. Likewise, for the other helper functions.
$ cat /proc/self/cgroup
0::/user.slice/.../user@1000.service/app.slice/shell-3608.scope
$ cgterm_quota
100
$ primesieve 1e12
Seconds: 5.224
$ cgterm_quota 50
$ primesieve 1e12
Seconds: 10.543
$ cgterm_attach 2
0::/user.slice/user-1000.slice/term-1
$ cgterm_quota 70
cannot apply quota: attached to a base cgroupcgterm_weight
The cgterm_weight [1..10000, idle, normal] function can be used to
get/set the relative priority for CPU and I/O resources, when there is
contention. A cgroup with a higher weight will receive a proportionally
larger share. The normal weight is 100 with a range of 1 to 10,000.
Specifying idle will mark the control group for "idle scheduling",
equivalent to running the tasks with the SCHED_IDLE policy.
The weight is applied to cpu.weight and io.weight.
# terminal one
$ cat /proc/self/cgroup
0::/user.slice/.../user@1000.service/app.slice/shell-3608.scope
$ cgterm_weight 100 (default)
$ primesieve 1e12
Seconds: 10.699
# terminal two
$ cat /proc/self/cgroup
0::/user.slice/.../user@1000.service/app.slice/shell-2715.scope
$ cgterm_weight 500
$ primesieve 1e12
Seconds: 7.041cgterm_nice
Alternatively, the cgterm_nice [-20..19, idle, normal] function can be
used to get/set the weight using the same values as the nice command,
ranging from -20 to 19 (normal 0). A cgroup with a lower nice value will
receive a relative larger share. Specifying idle will mark the control
group for "idle scheduling", its processes become low-priority tasks.
The resulting CPU weight is applied to io.weight, as well.
# terminal one
$ cat /proc/self/cgroup
0::/user.slice/.../user@1000.service/app.slice/shell-3608.scope
$ cgterm_nice 0 (default)
$ primesieve 1e12
Seconds: 6.039
# terminal two
$ cat /proc/self/cgroup
0::/user.slice/.../user@1000.service/app.slice/shell-2715.scope
$ cgterm_nice 9
$ primesieve 1e12
Seconds: 10.437cgterm_reset
The cgterm_reset function can be used to reset the control group files.
$ cgterm_reset
# cpu.max max 100000
# cpu.idle 0
# cpu.weight 100
# cpu.weight.nice 0
# io.weight 100
# cpuset.cpus 0 or 0-N
# cpuset.mems 0 or 0-NRun the uninstall script to undo the system-level changes. Separately, remove the entry from your shell's function folder or startup script.
$ ./cgconfig.uninstall
Removing pool of cgroup entries from /etc/cgconfig.conf
group user.slice/user-ID.slice/term-0 {
perm {
admin {
uid = [USER];
gid = [USER];
}
task {
tuid = [USER];
tgid = [USER];
}
}
}
...
Disabling and removing set-owner-cgroup-procs-ID.service
Removed '/etc/systemd/system/sysinit.target.wants/set-owner...
Please reboot the machine for the changes to take effect
Ghostty can isolate each terminal surface directly with the
linux-cgroup = always option: ~/.config/ghostty/config
Calling attach_shell_to_unique_cgroup is a NO-OP, does nothing.
Isolating GNOME Terminal windows is possible as well, simply adding
the CPUQuota entry to the service file.
Likewise, calling attach_shell_to_unique_cgroup is a NO-OP.
mkdir -p ~/.config/systemd/user
cd ~/.config/systemd/user
cp /usr/lib/systemd/user/gnome-terminal-server.service .Edit the file and add line CPUQuota=100% to the Service section.
Add also, AllowedCPUs=0-N matching your system specification.
[Unit]
Description=GNOME Terminal Server
PartOf=graphical-session.target
[Service]
Slice=app-org.gnome.Terminal.slice
CPUQuota=100%
AllowedCPUs=0-15
Type=dbus
BusName=org.gnome.Terminal
ExecStart=/usr/lib/gnome-terminal-server
TimeoutStopSec=5s
KillMode=process
Restart the service. This will close any opened GNOME Terminal windows. So, be mindful of any unsave work inside GNOME Terminal.
systemctl --user daemon-reload
systemctl --user restart gnome-terminal-server.serviceThank you, CachyOS community with sounding board and testing.
- ms178
- nutcase
- phusho
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or
at your option any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
