Exiting Niri in a systemd service does not wait for `graphical-session.target`, causing some services to immediately fail due to losing the Wayland socket

[Frontear]

It seems that niri.service exits before graphical-session.target is able to terminate, so services which come down alongside it are left in a broken state and usually terminate with error codes. Systemd then marks the system status as "degraded" due to the failed services.

The services which fail to shutdown is inconsistent, as their stop job is subject to race conditions. Ideally graphical-session.target is allowed to stop before Niri terminates, so that the wayland socket survives long enough.

Relevant journalctl output (last line is when graphical-session.target is stopped):

Sep 23 21:30:26 LAPTOP-3DT4F02 niri[117432]: 2025-09-24T01:30:26.607749Z  INFO niri::input: quitting after confirming exit dialog
Sep 23 21:30:26 LAPTOP-3DT4F02 niri[117432]: 2025-09-24T01:30:26.958939Z DEBUG niri::utils::watcher: exiting watcher thread for Regular { user_path: "/home/frontear/.config/niri/config.kdl">
Sep 23 21:30:26 LAPTOP-3DT4F02 .xdg-desktop-po[117540]: Lost connection to Wayland compositor.
Sep 23 21:30:26 LAPTOP-3DT4F02 .xdg-desktop-po[117583]: Error reading events from display: Broken pipe
Sep 23 21:30:26 LAPTOP-3DT4F02 .waybar-wrapped[117486]: Error reading events from display: Broken pipe
Sep 23 21:30:26 LAPTOP-3DT4F02 .thunar-wrapped[117483]: Error reading events from display: Broken pipe
Sep 23 21:30:26 LAPTOP-3DT4F02 foot[117476]: wayland: failed to read events from the Wayland socket: Connection reset by peer
Sep 23 21:30:26 LAPTOP-3DT4F02 foot[117476]:  err: wayland.c:1659: failed to read events from the Wayland socket: Connection reset by peer
Sep 23 21:30:26 LAPTOP-3DT4F02 foot[117476]: warn: server.c:74: client FD=8: terminal still alive
Sep 23 21:30:26 LAPTOP-3DT4F02 .swayosd-server[117482]: Lost connection to Wayland compositor.
Sep 23 21:30:26 LAPTOP-3DT4F02 foot[117476]: server: client FD=8: terminal still alive
Sep 23 21:30:26 LAPTOP-3DT4F02 polkit-gnome-au[117477]: Error reading events from display: Broken pipe
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: niri.service: Consumed 34.070s CPU time, 159.5M memory peak.
Sep 23 21:30:26 LAPTOP-3DT4F02 foot[117476]:  err: wayland.c:2302: failed to roundtrip Wayland display: Connection reset by peer
Sep 23 21:30:26 LAPTOP-3DT4F02 foot[117476]:  err: wayland.c:2302: failed to roundtrip Wayland display: Connection reset by peer
Sep 23 21:30:26 LAPTOP-3DT4F02 foot[117476]: wayland: failed to roundtrip Wayland display: Connection reset by peer
Sep 23 21:30:26 LAPTOP-3DT4F02 foot[117476]: wayland: failed to roundtrip Wayland display: Connection reset by peer
Sep 23 21:30:26 LAPTOP-3DT4F02 polkitd[1867]: Unregistered Authentication Agent for unix-session:11 (system bus name :1.249, object path /org/gnome/PolicyKit1/AuthenticationAgent, locale en>
Sep 23 21:30:26 LAPTOP-3DT4F02 foot[117476]: wayland: failed to roundtrip Wayland display: Connection reset by peer
Sep 23 21:30:27 LAPTOP-3DT4F02 foot[117476]:  err: wayland.c:2302: failed to roundtrip Wayland display: Connection reset by peer
Sep 23 21:30:27 LAPTOP-3DT4F02 foot[117476]: warn: server.c:74: client FD=20: terminal still alive
Sep 23 21:30:27 LAPTOP-3DT4F02 foot[117476]:  err: wayland.c:2302: failed to roundtrip Wayland display: Connection reset by peer
Sep 23 21:30:27 LAPTOP-3DT4F02 foot[117476]:  err: wayland.c:2302: failed to roundtrip Wayland display: Connection reset by peer
Sep 23 21:30:27 LAPTOP-3DT4F02 foot[117476]:  err: wayland.c:2302: failed to roundtrip Wayland display: Connection reset by peer
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: polkit-gnome.service: Main process exited, code=exited, status=1/FAILURE
Sep 23 21:30:26 LAPTOP-3DT4F02 foot[117476]: server: client FD=20: terminal still alive
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: polkit-gnome.service: Failed with result 'exit-code'.
Sep 23 21:30:26 LAPTOP-3DT4F02 foot[117476]: wayland: failed to roundtrip Wayland display: Connection reset by peer
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: Stopped GNOME PolicyKit Agent.
Sep 23 21:30:26 LAPTOP-3DT4F02 foot[117476]: wayland: failed to roundtrip Wayland display: Connection reset by peer
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: Stopped Idle manager for Wayland.
Sep 23 21:30:26 LAPTOP-3DT4F02 foot[117476]: wayland: failed to roundtrip Wayland display: Connection reset by peer
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: swayidle.service: Consumed 2.062s CPU time, 146.9M memory peak.
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: thunar.service: Main process exited, code=exited, status=1/FAILURE
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: thunar.service: Failed with result 'exit-code'.
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: Stopped thunar.service.
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: Stopped Dunst notification daemon.
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: xdg-desktop-portal-gnome.service: Main process exited, code=exited, status=1/FAILURE
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: xdg-desktop-portal-gnome.service: Failed with result 'exit-code'.
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: Stopped Portal service (GNOME implementation).
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: xdg-desktop-portal-gtk.service: Main process exited, code=exited, status=1/FAILURE
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: xdg-desktop-portal-gtk.service: Failed with result 'exit-code'.
Sep 23 21:30:26 LAPTOP-3DT4F02 systemd[1747]: Stopped Portal service (GTK/GNOME implementation).
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: swayosd.service: Main process exited, code=exited, status=1/FAILURE
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: swayosd.service: Failed with result 'exit-code'.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: Stopped swayosd.service.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: swayosd.service: Consumed 1.302s CPU time, 28.8M memory peak.
Sep 23 21:30:27 LAPTOP-3DT4F02 foot[117476]:  err: wayland.c:2267: failed to flush wayland socket: Connection reset by peer
Sep 23 21:30:27 LAPTOP-3DT4F02 foot[117476]: wayland: failed to flush wayland socket: Connection reset by peer
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: waybar.service: Main process exited, code=exited, status=1/FAILURE
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: waybar.service: Failed with result 'exit-code'.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: Stopped Highly customizable Wayland bar for Sway and Wlroots based compositors..
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: waybar.service: Consumed 2.512s CPU time, 25.1M memory peak.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: foot.service: Main process exited, code=exited, status=230/PERSONALITY
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: Stopping microsoft-edge...
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: Stopped target Startup of XDG autostart applications.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: Stopping Portal service...
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: Stopping flatpak document portal service...
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: Stopping sandboxed app permission store...
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1]: run-user-1000-doc.mount: Deactivated successfully.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: Stopped sandboxed app permission store.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: Stopped Portal service.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: Stopped flatpak document portal service.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: Stopped microsoft-edge.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: app-niri-microsoft\x2dedge-763a6496.scope: Consumed 1min 51.233s CPU time, 1.2G memory peak.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: app-org.chromium.Chromium-118028.scope: Consumed 55.911s CPU time, 1.2G memory peak.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: foot.service: Failed with result 'exit-code'.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: Stopped Fast, lightweight and minimalistic Wayland terminal emulator..
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: foot.service: Consumed 55.052s CPU time, 2.1G memory peak.
Sep 23 21:30:27 LAPTOP-3DT4F02 systemd[1747]: Stopped target Current graphical user session.

System Information

• niri version: niri 25.08 (Nixpkgs)

• Distro: NixOS

• GPU: Intel Iris Xe Graphics

• CPU: 11th Gen Intel(R) Core(TM) i7-1195G7

[Frontear]

Giving the code a look, I see that Niri handles the SIGTERM signal and shuts down exactly the same as if Action::Quit was used. This leads me to the idea that you could instead make quit tell systemd to stop niri.service, which would propagate stop jobs for relevant systemd services and still correctly shutdown Niri.

Alternatively, users could just replace quit with spawn "systemctl" "--user" "stop" "niri.service". Might need --no-block here maybe.

This would be a major improvement in controlling the shutdown sequence and making it more predictable and in-line with the systemd service's lifetime.

[Frontear]

I tried somewhat getting this to work by firing systemctl --user start niri-shutdown.target (I could have also just done systemctl --user stop graphical-session.target). This properly shuts down all services that were brought up with graphical-session.target, but it also causes Electron apps to coredump. This is a separate issue from Niri, but it becomes kind of relevant in this context.

[cameronaw13]

If you wish to retain a confirmation dialogue and still use systemctl --user stop graphical-session.target before it's fixed you can create a bash script with the following:

#!/usr/bin/env bash
message=$(cat << 'EOF'

  Are you sure you want to exit niri?

  Press 'ENTER' to continue.

EOF
)

read -rsN 1 -p "$message" key
case "$key" in
  $'\n')
    systemctl --user --no-block stop graphical-session.target
    ;;
esac

exit 0

Then in your config.kdl you can set your quit bind rule to something like this:

Ctrl+Alt+Delete { spawn-sh "alacritty -T niri-shutdown-prompt -o 'cursor.style=\"Underline\"' --command sh -c script-niri-shutdown"; }

You can change the terminal emulator "alacritty" and the script path "script-niri-shutdown" to fit your setup.

Then I use alacritty's -T flag to set the window title so that I can create a window rule like so:

// niri-shutdown-prompt: window styling
window-rule {
    match app-id="Alacritty" title="niri-shutdown-prompt"
    open-floating true
    default-column-width { fixed 355; }
    default-window-height { fixed 100; }
}

Title, column width, and window height will also depend on your setup.

This of course comes with its own set of issues due to the window being non-sticky and unfocusable but it works as a bandaid.

[cameronaw13]

Update: The method above didn't work for me 100% reliably since, according to my logs, niri also starts graphical-session.target too early during startup. If the wayland socket isn't ready in time, the user services could hit their restart rate limit before connecting to wayland.

You could use UpheldBy=graphical-session.target under the [Install] section of the user service to force the service to continuously restart when stopped during the graphical-session despite failure rate limits. Although oneshot services in this scenario would likely need a RemainAfterExit=true under [Service] so that they don't restart repeatedly. This of course prevents you from later stopping these user services after startup. Alternatively, you could probably just increase RestartSec under [Service] and StartLimitBurst/StartLimitIntervalSec under [Unit] by an arbitrary amount giving them enough time to connect to the wayland socket. This method is gross but that's the nature of bandaids.

Of course, you could also just make the user services depend on niri.service itself but that wouldn't allow them to be portable.

Edit: on hindsight I kinda strayed off topic with this one. I'm essentially dealing with the same issue with the race condition between niri/wayland-socket and graphical-session.target only that you show it affecting niri's shutdown and I show it affecting niri's startup. According to my user service logs, it seems that this race condition occurs on both niri-session on startup and shutdown.

Edit 2: Ugh, I think my issue actually is apart of a different issue. My user services that were failing to start were also using Requisite=graphical-session.target which, ironically despite starting with graphical-session.target via WantedBy, prevented them from starting up. It was misleading because my logs kept on showing that wayland-socket didn't exist on startup unless I increased their startup time. Removing the Requisite dependency allowed these services to properly start. This affects user services like waybar where, by default, they include a Requisite=graphical-session.target in their service file.

[Frontear]

I appreciate the extra debugging time that went into your posts. I don't think its unrelated since these targets are important to synchronise services appropriately to the current user session, and it touches onto the overarching issue that I'm facing as well.

[YaLTeR]

niri also starts graphical-session.target too early during startup. If the wayland socket isn't ready in time

At least using the service file provided in the niri repo, everything is made to work correctly in this regard. If it doesn't work for you then it's a misconfiguration of your system.

[cameronaw13]

Yeah its likely. Managed to fix it anyway for myself and if anyone else somehow experiences the same issue they can use it as reference. There has to be some grander issue at play on my setup but I'll figure it out eventually.

[YaLTeR]

Well seems like this was a system configuration issue

[bamarsha]

I thought the configuration issue was about services not starting when logging in, different from services not stopping when logging out? I have the problem where niri exits before graphical services have a chance to cleanly exit, so systemctl --user shows failed services after logging out, but everything starts correctly at login (well, at the first login - I think the unclean exit leaves some lingering processes that cause issues only when logging in a second time...).

[BBaoVanC]

At least for me, it's not a system configuration issue. I am using the official niri-session script, but the problem is it doesn't run the niri-shutdown.target until after the main niri process exits. If instead I exit niri by executing niri-shutdown.target, then everything shuts down fine. The quit keybinds don't do that, so units depending on the compositor fail when exiting using the keybind.