I went and studied some units in /usr/lib/systemd/system so I can get a better idea of how things are supposed to be done. I saw a pattern, so I mimicked it with good results.

For example, network.target will pull in a bunch of things as dependencies, but each one of those units will also contain a "Before=network.target" directive, to make sure that network.target is the "marker" that signifies all network things are brought up.

So I arranged my user services to x11.target pulling in startx.service
Then I have "Before=x11.target" in startx.service

Then in chrome.service, I have "After=x11.target"

This seems to fix my problem. It's strange since the original problem wasn't that things were killed out-of-order.. journalctl --user -b -1 shows that all user services are killed in the right order, just that chrome wasn't being killed in the way I specified.

Is this expected behavior? Why didn't just doing After=startx.service do the trick?



BTW, I reread my previous comment. The [Unit] I had posted wasn't a copy/paste, so I left out some details.. I had a Requires= directive to trigger deactivation. This is a copy/paste I found from a snapshot of my filesystem.

[Unit]
Description=Launch Chrome Browser
Requires=startx.service x11.target
After=startx.service

[Service]
KillMode=process
ExecStart=/usr/bin/google-chrome-unstable

TL;DR
I changed startx.service to have "Before=x11.target" and I changed chrome.service to have "After=x11.target" instead of "After=startx.service" and now it works as expected

The X11 libs kill processes hard if the connection is severed. THis is most likely what you experienced, and is out of systemd's hands. If you order things properly against X11 however, and make sure that X11 is terminated only after chrome then the X11 libs will not see the connectiosn terminated, and thus is the correct fix for the issue. Closing hence.

(In reply to comment #2)
> The X11 libs kill processes hard if the connection is severed. THis is most
> likely what you experienced, and is out of systemd's hands. If you order
> things properly against X11 however, and make sure that X11 is terminated
> only after chrome then the X11 libs will not see the connectiosn terminated,
> and thus is the correct fix for the issue. Closing hence.

It is kind of strange though. It now mostly works,
but still sometimes fails. I spoke too soon in my
last comment.

There's lots of strangeness going on with my
setup. I'm sure I have things ordered correctly,
and things mostly work, but occasionally things
are as they were before, and there are
occassional startup problems.

I'll do a 'systemctl --user start graphical.target
+some_other_gui_stuff' and X isn't started
correctly, but the GUI stuff are running and
active according to systemctl, which makes no
sense.

But back to the original problem


I've read the systemd documentation and looked
through some unit files in /usr/lib~, so I'm pretty
confident I have the dependency stuff figured out
(enough). 

X has:
Wants=x11.target
Before=x11.target
WantedBy=graphical.target

GUI stuff (including chrome) have:
Requires=x11.target
After=x11.target

So the startup order should be:
X
x11.target
gui stuff

Then at shutdown or deactivation, the Before and
After are reversed, so the units deactivate in
this order:
gui stuff
x11.target
X

The order should be correct, so Chrome shouldn't
be killed by X since X wouldn't even think about
shutting down until Chrome is gone. Is it because
Chrome takes too long to deactivate? Because it
_does_ spend a good while writing a bunch of stuff
to disk. I'm hesitant to add any sleeps or
timeouts to fix my startup/shutdown issues. If
there are any flaws in my Before/After config,
someone please let me know.