On Thu, 2020-02-13 at 13:15 +0100, Iñaki Ucar wrote:
On Thu, 13 Feb 2020 at 11:56, Patrick O'Callaghan
<pocallaghan(a)gmail.com> wrote:
> On Thu, 2020-02-13 at 00:34 +0100, Iñaki Ucar wrote:
> > This isn't a KDE thing, it's systemd. See e.g. [1] and [2].
> >
> > [1]
https://forum.xfce.org/viewtopic.php?id=12994
> > [2]
https://wiki.archlinux.org/index.php/Systemd/User#Kill_user_processes_on_...
>
> I know what causes it. My point is that the problem occurs with KDE and
> other desktops, but not apparently with Gnome, so it would be
> interesting to know why this is. Does systemd treat Gnome specially, or
> does Gnome have some kind of built-in workaraound that avoids this
> issue?
Are you sure that this doesn't happen with Gnome?
No, I'm simply relying on some comments in the above references.
There was a change in systemd some years ago [1] to
KillUSerProcesses
by default, and if you follow the trail, this triggered lively debate.
Particularly, we had [2] in Fedora, and there was even a change
proposal [3] to follow the decision upstream, but it was first delayed
and then never implemented [4]. Maybe Zbigniew (in CC), the owner,
could tell us why.
I'll take a look at those, thanks.
And maybe this topic could be resurrected in devel if there's
renewed
interest, but you are probably going to find opposition; because this
is fine for multi-user workstations, but it breaks things in servers.
IMHO, in a perfect world, each flavour (Workstation, Server, KDE...)
should decide and explicitly set the default option.
Sounds like a good idea. I don't see why it should break things in
servers, but maybe that's clarified in the links you mention.
The specific case that brought this to my attention was an apparent bug
in Evolution when running under KDE. With the default logind.conf a
session leaves gnome-keyring running after user logout, and a second
login often fails to start Evolution correctly, presumably because of a
race or something similar on the authentication DBus socket. But
there's no reason I can see to leave gnome-keyring running after
logout. It always runs as the user, so what's the point if the user has
gone? Changing logind.conf fixed this particular issue, but it seems
likely that the default behaviour could cause other problems.
poc