RFC: better systemd --user integration
#301
Conversation
intelfx
force-pushed
the
work/systemd-user
branch
from
038b071 to
68cd35e
Compare
Ignore $ppid if it is an instance of the systemd manager. This prevents `systemd --user` from being treated as a parent process of all non-double-forked `systemd --user` services.
Ideally, we'd do something more reasonable, like treating them similarly to system services and maybe even offering to restart some of them or at the very least printing them in a format suitable for copy-pasting, but for now just replace the process name with the service name and save them under a separate key to hopefully distinguish them from other processes under `systemd --user`.
intelfx
force-pushed
the
work/systemd-user
branch
from
68cd35e to
470397c
Compare
|
Can this be useful for #289? |
|
@Vladimir-csp yes, I believe it can. This needs a bit more UI work (the code I wrote piggybacks on existing data structures used for raw session processes, so it does not offer to perform a |
|
@liske ping, is there any chance this PR could get reviewed? There has been a new release since this PR was posted, so I'm assuming there must be some sort of problem with this code specifically? |
Sorry the security fixes for 3.8 have taken some time over the last few weeks and I was not able to take time for this more complex issue, yet. Scheduling it for the next release. |
|
Thanks. I just wanted to make sure this PR would actually be desired before I rebase it on top of the new changes. |
|
Systemd-managed user sessions with apps in units were already a reality in Gnome and at least some other major DEs. I helped a bit to make it happen in independent compositors also. |
|
I need to reschedule this for 3.10. Needrestart 3.9 will be release ASAP to fix the regression reported in #317. |
Right now, processes running under
systemd --userconfuse needrestart in several ways:systemd --useras the logical parent (which means that needrestart will bogusly reportsystemditself as the process that needs to be restarted);systemd --userinstance itself needs a restart, needrestart won't do anything about it;systemd --user(as opposed to an arbitrary process started in a user scope), needrestart won't use this information in any way.This PR offers a clean solution for (1) by disregarding the parent if it is a
systemd --userinstance (implemented via matching the cgroup of the parent against/init.scope$), and pretty hacky solutions for (2) and (3) (hence the RFC status).