Adding a related data point from Ubuntu 24.04.4.
This is a true *global kernel* OOM, complementary to the dbus-killed
scenario in the description.
System:
Ubuntu 24.04.4 LTS
GNOME Shell 46.0
libmutter 46.2-1ubuntu0.24.04.16
gnome-session-bin 46.0-1ubuntu4
systemd 255.4-1ubuntu8.16
X11/Xorg
NVIDIA 580.159.03
32 GB RAM + 8 GB swap.
Under heavy load the machine hit a true global kernel OOM.
gnome-shell's own protection (OOMScoreAdjust=-1000, shipped since GNOME
44) worked.
But, the kernel spared the shell and killed a browser renderer instead.
The whole X11 session still collapsed to the "please log out" screen one
second later:
kernel: systemd-oomd invoked oom-killer: ... oom_score_adj=-900
kernel: oom-kill:constraint=CONSTRAINT_NONE, ... global_oom,
task_memcg=.../session.slice/[email protected],
task=brave, pid=1306259, uid=1000
kernel: Out of memory: Killed process 1306259 (brave) ... oom_score_adj:300
systemd[user]: Started gnome-session-failed.service ... # +1 second
Two points relative to this report:
1. Protecting individual services by score isn't sufficient: here the shell WAS
protected and a non-critical app was correctly chosen, but the session died
anyway - so the instability isn't only about which core service gets a bad
score.
2. The killed app was accounted under
.../session.slice/[email protected],
not app.slice - it wasn't isolated into its own scope (cf. gnome-shell#6565).
That points at cgroup placement, not just OOMScoreAdjust values, as part of
the
fix - possibly relevant to enr0n's question about systemd vs. other packages.
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2089800
Title:
Ubuntu desktop is unstable under memory pressure due to undesireable
OOMScoreAdjust values
To manage notifications about this bug go to:
https://bugs.launchpad.net/lxd/+bug/2089800/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
