https://bugs.kde.org/show_bug.cgi?id=485024
Bug ID: 485024
Summary: systemd watchdog stops kwin_wayland during oom
Classification: Plasma
Product: kwin
Version: git master
Platform: Other
OS: Linux
Status: REPORTED
Severity: normal
Priority: NOR
Component: general
Assignee: kwin-bugs-n...@kde.org
Reporter: sit...@kde.org
Target Milestone: ---
SUMMARY
While the system is trying to fix an oom situation there is a good chance (so
long as the system is responsive enough) that the systemd unit watchdog
triggers and terminates kwin_wayland, thereby nuking the session.
Apr 04 13:28:07 ajax systemd[1007]: plasma-kwin_wayland.service: Watchdog
timeout (limit 15s)!
Apr 04 13:28:07 ajax systemd[1007]: plasma-kwin_wayland.service: Killing
process 1133 (kwin_wayland_wr) with signal SIGHUP.
...
Apr 04 13:28:11 ajax systemd[1007]: plasma-kwin_wayland.service: Failed with
result 'watchdog'.
Apr 04 13:28:11 ajax systemd[1007]: Stopped KDE Window Manager.
STEPS TO REPRODUCE
1. run out of memory. have many VMs, or a leaky app, or just too much stuff
open
2. system becomes slightly unresponsive
OBSERVED RESULT
you get thrown to sddm because your session has been stopped by the watchdog
EXPECTED RESULT
oom handling should be allowed to take place and kill a suitable client. note
the additional info below though
SOFTWARE/OS VERSIONS
KDE Plasma Version: 6.0.80
KDE Frameworks Version: 6.1.0
Qt Version: 6.6.3
Kernel Version: 6.8.2-arch2-1 (64-bit)
Graphics Platform: Wayland
Processors: 12 × AMD Ryzen 5 3600X 6-Core Processor
Memory: 31.2 GiB of RAM
Graphics Processor: AMD Radeon RX 5700 XT
ADDITIONAL INFORMATION
In a way this is actually useful because it likely terminates what was causing
the oom situation in the first place, not sure if we can make this a feature
somehow. Maybe instead of terminating, wildly shoot at clients on SIGHUP? But
then I suppose we'd be implementing yet another user space oom handler. Indeed
perhaps we can consider this a feature? If the oom handler hasn't been able to
fix the situation in 15 seconds it probably won't for a lot longer. That said,
we definitely need to tell the user what happened.
--
You are receiving this mail because:
You are watching all bug changes.
