On 14/11/2023 12:05, Ralph Corderoy wrote:
I'm surprised a Linux machine is configured to shutdown on persistent
high CPU or memory load. Rather, the kernel's out-of-memory killer will
kick in and guess a process to kill to free memory. It sometimes
nobbles the wrong horse. Perhaps it sent some vital process to the glue
factory which initiated an orderly shutdown, but I doubt it.
Hmmm. Several times prior to the big crash that trashed the machine, I
observed Chromium consuming CPU in all eight cores and all of the
available RAM including swap. I can't exactly remember what the process
was that was consuming the most resources, because it was a very
long-winded name. I've had another look at the output of htop with the
browser benign and there are dozens, if not hundreds of tasks listed (3
to 4 screenfuls)
What was clear was that as the CPU and RAM was consumed, the desktop
became less and less responsive until I was unable to do anything about
it. Previously I have hit the power button at this point, but on the day
in question the machine shut down on its own.
Could it be the graphics driver in the kernel which is running amok?
Chromium will be making it work hard by compositing lots of pixel
rectangles.
https://superuser.com/questions/1716854/disabling-gpu-compositing-for-all-chrome-and-electron-instances-on-a-mac
As with the Mac, it seems the only way to invoke the setting mentioned
is to launch Chromium in the console with the switch to disable GPU
Compositing. If all else fails, I'll try it and see what happens.
The biggest problem I've had is that this all happens so fast that I've
not been able capture any meaningful data to allow me to raise a bug
report on the Chromium bug tracker.
You might find some indication of past problems with ‘sudo -i
journalctl’. It will place you in less(1). There's the date to go by
and you can search with ‘/’, e.g. ‘oom-killer’.
There was only one instance of oom-killer:
Nov 02 08:46:48 OptiPlex kernel: systemd invoked oom-killer:
gfp_mask=0x140cca(GFP_HIGHUSER_MOVABLE|__GFP_COMP), order=0, oom_score_adj=0
That is some time before all this happened.
--
Terry Coles
--
Next meeting: Online, Jitsi, Tuesday, 2023-12-05 20:00
Check to whom you are replying
Meetings, mailing list, IRC, ... http://dorset.lug.org.uk
New thread, don't hijack: mailto:dorset@mailman.lug.org.uk