Per this suggestion [1] by hakavlad (Alexey Avramov), I did a single
test with earlyoom, a user space service that's already packaged for
Fedora. I have not yet tested nohang, also mentioned.

I chose a configuration that has fairly consistently (>80%) resulted
in a total hang for more than 30m. And the result with earlyoom is
that while responsiveness in the GUI was still bad, the longest GUI
freeze lasted ~8 minutes ending in oom which is certainly a lot better
than a 30+ minute hang. It's entirely subjective, but my opinion is
the system was legitimately lost within the 1st minute of hang, and a
reasonable user can choose to give up at that point. Is this an
improvement? Yes, the oom happens sooner, and maybe more testing will
prove it's more predictable. Is it good enough? No. Should Fedora
enable earlyoom in Fedora 32 Workstation? Maybe.

Configuration:
CPU i7-2820QM
RAM: 7837M
swap on ZRAM (lz4): 7836M (there is no other swap)
BOOT_IMAGE=(hd5,gpt6)/boot/vmlinuz-5.3.0-0.rc6.git1.1.fc32.x86_64
root=UUID=72df6d5b-26d1-47ff-a9ab-33f6a0b2c4cf ro
rootflags=subvol=root log_buf_len=4M systemd.debug-shell=1
printk.devkmsg=on slub_debug=FZPU

Logs [2] [3] [4] all monotonic time, and screenshot [5]. The
screenshot monotonic time equivalent is ~ [ 4890.404675] which
coincides with GUI totally hung, and a sysrq+t issued via ssh. There's
a lot going on including some i915 weirdness that I've been informed
by upstream is swap related, but I don't know the significance of
these kernel/gnome-shell page allocation failure complaints (they
don't taint the kernel). But suffice to say there are questionable
things happening well before the oom kill is issued.


[1]
https://pagure.io/fedora-workstation/issue/98#comment-594295
[2]
Full journal
https://drive.google.com/open?id=1nnZupWFnGfqu41aYs-u3UL8ir180YiL_
[3]
dmesg only
https://drive.google.com/open?id=1FNc7e-XuIiIAzSdBkOg98x9jgYEQQWbr
[4]
earlyoom messages only
https://drive.google.com/open?id=1drd570PRbUiiSnCwoP26DMSgSo93SdAC
[5]
screenshot shows top and iotop at the time of heavy CPU, IO, memory
and swap pressure
https://drive.google.com/open?id=1unLv11HmHW3bYOJvlLSuZYSiv6gyw7L3
_______________________________________________
devel mailing list -- devel@lists.fedoraproject.org
To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Fedora Code of Conduct: 
https://docs.fedoraproject.org/en-US/project/code-of-conduct/
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org

Reply via email to