On 4/12/23 10:26, Max Nikulin wrote:
> I am curious if this creature may provide a summary on user-space OOM
> killers. I have never tried them, but I expect that they may be more
> intelligent than the kernel-space one. I have seen mentions of the >
following ones: earlyoom, nohang, oomd.
I think I've found a potential culprit using about:processes
https://openai.com 110% CPU
(That's GPT4)
Firefox at this time is running around 1GB memory
The super high CPU for openai is certainly a whole bunch of javascript.
I closed that tab and the CPU load dropped to normal. Reloading the
openai website and CPU oscillated between 1% and 20% till it settled
around 1%
I have also been monitoring global memory use. The last entry below is
after I closed the openai tab
root@client:~# date; free -m
Mon 04 Dec 2023 09:14:29 AWST
total used free shared buff/cache
available
Mem: 32023 12400 10521 426 10136 19622
Swap: 976 8 968
root@client:~# date; free -m
Mon 04 Dec 2023 10:04:56 AWST
total used free shared buff/cache
available
Mem: 32023 12913 9959 437 10197 19110
Swap: 976 8 968
root@client:~# date; free -m
Mon 04 Dec 2023 10:28:19 AWST
total used free shared buff/cache
available
Mem: 32023 13113 9732 419 10206 18910
Swap: 976 8 968
root@client:~# date; free -m
Mon 04 Dec 2023 10:32:58 AWST
total used free shared buff/cache
available
Mem: 32023 12545 10297 428 10217 19478
Swap: 976 8 968
This does not explain the OOM problem, but it seems memory is slowly
being consumed at a rate where in 24 hours I'll start seeing problems again.