Re: EarlyOOM +ZRAM Only

[Sergio Belkin <sebelk@xxxxxxxxx>]

El vie., 14 ago. 2020 a las 9:28, Przemek Klosowski via devel (<devel@xxxxxxxxxxxxxxxxxxxxxxx>) escribió:

On 8/14/20 7:33 AM, John M. Harris Jr wrote:
> On Wednesday, August 12, 2020 1:16:34 PM MST Przemek Klosowski via devel
> wrote:
>>
>> This is weird---your swap was 100% full, and ram almost full, and yet
>> killing 4GB VirtualBox didn't seem to free up memory. I suspect some
>> sort of measurement or reporting error---if these numbers were accurate,
>> EarlyOOM did have a reason to panic and kill zoom. BTW, zoom taking 721
>> MiB is crazy.

> Are you kidding? The system still had over a quarter of a gigabyte of free
> RAM. There's no reason to start killing off processes at that point. That's
> tons of free memory. To put that into perspective, that's enough free memory
> to store over 1000 average-length novels directly in memory.

When the swap is 100% full, it is not like you have a bunch of processes
neatly stored in it, and some free memory available---you have a mess of
partly swapped out processes reading back their pages from swap and
pushing other processes' RAM pages onto the fragmented swap, when they
are trying to run.

This is the worst case of disk usage, and as we discussed before, the
transfer speeds for such traffic will be hundreds/thousands times
slower---I would expect latencies going into tens of seconds. So, no, I
am not kidding.
_______________________________________________

2 comments:

- I don't use disk-based swap, only zram.

- It happened again, and in this case there is no Virtual Machine nor Zoom app running:

ago 14 15:08:37 dublin.ireland.home earlyoom[888]: sending SIGTERM to process 2052260 uid 1000 "Web Content": badness 322, VmRSS 447 MiB
ago 14 15:08:40 dublin.ireland.home earlyoom[888]: process exited after 2.7 seconds
ago 14 15:12:15 dublin.ireland.home earlyoom[888]: mem avail:   395 of 15887 MiB ( 2.49%), swap free:    0 of 4095 MiB ( 0.00%)
ago 14 15:12:15 dublin.ireland.home earlyoom[888]: low memory! at or below SIGTERM limits: mem  2.52%, swap 10.00%
ago 14 15:12:15 dublin.ireland.home earlyoom[888]: sending SIGTERM to process 2055755 uid 1000 "Web Content": badness 319, VmRSS 392 MiB
ago 14 15:12:17 dublin.ireland.home earlyoom[888]: process exited after 2.8 seconds
ago 14 15:28:35 dublin.ireland.home earlyoom[888]: mem avail:   371 of 15887 MiB ( 2.34%), swap free:    0 of 4095 MiB ( 0.00%)
ago 14 15:28:35 dublin.ireland.home earlyoom[888]: low memory! at or below SIGTERM limits: mem  2.52%, swap 10.00%
ago 14 15:28:35 dublin.ireland.home earlyoom[888]: sending SIGTERM to process 2062157 uid 1000 "Web Content": badness 327, VmRSS 553 MiB
ago 14 15:28:37 dublin.ireland.home earlyoom[888]: process exited after 2.3 seconds

Perhaps this ps_mem snippet is useful:

<snip

Private  +   Shared  =  RAM used       Program

</snip>

....

<snip>

294.8 MiB +   4.7 MiB = 299.4 MiB       telegram-desktop.bin
290.3 MiB +  22.1 MiB = 312.4 MiB       rocketchat-desktop (5)
323.8 MiB +   9.5 MiB = 333.2 MiB       kwin_x11 (9)
344.5 MiB +   1.4 MiB = 345.8 MiB       nextcloud
373.2 MiB +  21.2 MiB = 394.4 MiB       spotify (5)
416.6 MiB +   1.3 MiB = 417.9 MiB       plasma-discover
448.5 MiB +  16.6 MiB = 465.2 MiB       MainThread
892.8 MiB + 444.5 KiB = 893.2 MiB       packagekitd
  1.0 GiB +   8.6 MiB =   1.0 GiB       plasmashell
  1.9 GiB +  81.0 MiB =   1.9 GiB       Web Content (9)

---------------------------------
                          9.7 GiB
=================================

</snip>

Well I'll report soon, any  idea (remember I have 16 GB of RAM with 4G of zram-based swap) will be welcome

--

--
Sergio Belkin
LPIC-2 Certified - http://www.lpi.org

_______________________________________________
devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx
To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx
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@xxxxxxxxxxxxxxxxxxxxxxx