Abnormal memory usage

develux · 2025-06-03 18:17:49

topcat01 wrote:

Can you revert this temporarily, reboot, and then post meminfo with no graphics running?

Right after reboot with only plasma running: http://0x0.st/8YYv.txt

After doing

systemctl isolate multi-user.target

http://0x0.st/8YYw.txt

Last edited by develux (2025-06-03 18:23:11)

seth · 2025-06-03 19:22:04

Drops from ~690 to ~660 MB unaccounted for.
Test @topcat01's suggestion, though nb. that "Total - Available" is the active memory (+kernel slabs, but they're usually not much)

topcat01 · 2025-06-03 19:52:21

I have found that activating the zram related sysctls bumps the kernel memory usage (see, for example, https://unix.stackexchange.com/question … onsumption). I think it also increases the kernels internal memory thresholds, which might not be properly reflected in procinfo.

develux · 2025-06-04 02:17:05

seth wrote:

Test @topcat01's suggestion

Already disabled those sysctls in previous post, will continue monitoring today

develux · 2025-06-04 04:23:46

With zram rules disabled and plasma not running: http://0x0.st/8YIQ.txt
Still using +-2.5gb without anything running in tty

seth · 2025-06-04 05:21:41

More like 3.7 GB but

Active:          1406972 kB
Inactive:        1138252 kB
Slab:            1398472 kB

they're accounted for and

Active(file):    1161956 kB
Inactive(file):  1138252 kB
SReclaimable:    1193352 kB

most of that is in reclaimable kernel memory and file caches

Active(anon):     245016 kB
Inactive(anon):        0 kB

Only around 240MB are actually actively used by something.

Stop looking at data aggregators or at least look at "free -h" which lists file caches separately.

Wrt #27 it will also be perfectly normal that the GPU allocates *some* memory, the question is whether the amount of unaccounted memory (significantly) increases over time and then also cannot be (largely) reclaimed by temporarily allocating a lot of RAM while there's no VRAM demands.

develux · 2025-06-06 07:52:24

Seems this problem randomly disappeared when I switched to 6.15.0-1-mainline. I had 1.8gb used after entering plasma session
And it is using 2gb ram when everything is closed (except koi, yakuake, easyeffects, openfreebuds and kde connect)

Last edited by develux (2025-06-06 07:54:04)

seth · 2025-06-06 12:10:55

when everything is closed (except koi, yakuake, easyeffects, openfreebuds and kde connect)

You do understand that all of that +the GUI +the GPU /will/ require some RAM?
Plasma is morbidly obese desktop environment. It /will/ take some of your resources.
The thing you're looking for is, as pointed out, whether some memory isn't accounted for.

develux · 2025-06-06 13:18:59

seth wrote:

You do understand that all of that +the GUI +the GPU /will/ require some RAM?

Yes. I meant that it is good compared to 3gb usage after entering plasma session on 6.14.9 kernel. Seems this memory issue is fixed with 6.15.0-1-mainline

topcat01 · 2025-06-06 20:15:54

As a test, I tried many (silly) things with a couple of my machines, but everything is behaving normally.

develux · 2025-06-07 17:31:32

Nah it wasn't fixed. Got this after few hours of usage http://0x0.st/8g4u.txt
Maybe I need to reinstall Arch

seth · 2025-06-07 19:36:43

MemTotal:       32235876 kB
MemFree:        10267832 kB
Active:          3306700 kB
Inactive:       14721832 kB
Slab:             546036 kB

Around 3GB are unaccounted for, most RAM is in

Inactive(file): 14386952 kB

inactive file caches (likely the indexer baloo)
If you cannot reclaim the lost RAM from the rescue target (ie. it's not GART/GTT) you're looking at some leak

Maybe I need to reinstall Arch

If the kernel is leaking memory, it will do so despite any re-installations.

lsmod

Have you tried eg. the LTS kernel?

develux · 2025-06-08 02:58:09

seth wrote:

Have you tried eg. the LTS kernel?

Not yet, will try today

develux · 2025-06-08 03:38:28

Got the same issue, no apps are opened at all, lts kernel. http://0x0.st/8gxn.txt

develux · 2025-06-08 06:21:36

seth wrote:

lsmod

http://0x0.st/8g3m.txt

seth · 2025-06-08 06:34:29

"no apps" is not a thing, again: plasma itself is morbidly obese.
There're ~1.7GB unaccounted for.
Do you use an overlay filesystem?? Why? for what? Of what size?

lsblk -f
df -h

develux · 2025-06-08 09:58:27

seth wrote:

Do you use an overlay filesystem?? Why? for what? Of what size?
lsblk -f
df -h

https://0x0.st/8ggi.txt https://0x0.st/8ggs.txt
Yes, I am using overlayfs for psd https://wiki.archlinux.org/title/Profile-sync-daemon

seth · 2025-06-08 10:05:27

And do you think that maybe this is were your RAM gets "lost"…?
(In case: that's a rethorical question, check the behavior w/ psd but I'm already betting your right arm on it…)

develux · 2025-06-08 11:33:54

seth wrote:

And do you think that maybe this is were your RAM gets "lost"…?
(In case: that's a rethorical question, check the behavior w/ psd but I'm already betting your right arm on it…)

I disabled psd service and got same ~4GB memory usage even in tty (http:0x0.st/8ggE.txt)

seth · 2025-06-08 12:07:46

3.3GB memory gap… for clarification, the "overlay" module wasn't loaded at the time?
Next to that and zram, xfs is really the only major outlier in that list… and then try

amdgpu.gttsize=512 amdgpu.gartsize=512

, as mentioned, you might run into GUI issues w/ plasma, but amdgpu should™ no longer be able to snatch more than 1GB
So if that constrains the leak, we can at least tell where it's likely going.

develux · 2025-06-08 16:33:14

seth wrote:

"overlay" module wasn't loaded at the time?

Don`t know really, but I disabled psd and then rebooted. Will try those parameters

seth · 2025-06-08 19:58:13

Also make sure that the overlay module isn't loaded - the memory simply taken by the overlayfs would be the most obvious explanation.
If it's not that nor amdgpu, nor zram, we'll have to consider a leak in xfs (because it's less commonly used this might rather go undetected)

topcat01 · 2025-06-08 20:19:24

I have been using xfs on Amazon Linux 2023 instances (so, relatively recent kernel) and even after heavy cpu, ram, and fs usage haven't noticed any leaks. These instances are up for weeks. Haven't tried it on Arch though.

develux · 2025-06-09 12:05:32

After adding limits parameters to cmdline: http://0x0.st/8g5m.txt
System was up for ~1 hour so not sure that i have enough memory usage here

seth · 2025-06-09 12:17:35

Seems ~1.4GB unaccounted for… keep an eye on it but it starts to look like it's just GART/GTT after all?

Arch Linux

#26 2025-06-03 18:17:49

Re: Abnormal memory usage

#27 2025-06-03 19:22:04

Re: Abnormal memory usage

#28 2025-06-03 19:52:21

Re: Abnormal memory usage

#29 2025-06-04 02:17:05

Re: Abnormal memory usage

#30 2025-06-04 04:23:46

Re: Abnormal memory usage

#31 2025-06-04 05:21:41

Re: Abnormal memory usage

#32 2025-06-06 07:52:24

Re: Abnormal memory usage

#33 2025-06-06 12:10:55

Re: Abnormal memory usage

#34 2025-06-06 13:18:59

Re: Abnormal memory usage

#35 2025-06-06 20:15:54

Re: Abnormal memory usage

#36 2025-06-07 17:31:32

Re: Abnormal memory usage

#37 2025-06-07 19:36:43

Re: Abnormal memory usage

#38 2025-06-08 02:58:09

Re: Abnormal memory usage

#39 2025-06-08 03:38:28

Re: Abnormal memory usage

#40 2025-06-08 06:21:36

Re: Abnormal memory usage

#41 2025-06-08 06:34:29

Re: Abnormal memory usage

#42 2025-06-08 09:58:27

Re: Abnormal memory usage

#43 2025-06-08 10:05:27

Re: Abnormal memory usage

#44 2025-06-08 11:33:54

Re: Abnormal memory usage

#45 2025-06-08 12:07:46

Re: Abnormal memory usage

#46 2025-06-08 16:33:14

Re: Abnormal memory usage

#47 2025-06-08 19:58:13

Re: Abnormal memory usage

#48 2025-06-08 20:19:24

Re: Abnormal memory usage

#49 2025-06-09 12:05:32

Re: Abnormal memory usage

#50 2025-06-09 12:17:35

Re: Abnormal memory usage

Board footer