Huge memory usage in Fedora KDE (20GB+). Help finding leak?

Software & Operating Systems Linux
1

After working for a couple of days, primarily with Chrome and Pycharm open, I noticed things slowing down a bit. 21GB of memory was being used, reported by htop and free. Upon closing all programs, still 17.1GB was being used, but no process seems to be listed as using anything like that. I’m sure there are better ways of analysing memory usage than htop but I don’t know much about them, and some googling and looking idly at slabtop hasn’t enlightened me much.

Here’s a screenshot of htop. Can anyone help? Firing up a VM at this point (with 16gb of memory allocated) causes the machine to lock up entirely, leaving me unable to SSH in or ctrl-alt-F6 even.

Screenshot_20191118_205636
Screenshot_20191118_205636.png935×566 93.2 KB

kernel version 5.3.11-100.fc29.x86_64

slabtop:

Screenshot_20191118_210511

restarting plasma shell doesn’t free the missing RAM

2

I’ve noticed the GUI leaking on Fedora too, using both Cinnamon and Mate. I’ve made it a point to switch to runlevel 3 when I’m not using the console. The machines primary roles are file server and virtualization host.

3

That’s super unfortunate. I remember having leak issues with the redshift plasma widget, but that doesn’t seem to be it this time. Any idea how I could isolate the problem to a process or something?

4

HTOP has a sort function, here’s mine sorted by mem % … I am at runlevel 3 right now, haven’t been at the console in a few days.

image
image.png2016×1086 322 KB

Next time you notice high memory usage, try these commands as root and see if you get your memory back

switch to text console mode

# systemctl isolate multi-user.target

then to switch back to gui mode

# systemctl isolate graphical.target
5

Interestingly, switching run level like that caused me to be unable to log back in in GUI mode without restarting. Now I’ve done that, memory use is back to 1.4gb, but I’m not sure it’ll last.

6

So you’ve confirmed the leak is in the GUI system somewhere, hopefully someone with more experience can share with us how to narrow down where exactly.

1 Like
7

Have I? Restarting plasmashell didn’t recoup the missing memory, but rebooting the computer did. I still feel a bit like it could be anything. I guess I’ll try to keep track of memory use/programs used and see if I can notice anything significant happening

8

Oh, I misunderstood. I thought your memory use dropped when you switched to runlevel 3. Sounds like our problems are different. What video card & driver?

I have a GTX 1660 and using the nvidia binary driver, 430 I believe.

9

R9380 with Mesa drivers. I also have a GTX1080 in the system with vfio stub driver bound for VM passthrough, but I don’t reckon that’s the problem.

10

Aight I’mma say it: NUKE IT

11

You mean reinstall the OS entirely? I’d really, really like to avoid that, but I guess if it comes to that I should get on with it sooner rather than later. What makes you think this is such a complicated/not-worth-solving issue?

12

Oh it’s not the problem itself it’s me. I’m lazy.

But since you have more than one DE installed try to remove and purge the others except from kde :man_shrugging:

13

free -g

15

free -g shows that the memory is being used, but not what is using it?

1 Like
16

Could always use a memory dumper and see whats actually in there.

17

Looks like it’s been a thing for a while.

killall plasmashell && kstart plasmashell

That will give you a temporary reprieve.

3 Likes
18

Are you using the Oxygen theme? Some have reported that when going to Breeze or Breeze dark the consumption goes back to normal.

1 Like
19

I had the same idea, but weirdly, restarting plasmashell doesn’t free the memory

1 Like
20

I’m using Breeze Dark

1 Like
21

Are you using zfs ?

Some have expressed success with disabling application status, as well.

systemsettings -> Startup and Shutdown -> Service Manager -> uncheck “Status Notifier Manager”