View Issue Details

IDProjectCategoryView StatusLast Update
0015811CentOS-7xorg-x11-serverpublic2019-02-14 10:12
Reporterrdrago 
PrioritynormalSeveritymajorReproducibilityrandom
Status newResolutionopen 
Product Version7.6.1810 
Target VersionFixed in Version 
Summary0015811: kdeinit4 [plasma-desktop] using 100% CPU with latest Xorg server
DescriptionHello.
Ever since we've updated our machines to Centos-7.5 and now Centos-7.6, I see all kdeinit4: plasma-desktop processes go into 99-100% CPU (at the same time), every few minutes, while the taskbar and desktop become completely unresponsive for up to 2 minutes, sometimes even more. During that time, apps and windows behave normally, as long as they are not minimized.
I have a server to which 5-15 users connect using Xvnc or ExceedOnDemand, each getting a whole remote KDE desktop and I can't tell what triggers it. It also happens on my Dell T1500, with an Nvidia NVS 310 card.

It doesn't seem to happen on Centos-7.4. Only later distro versions.
I ran an strace -p against the plasma-desktop process and it doesn't seem stuck on something.
Last time, it was actually going through some NFS mounts, possibly someone was searching using krusader. Then the strace showed this same line for a 10-30 seconds:

 stat("/etc/localtime", {st_mode=S_IFREG|0644, st_size=2195, ...}) = 0

Workarounds include: switching to tty2-6 and back to tty1 using ctrl+alf+fX; or killing and restarting plasma-desktop (doesn't always work properly), log out or reboot.
Those steps usually un-freeze the taskbar and desktop, and the plasma-desktop usage drops again.

Is this related to a known bug? I haven't found any recent similar reports.
Additional InformationKernel:
3.10.0-957.5.1.el7

Installed KDE:

kdeplasma-addons-4.10.5-5.el7
kde-workspace-4.11.19-12.el7
kde-baseapps-4.10/5-6.el7

Installed X server:
xorg-x11-server-Xorg-1.20.1-5.2.el7_6

All from official mirrors.
TagsNo tags attached.
abrt_hash
URL

Activities

rdrago

rdrago

2019-02-14 10:06

reporter   ~0033829

Correction!

I am able to reproduce it on Centos-7.4, 7.5 and 7.6:
1. start a KDE session
2. have 40+ NFS shares mounted, by listing them or any other means
3. have another NFS share mounted
4. notice plasma-desktop process run 100% CPU
5. notice in `strace -p` of the above mentioned process that KDE/plasma is re(reading) all previously mounted NFS shares (gets much worse when having 100+ NFS mounted)
To make matters even worse, multiple users with KDE sessions amplify the effect, as all their mounts are read too, whenever any of the users triggers mounting another NFS share (tested with a simple ls command)

Was able to reproduce this even on previous versions of xorg-x11-server-Xorg, all found in the base/update repos for the respective distro.
Don't know if it's a KDE bug, but apparently, I can only file bugs for newer KDE/plasma versions (5+).
rdrago

rdrago

2019-02-14 10:12

reporter   ~0033830

Note: I'm using automounter for all NFS shares.
I've also tested by mounting one of the shares via `mount`. The same thing happens, all plasma-desktop processes from all KDE sessions jump to 100% CPU (can be lower with small number of mounted shares) and all desktops and panels freeze, until the re-reading operation of all mount points is done (apparently done by each plasma-desktop/kdeinit4 PID)

Issue History

Date Modified Username Field Change
2019-02-11 12:03 rdrago New Issue
2019-02-14 10:06 rdrago Note Added: 0033829
2019-02-14 10:12 rdrago Note Added: 0033830