SUMMARY When the system is heavily loaded (in this case compiling GCC at nice 19 while running No Man's Sky at nice -8), plasma-systemmonitor "processes" table becomes so unresponsive as to be useless for it's intended purpose. Opening plasma-systemmonitor from the M+Esc shortcut takes >5s. Resizing the application window is incredibly laggy, with obvious redraw. Ditto for anything else that updates/rearranges any part of the process table (i.e. the default page on open). Opening the "tools" menu takes ~10s for the highlight selection, then another ~15s waiting for the click to register. Switching to one of the two chart tabs I have enabled restores UI perf, this appears to be down to the "processes" page specifically. Closing plasma-systemmonitor while the "processes" page is focused takes another ~15s, long enough to trigger the "application not responding" dialog. Other KDE applications (e.g. Dolphin) appear unaffected, and their UIs perform as expected in this scenario. The browser I am using to post this (firefox) is running just fine as well. STEPS TO REPRODUCE 1. Load up both GPU and CPU. 2. Try to use plasma-systemmonitor. 3. Cry for ksysguard, which ran blazing fast while having a better UI, more features, and fewer problems. OBSERVED RESULT "System monitor" is useless for the purpose of monitoring a heavily loaded system, particularly if the GPU is busy. EXPECTED RESULT Any "system monitor" application is light enough on system resources itself that it can be used monitor the resource use of other processes, even when the system is under load. SOFTWARE/OS VERSIONS KDE Plasma Version: 6.1.5 KDE Frameworks Version: 6.5.0 Qt Version: 6.7.2 Graphics Platform: X11 Kernel: 6.6.47-gentoo-dist ADDITIONAL INFORMATION I suspect this may well be GPU related, as NMS is pegging my (RX6700XT) GPU at ~97%, but total CPU load is only 50-60%. A "system monitor" that's GPU-bound is kinda ridiculous though if you ask me, so...?
Screen recording (~23MB): https://damnation.dynu.net/nextcloud/s/TQ9Q8KGwGcM2g6d/download/foo-2024-09-24_18.40.35.mkv
After some mucking around, I take back the comments on GPU load. While it does make the situation worse, plasma-systemmonitor performance really just sucks in general, and the UI becomes unusable if there is significant CPU-time contention of any kind... i.e. the exact opposite of what one expects from a class of tool primarily used to diagnose performance problems. Add this one to the long list of bugs, jank and missing functionality, cross reference countless threads on reddit et al. and that big discuss thread that was locked because it was "causing stress"... And it quickly becomes apparent that pretty much everyone outside KDE/Plasma developers hates this thing, because it's demonstrably and consistently worse than ksysguard in every metric that actually matters to an end-user. To put this another way, for all the (ex)windows types around here these days: Compare performance of the Windows Task Manager... Go on, I dare you. Can we please, please return to shipping a system resource monitor that is fit for purpose? (*cough* https://github.com/zvova7890/ksysguard6). Ksysguard was the darling of many a power-user for over 15 years, and not only has it been deprecated in favour of something missing most of those "power-user" features, that replacement isn't even usable under the very situation where one might want a system resource monitor in the first place. If anyone has suggestions on making this thing usable when it matters, I'm all ears. Otherwise I think I'll just go write an ebuild for ksysguard6, because it's becoming abundantly clear that nobody here cares. Yes, I realise that with GPU contention (mostly) off the hook this report is probably yet another duplicate. At this point I don't care either.
*** This bug has been marked as a duplicate of bug 479321 ***