plasma-systemmonitor in kdereview

[Arjen Hiemstra]

On Thursday, 1 October 2020 14:11:16 CEST Harald Sitter wrote:
> On 01.10.20 11:36, Arjen Hiemstra wrote:
> > Hello,
> > 
> > I'd hereby like to announce that plasma-systemmonitor is in kdereview. It
> > can be found at https://invent.kde.org/plasma/plasma-systemmonitor .
> > 
> > plasma-systemmonitor is a new system monitor UI built with Kirigami. It
> > makes use of the ksystemstats daemon and the faces system for system
> > monitor plasmoids that were both introduced in Plasma 5.19.
> > 
> > Our current plan is to do a "preview release" alongside Plasma 5.20, then
> > have it be an official part of Plasma with 5.21.
> 
> Cool stuff.
> 
> L10n is currently a bit incomplete.
> Notably
> - the pages files lack any localization at all and I'm also not sure how
> those could be best localized.

Yeah that is a good point. If we can somehow extract the strings from these 
files I think we can make it work. But that may need some custom scripting.

> - the 'configure columns' feature doesn't load the translated strings
> into the checkboxes

Huh, weird. They are being populated by i18n calls, wonder what's going on 
there.

> 
> On the subject of configure columns I feel like word wrapping would look
> better than eliding there https://i.imgur.com/RbAeUhH.png
> 
> Something is astray with memory calculation. See konsole which is
> somehow listed as 2.6gib https://i.imgur.com/MjIpj3O.png but its actual
> children aren't getting nowhere near that number. I'm guessing it's
> because even detached processes (notably vscode here) are in the same
> cgroup? If so we probably can't do much about this?
> 

There's two things going on here I expect: everything started from konsole 
will be part of the konsole cgroup unless started through kstart5 or a similar 
tool that explicitly moves things to a new cgroup. I doubt this is solvable 
without patching all major shells. 

The other thing is that the "memory" column here is what is called "Total 
Memory" in KSysGuard, which is the process' private memory with the shared 
memory added divided over the processes that share that memory. This provides 
a more accurate number for the actual memory usage of the process, but does 
mean that it will most likely seem like it suddenly uses more memory, since 
the "memory" in KSysGuard is only the private memory.

> The table highlights are getting messed up when new apps appear/disappear.
> New app appears: https://i.imgur.com/tacWeoI.png mind that suddenly the
> ksysguard memory cell is selected as well. More or less the same happens
> when a process disappears. The cells seem somewhat random though Memory
> is almost always a highlighted one.

Ugh, I thought we fixed that. The trouble here is that QML's TableView has a 
bug where certain bindings aren't properly reevaluated. We have a workaround 
in place but apparently it doesn't work in all cases. :/

> 
> Sometimes the CPU column shows nan% for new apps. Happens a lot with
> bustle for me. In fact, sometimes I have two entries one of which is
> nan% for all eternity https://i.imgur.com/HLavLsI.png

The lingering one is odd, but may have something to do with cgroups not 
getting cleaned up properly. If you can reproduce and provide the output of 
systemd-cgls that would be helpful.

> 
> Possibly related: when editing the Processes page and discarding all
> changes the R/W columns show negative values for a second or two.
> https://i.imgur.com/amrfewG.png

Yeah I've seen that happen as well, it has something to do with the underlying 
ProcessCore processes not being updated yet I expect. This probably also 
causes the NaN above, since the application table currently uses the sum of 
all values of the processes to display values.

> 
> Also the columns in the processes page lack labels.

Gah. And applications works correctly?

> 
> HS