Bug 164851 - ksysguard gives wild values after a few minutes
Summary: ksysguard gives wild values after a few minutes
Status: RESOLVED FIXED
Alias: None
Product: ksysguard
Classification: Applications
Component: general (show other bugs)
Version: unspecified
Platform: Ubuntu Linux
: NOR normal
Target Milestone: ---
Assignee: KSysGuard Developers
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2008-06-24 19:31 UTC by stgallag
Modified: 2008-12-05 01:26 UTC (History)
3 users (show)

See Also:
Latest Commit:
Version Fixed In:


Attachments
An example of the ridiculous values (51.40 KB, image/png)
2008-06-24 19:33 UTC, stgallag
Details
Same, but notice the sensor browser is empty (82.71 KB, image/png)
2008-06-24 19:39 UTC, stgallag
Details
A tar.gz of the strace and i/o (24.98 KB, application/x-tgz)
2008-07-09 03:04 UTC, stgallag
Details
tar.gz of requested SystemLoad and ProcessTable and ksysguardrc (1.21 KB, application/x-tgz)
2008-10-28 18:39 UTC, stgallag
Details
Another sample of the requested files (28.74 KB, application/x-compressed-tar)
2008-11-01 17:50 UTC, Oliver Putz
Details
1-min-load of 2000 (143.09 KB, image/png)
2008-11-02 17:24 UTC, info
Details
15-min-load of 60, temp 1500 degrees (164.50 KB, image/png)
2008-11-02 17:31 UTC, info
Details
The sgrd file (5.77 KB, application/x-ksysguard)
2008-11-02 17:36 UTC, info
Details

Note You need to log in before you can comment on or make changes to this bug.
Description stgallag 2008-06-24 19:31:15 UTC
Version:           4 (using KDE 4.0.80)
Installed from:    Ubuntu Packages
OS:                Linux

Ksysguard gives wildly ridiculous values for it's sensors. For example, it shows 200k+ processes, negative cpu usage, and a load average in the millions. What's more, it seems the sensors switch occasionally, as some return to reasonable values while others go wild.
Comment 1 stgallag 2008-06-24 19:33:45 UTC
Created attachment 25568 [details]
An example of the ridiculous values
Comment 2 stgallag 2008-06-24 19:39:24 UTC
Created attachment 25569 [details]
Same, but notice the sensor browser is empty

What's more, the disk fills on the lower-right oscillate between those values
and 0, 81, and 6 (respectively), on every refresh.
Comment 3 stgallag 2008-06-24 19:50:15 UTC
It's happened three times so far, but I have not done anything unique with them, just left them on.
I'm a programmer, and I'd like to help, but I know Python, not C (yet).
Is it possible that some list of sensors could be out of order, or an incorrect pointer of some kind? (Just a guess)
Comment 4 stgallag 2008-06-24 19:51:27 UTC
I have not looked at stdout for the last three, but I started it in a shell so if it happens again I might catch what caused it.
Comment 5 John Tapsell 2008-06-26 10:32:41 UTC
Thanks,

 I haven't seen this bug myself - I'm not sure what's happening.  I'll have a look in the code.
Comment 6 stgallag 2008-07-09 03:04:31 UTC
Created attachment 25972 [details]
A tar.gz of the strace and i/o

Okay, It happened again.

I monitored it input/output with another ksysguard and strace'd it (about 10
seconds or 5 refreshes).
Comment 7 Oliver Putz 2008-10-16 13:14:46 UTC
I am not exactly sure if this is related, but I think I have a similar problem: Sometimes the sensors still report the correct value, but scale completely wrong. I uploaded a screenshot where the network load sensor just goes off and scales from -1.5TiB/s to 105MiB/s instead of scaling from 0 to maybe 10MiB/s at http://www.oliver-putz.de/bugreports/Bug164851NetworkLoad.png
Comment 8 stgallag 2008-10-16 19:55:45 UTC
I speculate that the cause for your problem may have to do with the way Linux measures network usage. Linux counts the bytes that have been sent and received, but about every four gigabytes, the number overflows and returns to 0. So on the next interval, the code calculates a 4 gigabyte loss, making that negative spike.

Now that's just speculation because I don't know how Ksysguard calculates it, but it's not uncommon.
Comment 9 John Tapsell 2008-10-25 00:20:53 UTC
Could you attach your .kde/share/apps/ksysguard/ProcessTable.sgrd and .kde4/share/config/ksysguardrc  files please?
Comment 10 John Tapsell 2008-10-25 00:22:51 UTC
Opps, I mean ~/.kde4/share/apps/ksysguard/SystemLoad.sgrd   please - thank you :)
Comment 11 Oliver Putz 2008-10-25 10:57:29 UTC
John, attach a SystemLoad.sgrd from a session where ksysguard is spitting out crazy numbers or is one from a session where everything is tracked right fine too?
Comment 12 John Tapsell 2008-10-25 12:35:32 UTC
THe session where it goes crazy please
Comment 13 stgallag 2008-10-28 18:37:25 UTC
I'm not sure if you need those files from me too, but I attached them anyway, while running a session that went wild.
Comment 14 stgallag 2008-10-28 18:39:15 UTC
Created attachment 28211 [details]
tar.gz of requested SystemLoad and ProcessTable and ksysguardrc
Comment 15 Oliver Putz 2008-11-01 17:50:50 UTC
Created attachment 28266 [details]
Another sample of the requested files

Here are my files after ksysguard went crazy. The package includes:

ksysguardrc
ProcessTable.sgrd
SystemLoad.sgrd
A screenshot of the crazy ksysguard session
Comment 16 info 2008-11-02 17:10:38 UTC
I have this almost every time after ksysguard has been running for a while (sometimes after a few minutes, sometimes half an hour, never longer than an hour). I attach screenshots. Note how the same values are dosplayed inconsistently between different screens (on the second screenshot, the left and the right are identical exept for the update interval, so the left shows a longer timeseries).

Note how the average load (1min) is 20000 in one screen, but a reasonable ~1 in another. The 15minute load is 60, and the temperature is 1500 degrees.

Comment 17 info 2008-11-02 17:24:00 UTC
Created attachment 28281 [details]
1-min-load of 2000

see the top right panel.
Comment 18 info 2008-11-02 17:31:18 UTC
Created attachment 28282 [details]
15-min-load of 60, temp 1500 degrees

Note the bottom two panels (average load 15 mins, and temperature). Left and
right have exactly the same setting, only that the right panels have 5 pixels
per timestep, and the left ones 1 pixel per timestep (i.e. the right are
closeups of the left panels).
Also compare the second panel from the top with the other attachment. Both show the 1-min-load, here it is ok, in the other one it is badly wrong.

This is on a Fujitsu Amilo 1650G laptop with Fedora 9. The problem has been there since kde 4.0 back in April or May.
Comment 19 info 2008-11-02 17:36:33 UTC
Created attachment 28283 [details]
The sgrd file
Comment 20 John Tapsell 2008-12-05 01:26:27 UTC
SVN commit 892692 by johnflux:

Don't go crazy if told to reconfigure.
This fixes some very old bugs

BUG:99377
BUG:164851
BUG:176160
BUG:78138



 M  +3 -7      SensorAgent.cc  


WebSVN link: http://websvn.kde.org/?view=rev&revision=892692