Version: unspecified (using KDE 4.5.5) OS: Linux dbus-daemon hits 1024 file descriptors fairly frequently (and thus wedges up several other things). Generally I log out and back in to start a new dbus-daemon, but I've discovered that kpackagekiticon also had almost 1000 fds, and killing it reduced dbus-daemon's open fds by about that number, to below 100: cd /proc; do echo $i `ls $i/fd 2>/dev/null|wc`;done | awk '$2 > 300 { print $0 }' 2886 1024 1024 4010 # dbus-daemon, see below 3960 959 959 3726 # kpackagekiticon, see below user 3960 1 0 Feb08 ? 00:00:25 /usr/libexec/kde4/kpackagekitsmarticon user 2886 1 13 Feb08 ? 15:24:39 /bin/dbus-daemon --fork --print-pid 5 --print-address 7 --session [root@host proc]# kill 3960 [root@host proc]# kill 3960 bash: kill: (3960) - No such process [root@host proc]# uptime 08:36:10 up 4 days, 14:27, 10 users, load average: 0.49, 0.86, 1.29 [root@host proc]# ls /proc/2886/fd | wc # dbus-daemon now many fewer fds open 80 80 232 RPM versions are: dbus-1.4.0-2.fc14.x86_64 kpackagekit-0.6.3.3-1.fc14.x86_64 Reproducible: Always Steps to Reproduce: Reproducibility of many-fds in dbus-daemon "happens every time". This is the first time I've killed kpackagekiticon to fix it, but it was a pretty definite fix. Actual Results: Normal desktop use ~4days, dbus-daemon open fds goes to 1024 (mostly sockets in /tmp/dbus[stuff]), kpackagekiticon open fds goes to nearly 1000. Kill kpackagekiticon, dbus-daemon back to normal. I've been a Linux sysadmin for a long time and this is my personal desktop. You can ask me to try fairly odd things :-) (debug package? restart kpackagekiticon and strace it?).
I encountered the same bug with KDE 4.5.95 (openSUSE 11.4 milestone 6). After running normally for about 24 hours, kpackagekitsmarticon starts opening a new socket every 6 minutes. This is consistent with the observation by the original poster that it takes ~4 days to hit 1024 open file descriptors.
I can confirm this bug. After a while the KDE session behaves strangely, screen lock breaks, new apps won't start, etc. Top shows that Dbus is eating 100% of a core. Killing kpackagekitsmarticon manually resolves the issue. Thanks for the workaround suggestion, and I can too try out debug packages.
I experienced this as well, after leaving my desktop on through a weekend . Kde 4.6.1 from fedora KDE testing repo, x86_64. Killing kpackagekitsmarticon returned dbus to normal.
Hmmm this is probably why when I test kpackagekit on fedora it stops working after a while. To be honest I don't know how to close a DBus connection since this seems to be related to a DBus limit on how many questions should be asked on the bus. Of course the code to look for a fix is at packagekit-qt and not kpackagekit, I'm out of time these days but as soon as I get some free time I'll try to investigate.
Bug still occurs in Fedora 15 following upgrade to original reporting system: dbus-1.4.6-4.fc15.x86_64 kpackagekit-0.6.3.3-2.fc15.x86_64
*** This bug has been marked as a duplicate of bug 261180 ***
This issue was set as resolved however I cannot find the version in which it is expected to be fixed. I'm currently using kde v4.6.2 and I'm still experiencing this issue. Could anyone provide an update? Thanks!
I'm not too sure if Aaron backported it to KDE 4.6.x but surelly KDE 4.7 will have the fix