WCPU footprint is large.

[anonymous]

The CPU footprint of Gajim is rather large on FreeBSD. Would it be possible to reduce it?

FreeBSD hovel.york.ac.uk 6.0-STABLE FreeBSD 6.0-STABLE #5: Wed Apr 19 08:47:44 BST 2006 root@…:/usr/obj/usr/src/sys/GENERIC i386

last pid: 27796; load averages: 0.84, 0.90, 0.76 up 27+00:33:09 09:36:26 85 processes: 6 running, 78 sleeping, 1 zombie CPU states: 91.0% user, 0.0% nice, 8.6% system, 0.4% interrupt, 0.0% idle Mem: 209M Active, 171M Inact, 93M Wired, 16M Cache, 60M Buf, 4160K Free Swap: 512M Total, 49M Used, 463M Free, 9% Inuse

PID USERNAME THR PRI NICE SIZE RES STATE TIME WCPU COMMAND

27781 yann 1 123 0 35840K 29320K RUN 1:11 88.58% python 17422 yann 1 97 0 111M 87524K RUN 199:07 2.20% Xorg

[dkirov]

Yes, I just e-mailed the maintainer. This bug is already fixed in svn and

he have to apply the patch. http://trac.gajim.org/changeset/6314?format=diff

[antemil]

Although this was fixed, I realize this is happening again.

The Gajim About dialog: Gajim 0.11.1 A GTK+ jabber client GTK+ Version: 2.10.14 PyGTK Version: 2.10.6

antemil@caleac:~$ uname -a FreeBSD caleac 6.2-STABLE FreeBSD 6.2-STABLE #1: Wed May 9 20:05:13 CEST 2007 root@…:/usr/src/sys/amd64/compile/MYKERNEL amd64

antemil@caleac:~$ top last pid: 10360; load averages: 3.04, 1.97, 1.84 up 0+03:27:30 12:11:37 83 processes: 5 running, 78 sleeping CPU states: 81.2% user, 0.0% nice, 18.4% system, 0.4% interrupt, 0.0% idle Mem: 324M Active, 295M Inact, 110M Wired, 5916K Cache, 111M Buf, 204M Free Swap: 1024M Total, 1024M Free

PID USERNAME THR PRI NICE SIZE RES STATE TIME WCPU COMMAND

10325 antemil 1 132 0 152M 67332K RUN 2:33 94.20% python2.4

X Window System Version 7.2.0 Release Date: 22 January 2007 X Protocol Version 11, Revision 0, Release 7.2 Build Operating System: FreeBSD 6.2-STABLE amd64

The provided old patch is just the actual code now. I tried python2.5 too (in fact, I get back to 2.4 trying to avoid this behaviour). Any suggestion about how to know which part of gajim is really eating the CPU?

[asterix]

could you try to add 56 to the list in src/common/proxy65_manager.py line 256:

if errnum in (errno.EINPROGRESS, errno.EALREADY, errno.EWOULDBLOCK, 56):

does that help ?

see #3396 and [8572]

[antemil]

Replying to asterix:

could you try to add 56 to the list in src/common/proxy65_manager.py line 256: {{{ if errnum in (errno.EINPROGRESS, errno.EALREADY, errno.EWOULDBLOCK, 56): }}} does that help ? see #3396 and [8572]

Using Gajim 0.11.1.0 with that modification solves the problem:

last pid: 76040; load averages: 1.28, 1.68, 1.60 up 0+20:42:38 16:36:06 93 processes: 2 running, 90 sleeping, 1 zombie CPU states: 1.5% user, 95.9% nice, 1.1% system, 1.5% interrupt, 0.0% idle Mem: 352M Active, 368M Inact, 125M Wired, 40M Cache, 111M Buf, 1736K Free Swap: 1024M Total, 72K Used, 1024M Free

PID USERNAME THR PRI NICE SIZE RES STATE TIME WCPU COMMAND

70931 boinc 2 171 52 78660K 70172K RUN 9:48 89.11% setiathome-5. 57634 antemil 1 96 0 336M 73920K select 1:41 2.83% Xorg 70892 antemil 1 96 0 90448K 50300K select 2:40 0.88% firefox-bin 76024 antemil 1 4 0 199M 68392K kqread 0:05 0.49% python2.5

Thanks!

[asterix]

(In [8585]) fixes large WCPU footprint under freebsd. fixes #1963

[roidelapluie]

(In [8587]) fixes large WCPU footprint under freebsd. fixes #1963 for 0.11.2.

[f@…]

This is a bad idea, under freebsd, 56 is EISCONN, so you probably do not want to retry. With this patch my gajim just hangs, resolving one proxy over and over.

[asterix]

ok so EISCONN was handled in the next if. do you see large WCPU footprint if you simply reverse this patch ?

[f@…]

No, nor have I seen it before the patch was applied.

[antemil]

Without the patch

[antemil]

With the 56 patch

[antemil]

I generated two profile logs using Gajim 0.11.1.0, the first without the patch, and the second with the 56 at line 256. Without the patch, WCPU tends to 100% while with the patch, WCPU tends to 0%

I hope this files to be someway useful. I ordered them using internal time, and generated it with the following command line:

....gajim/src $ python -m cProfile -s 'time' ./gajim.py

[asterix]

I reverted last commit for 0.11.2 as it makes Gajim hangs. We'll see if a better solution is found

[asterix]

I don't understand why behaviour is diferent under BSD. Gajim doesn't use 100% CPU under linux ...

[asterix]

I don't have a BSD installed, so I can't test, but I don't see why behaviour under BSD would be different...

[anonymous]

I use FreeBSD (7.0) too and I still have the same problem.

Now I disabled the function "do_connect" (return at first line) and it works fine.

What is the task of this function ?