indicator-sound-service eating CPU on second user login

Yes, it's reproducible--at least in this session.

This affects me as well. Here is what strace says:

open("/dev/shm/pulse-shm-87918442", O_RDONLY|O_NOFOLLOW|O_CLOEXEC) = 10
fstat(10, {st_mode=S_IFREG|0400, st_size=67108904, ...}) = 0
mmap(NULL, 67112960, PROT_READ, MAP_SHARED, 10, 0) = 0x7dd86cb71000
close(10) = 0
kill(5289, SIG_0) = 0
munmap(0x7dd86cb71000, 67112960) = 0
open("/dev/shm/pulse-shm-4214229396", O_RDONLY|O_NOFOLLOW|O_CLOEXEC) = 10
fstat(10, {st_mode=S_IFREG|0400, st_size=67108904, ...}) = 0
mmap(NULL, 67112960, PROT_READ, MAP_SHARED, 10, 0) = 0x7dd86cb71000
close(10) = 0
kill(5289, SIG_0) = 0
munmap(0x7dd86cb71000, 67112960) = 0

Also, the process as allocated 500M of physical memory (1.7TB virtual) and is spinning on one CPU. The problem *might* be related to the gmail checker plugin in Chrome. At least I didn't notice the problem before installing that extension.

I don't have Chrome installed, myself. Confirming due to second report.

I am also experiencing this problem using nx client/server. It happened multiple times in 1 session.

It affects me - on a dual processor it causes about roughly full use on one processor and causes indicator-applet to use most of the remaining CPU!

Note that signing out of the second user and returning to the first user "cures" the problem...

PS I don't have anything else but the calendar,date/time, wireless and pulse-audio in indicator.
(No gmail etc)

Killing the looping processes does not help.

I also tried re-installing the indicator-applet, which didn't help.

Jeremy

Okay i cannot reproduce this but I have heard about this issue since Lucid. Whereby when indicator-sound-service attempt to connect the pulse daemon and for whatever reason the daemon is not running but it should be, the respawn behaviour of pulse goes into overdrive and cripples a CPU. I am going to assign this also to pulse because indicator-sound does not do anything unorthodox there.

There's already a task for indicator-sound.

If it is a problem with pulseaudio, we should confirm before making an upstream task.

Connor: if you can't reproduce, how do you know what the underlying problem is? Do you have a source for this?

Sorry David,

My mistake.

The service was trying to recreate a pulse connection again if it had failed even though the initial attempt had the reconnect flag. Will be fixed in 0.4.8.

Will be released in lucid as part of 0.2.4

Cool, thanks! Sorry about invalidating your upstream task, didn't realize that you _were_ upstream. :)

Please provide an SRU for maverick as well, so that this can be moved to -updates after verification.

Accepted indicator-sound into lucid-proposed, the package will build now and be available in a few hours. Please test and give feedback here. See https://wiki.ubuntu.com/Testing/EnableProposed for documentation how to enable and use -proposed. Thank you in advance!

Accepted indicator-sound into maverick-proposed, the package will build now and be available in a few hours. Please test and give feedback here. See https://wiki.ubuntu.com/Testing/EnableProposed for documentation how to enable and use -proposed. Thank you in advance!

@Martin: maverick-proposed update fixes it for me, thank you!

verification done for maverick
verification needed for lucid

Verifying for Lucid.

I've been seeing this in LTSP thin client labs, where all user sessions run on the server. Sometimes there were more than 10 indicator-sound processes (one for each user) eating all the CPU.

I've not been able to reproduce it after upgrading to the version in lucid-proposed.

Thanks for testing. Marking as verification-done for lucid.

This bug was fixed in the package indicator-sound - 0.2.5-0ubuntu1

---------------
indicator-sound (0.2.5-0ubuntu1) lucid-proposed; urgency=low

  * New upstream release.
    - Fixed high CPU load when there is problems connecting to
      pulse (LP: #581173)
 -- Ken VanDine <email address hidden> Tue, 05 Oct 2010 10:18:07 -0400

This bug was fixed in the package indicator-sound - 0.4.9-0ubuntu1

---------------
indicator-sound (0.4.9-0ubuntu1) maverick-proposed; urgency=low

  * New upstream release.
    - fixes tranlation mpris name regression (LP: #654140)

indicator-sound (0.4.8-0ubuntu1) maverick-proposed; urgency=low

  * New upstream release.
    - indicator-sound-service eating CPU on second user login (LP: #581173)
    - Right border of transport control is different colour while using
      Ambiance (LP: #632774)
    - Incorrect ellipsing (LP: #637580)
    - always show unmuted at startup (LP: #648238)
    - click or enter on metadata widget does not copy contents to
      clipboard (LP: #651246)
    - Sound menu has two Rhythmbox elements when using
      translations (LP: #654140)
 -- Ken VanDine <email address hidden> Thu, 14 Oct 2010 10:40:13 -0400

Least we think this is completely resolved, this just occurred on my up-to-date 10.04/Lucid system.

Marc, the bug is fixed for me so please submit a new report for yourself.