Skip to content

CPU usage increases over time until the UI becomes unresponsive. [mac] #1643

Closed

Description

@pianojoe

On Mac, after two hours of rehearsal with ~40 participants, the UI becomes unresponsive. A click may take 20s to execute. top shows a processor usage of 103%. (As compared to 35% one hour earlier, with the same number of participants.)
The Mac mouse cursor changes into the "spinning beachball of death", as it is commonly reffered to.

CPU usage shouldn’t grow over time.

I’m on macos 11.3.1 on a 2.9 GHz 6-Core Intel Core i9 MacBookPro (definitely not underpowered). I’m using a USB audio interface Focusrite Scarlett 4i4.

The behaviour was observed on Jamulus 3.7.0 (obtained as a .dmg from jamulus.io) AND on Jamulus 3.8.0beta1 (which I compiled myself).

Activity

  1. hoffie commented on May 6, 2021

    @hoffie
    Member

    Thanks for the report.

    1. Just to confirm, you are talking about the Jamulus client, right?

    2. Just a blind guess because it seems to be the reason for another recent Mac-related issue: Which version of Qt did you use on your local build? The official .dmg is built against rather old Qt (5.9) which seems to cause problems (UI glitches) on recent Mac versions.
      Could you try this build and report back if this improves things for you?
      ActiveTab heading unvisible at MAC 3.8.0beta #1628 (comment)

    3. Does memory usage also increase?

  2. pianojoe commented on May 6, 2021

    @pianojoe
    Author

    hoffie,

    1. Yes, I’m talking about the Jamulus Client.
    2. Qt 5.15.2. This is the very same Qt version as your build.
    3. Need to check. I’ll get back to you on this.
  3. hoffie commented on May 6, 2021

    @hoffie
    Member

    Ok. My only idea for getting close to the cause of the issue would be to try different working/non-working versions. Does 3.6.2 exhibit the same problem? If not, that would be something to start from. Are you familiar with git bisect? That way you could generate different builds and check if they are broken or not. This will likely require some iterations of building / verifying / building again, etc., but will lead to a single commit which is the cause for the change in behavior.

    This requires effort, but I don't know where else to start. :(

  4. softins commented on May 6, 2021

    @softins
    Member

    Also, when it gets into this state, does clicking Disconnect and reconnecting give any improvement, or is it necessary to exit and restart the program?

  5. MattMerz-Violin commented on May 10, 2021

    @MattMerz-Violin

    I have the same problem. Jamulus 3.7 becomes unresponsive after playing for some time and not touching any UI item. It then takes up to 20 seconds until the mouse click is recognized and the UI is reacting again. MacOS 11.3.1.

    I use the official Jamulus for Mac version that can be downloaded from SourceForge.

  6. hoffie commented on May 10, 2021

    @hoffie
    Member

    @MattMerz-Violin Thanks for your information.

    If someone is able to capture a trace in the problem situation, this may also help (I think Mac's strace equivalent would be dtruss?).

    Just a quick idea -- #1075 was a Mac-only improvement which got into 3.7.0. I don't see how it would cause this, but this might be one thing to try.

  7. softins commented on May 10, 2021

    @softins
    Member

    For the first time, I experienced this tonight. Macbook Pro on Mojave, running the 3.8.0beta1 release.

    This was after playing with four others for about 90 minutes.

    Interestingly, the audio continued without significant delay, although the ping and delay stopped updating, and attempting to use the GUI gave the spinning beach ball. This prevented disconnecting.

    Running top in a terminal showed Jamulus using over 90% of the CPU.

    I quit and restarted Jamulus. I have a slight suspicion that using the menu bar might have unstuck it, but I had already clicked on Quit.

    Will investigate more...

  8. hoffie commented on May 10, 2021

    @hoffie
    Member

    Good to know that this is reproducible with just five participants which should be easier to simulate than 40.
    I've ran multiple 90+ minute sessions with more than 4 people with both 3.7.0 and ~3.8.0beta1 on Linux, so this is either really Mac-specific (as suspected) or needs another factor to become an issue.

  9. MattMerz-Violin commented on May 11, 2021

    @MattMerz-Violin

    Hi, a few answers (more to come):

    What is the minimal number of users to trigger this for you? MM: happens for me with 3 Players and also with up to 20 (did not try with more)
    What is the minimal number of minutes after this happens? MM: will try to find out
    Can you answer @softins's question regarding the Disconnect test above? MM: as soon as the program reacts again, it reacts totally normal again without restarting.
    Can you try with an empty .ini file (e.g. start via Jamulus --inifile /tmp/empty.ini or locate the default ini and rename/remove it)? MM: will do so and Report.
    If you are familiar with building and using git, it would be really helpful to run the steps from #1643 (comment) MM: I am not familiar with building programs myself :-(

  10. softins commented on May 11, 2021

    @softins
    Member

    Please could those affected by this issue try the macOS build from the new 3.8.0beta2 version?

    It has come to light that both 3.7.0 and 3.8.0beta1 were built using Apple SDK 11.1, which Qt5 claims is not fully compatible. So it's possible, although not certain, that could be related to this issue.

    See PR #1655 for more information.

  11. MattMerz-Violin commented on May 13, 2021

    @MattMerz-Violin

    Hi, I tried 3.8 beta 2 today in a 2 hour Jamulus session and it still shows the same behaviour under MacOS 11.3.1. After a certain time 30 to 60 minutes in the background, the app becomes unresponsive when I click again in the Jamulus window. Then the spinning wheel shows for 10 or 15 seconds an then Jamulus reacts again normal to every input. Between 5 and 15 players were in the session.

  12. JWuStudio commented on May 15, 2021

    @JWuStudio

    now that this is brought up, I have totally have had this experience before, since I started using this for large ensemble last September (thus I do not think it is version/build sepcific). During the school year we had 2 hours rehearsals with 40+ people twice a week, and i do not know about everyone in the ensemble, but I know that a couple of my friends who are on Mac would need to disconnect and reconnect during the break at 1 hour, otherwise the clicks becomes VERY irresponsible, ie clicking on "Solo" might take 10 seconds to actually react and solo the track, and same with un-soloing.

    I just assumed that this is just because of large ensemble and extended time, so that is why I never bothered reporting it. I will ask the ensemble and come back here.

  13. Sommi486 commented on May 18, 2021

    @Sommi486

    The behavior ist still in 3.8.0rc1

    After 8 minutes not touching the client an sending no sound the client ist frozen. Happened with Catalina and BigSur

  14. softins commented on May 18, 2021

    @softins
    Member

    The behavior ist still in 3.8.0rc1

    After 8 minutes not touching the client an sending no sound the client ist frozen. Happened with Catalina and BigSur

    Was this while connected to a server, or not?

  15. 35 remaining items

  16. unpinned this issue on May 26, 2021
  17. MattMerz-Violin commented on May 29, 2021

    @MattMerz-Violin

    Hi, just played one hour with RC1 under MacOS 11.4 and had the settings window closed. When trying to disconnect at the end of the session (3 players) the Beachball showed up and it took 10 to 15 seconds until the program reacted. The RAM usage had not increased.

    So for me it looks like the settings window has a big problem that can be reproduced easily, but there seems to be another issue with the program also causing the GUI not reacting.

  18. softins commented on May 29, 2021

    @softins
    Member

    @MattMerz-Violin thanks for the report! I suspect that this is a long-standing issue deep within the Qt implementation on Mac. We will not be able to solve it (or even understand it) before the 3.8.0 release.

  19. moved this from Triage to Done in Trackingon Jan 25, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

bugSomething isn't working

Type

No type

Projects

  • Status
    Done

Milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions