blueman-applet excessive CPU usage after resume

[tommie]

blueman: 2.3.5-3build1 (applet c07e8a4)
BlueZ: 5.72-0ubuntu5
Distribution: Ubuntu 24.04
Desktop environment: Gnome

This is a continuation on the discussion in #2499.

Today blueman-applet went into using constant excessive CPU after suspend/resume. It's at 10% rather than 100%, but that might be due to debug logging slowing it down.

https://gist.github.com/tommie/0ec3fa00f013995ab87f6e136d56fac7

It seems to be flooding

blueman-applet 07.50.43 DEBUG    Base:60 do_g_properties_changed: /org/bluez/hci0/dev_04_B9_E3_48_CF_87 {'RSSI': -85}

with different RSSI for the same device. Normally this only happens 1-2 seconds per minute as devices are added/removed, but now it keeps running constantly, logging several times per second. It keeps happening even if I restart blueman-applet.

The assertion error

Traceback (most recent call last):
  File "/usr/lib/python3/dist-packages/blueman/bluez/Manager.py", line 76, in _on_object_removed
    assert isinstance(device_proxy, Gio.DBusProxy)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError

seems unrelated as it occurs previously in the log file without causing the behavior.

[infirit]

We get these RSSI property changes from BlueZ's dbus api for devices when discovery is in progress. And I've never seen so many in such a short time ever. Like you said, normally BlueZ would update every couple seconds.

Is there anything interesting in the kernel log journalctl --dmesg? Or maybe in the bluetoothd daemon log journalctl -u bluetooth.service.

The assertion error should not happen but afaics harmless.

[tommie]

Kernel logs are very quiet about Bluetooth. bluez has nothing that seems related. It complains about my Bose headphones as I turn them on and off (and sometimes picks the wrong audio profile.)

I killed the applet and restarted it a couple of times. Right now it's off, and now /usr/lib/x86_64-linux-gnu/indicator-bluetooth/indicator-bluetooth-service is at 82% CPU and bluetoothd is around 6%. If I kill indicator-bluetooth, bluetoothd still sits around 6%.

So perhaps this is rooted in bluez having a mental breakdown.

[tommie]

Just realized: I am running bluetoothd --experimental because I needed access to some experimental API at some point.

[tommie]

Interestingly, restarting bluetoothd seems to make my whole process tree just... quiet down... Even Chrome stops using a bunch of CPU, it seems.

If I now start blueman-applet, it is quiet as it should be.