dbus-broker caused the system OOM issue.

[Sui Chen]

On Mon, Feb 1, 2021 at 5:40 PM Fran Hsu (徐誌謙) <Fran.Hsu at quantatw.com> wrote:
>
> Hello,
>         Sometimes my BMC will run out of memory so that I try to find out which process caused the OOM issue.
> From the htop information, it looks like the dbus-broker uses huge memory space.
> What I'm doing is to send the ipmi sdr list command from Host OS thru the kcs interface every 2 seconds.
> In the begging, I could see the dbus-broker uses up to 400MB and then reduced to less than 100MB.
> But after some hours I could see the memory usage keep increasing and the dbus-broker caused system OOM eventually.
>
>   1  [||||||||||||||||||||||||||||||||          67.1%]   Tasks: 104, 20 thr; 2 running
>   2  [||||||||||||||||||||||||||||||            60.4%]   Load average: 3.12 3.34 3.54
>   Mem[|||||||||||||||||||||||||||||||||||||969M/1003M]   Uptime: 12:59:12
>   Swp[                                          0K/0K]
>
>   PID USER      PRI  NI  VIRT   RES   SHR S CPU% MEM%   TIME+  Command
>   281 messagebu  20   0  870M  868M   376 S  0.6 86.6 40:54.32 dbus-broker --log 4 --controller 9 --machine-id 932c
>   265 messagebu  20   0  259M   284     4 S  0.0  0.0  0:00.34 /usr/bin/dbus-broker-launch --scope system --audit
>
> Version:
>         dbus-broker :25-r0
>
> I'd appreciate any comments or feedback.
>
> Thanks,
> Fran Hsu
> E-Mail : Fran.Hsu at QuantaTW.com
> Tel: +886-3-327-2345 Ext: 16935
>
>

Hello Fran,

I would do a DBus message capture and use it to examine: 1) what
method calls, signals and errors happened on the system, and 2)
whether there are method calls that take too long to complete, or
never return due to some errors.

The capture may be done using a command like "busctl capture >
dbus_capture.pcap"; the command monitors and dumps the DBus messages
to the output file similarly to Wireshark's capture mode. After
transferring the output file from the BMC to the host, dbus-pcap
(https://github.com/openbmc/openbmc-tools/tree/master/dbus-pcap) or a
work-in-progress visual tool
(https://gerrit.openbmc-project.xyz/c/openbmc/openbmc-tools/+/34263)
may be used to examine it to see if there are method calls that take a
very long time to complete or not return at all.

Hope this helps and thanks,
Sui