Consultation about QMI CID Leaks

[Aleksander Morgado]

> >> We have created the issue in Gitlab and uploaded all the logs that related.
> >> https://gitlab.freedesktop.org/mobile-broadband/libqmi/-/issues/81
> >>
> >> Please contact us if there is still anything missing.
> >> Waiting for your kind reply. Thanks!
> >>
> >
> > You missed to reply to my question before: can you confirm you're NOT
> > running qmicli commands in addition to MM managing the modem? or are
> > you running them?
> > If you're running custom qmicli commands you may be the one triggering
> > the client ids exhausted. MM by itself doesn't (shouldn't) do that.
>
> In later reply he says no, but we are.
>
> On a long running system I saw some kind of resource exhaustion; unfortunately
> I was not able to examine the system myself, only that the system no longer
> attempted to connect.
>
> I've got some possible tests that may trigger the issue, but can you give me
> some clues as to what I'm looking at, or what to avoid?
>

You should obviously avoid explicit CID leaks happening due to missing
"release client" operations. E.g. if you run your custom qmicli
commands and you're using --client-cid and --client-no-release-cid,
please make sure you correctly balance allocated and released CIDs.
You should also avoid killing ModemManager abruptly; e.g. a kill -9 on
the process will leak CIDs. I don't think there are many more things
to avoid.

If you're testing this, please make sure to run the qmi-proxy with
verbose logs; at the end the proxy is the one controlling which CIDs
are allocated from all processes, and the only one that could detect
the leaks.

-- 
Aleksander
https://aleksander.es