[libnice] TRANSITION(CONNECTING, GATHERING) in agent.c/agent_signal_component_state_change

[Olivier Crête]

Hi,

And here is the MR

https://gitlab.freedesktop.org/libnice/libnice/-/merge_requests/204

I'm fixing the CI now to be able to merge it.

Olivier

On Tue, 2021-07-20 at 13:11 -0400, Olivier Crête wrote:
> Hi,
> 
> You were actually right, this transition was made possibly by a
> recent commit, I filed a MR to allow it.
> 
> Olivier
> 
> On Tue, 2021-07-20 at 17:57 +0100, Graham Hazel wrote:
> > Hi,
> > 
> > AFAICS the HEAD of master (commit a0cfef727...) which helpfully
> > predates all my hackery, so I'm fairly sure both failing and
> > "succeeding" builds are using the same version.
> > 
> > 
> > 
> > On Tue, Jul 20, 2021 at 5:25 PM Olivier Crête
> > <olivier.crete at collabora.com> wrote:
> > > Hi,
> > > 
> > > That actually really helps! Which exact version of libnice are
> > > you using ?
> > > 
> > > On Tue, 2021-07-20 at 17:20 +0100, Graham Hazel wrote:
> > > > Hi Olivier,
> > > > 
> > > > Thanks for your reply. It happened ~8-10 times over ~20k CPU-
> > > > hours of execution (and has happened 0 times in well over 10k
> > > > hours since I vandalised the assertion statement).
> > > > I'm afraid I don't have an easy way to reproduce it, nor a
> > > > useable stack trace, but by code inspection and correlation
> > > > with Janus logs I infer that the call stack is:
> > > > 
> > > > Janus function janus_ice_restart calls nice_agent_restart ->
> > > > nice_agent_restart calls nice_stream_restart ->
> > > > nice_stream_restart calls agent_signal_component_state_change
> > > > 
> > > > I hope that helps!
> > > > 
> > > > Graham
> > > > 
> > > > 
> > > > 
> > > > 
> > > > On Tue, Jul 20, 2021 at 4:48 PM Olivier Crête
> > > > <olivier.crete at collabora.com> wrote:
> > > > > Hi,
> > > > > 
> > > > > Can you get a stack trace when the assertion is triggered?
> > > > > It's
> > > > > definitely not a valid transition. How often can you
> > > > > reproduce this? Do
> > > > > you have an easy way to trigger it ?
> > > > > 
> > > > > If you're running a production server, you can always compile
> > > > > with -
> > > > > DG_DISABLE_ASSERT if you want to not abort. But it would
> > > > > really help me
> > > > > if you have a way to know where this invalid transition is
> > > > > called from.
> > > > > 
> > > > > Olivier
> > > > > 
> > > > > On Tue, 2021-07-20 at 14:00 +0100, Graham Hazel wrote:
> > > > > > Hi,
> > > > > > 
> > > > > > I'm running the Janus WebRTC Server to handle some video
> > > > > transport and
> > > > > > on several occasions have observed an assertion firing in
> > > > > libnice
> > > > > > (agent/agent.c, line 2573) during an ICE restart.
> > > > > > 
> > > > > > It appears likely that adding TRANSITION(CONNECTING,
> > > > > GATHERING) to the
> > > > > > list of valid state changes in the assert statement "fixes"
> > > > > the issue.
> > > > > > 
> > > > > > Now, I fully admit both (a) I'm probably doing "something
> > > > > weird" in my
> > > > > > setup to hit this case surprisingly often; and (b) I have
> > > > > very little
> > > > > > idea what any of the libnice code is doing. So my very
> > > > > naive question
> > > > > > is: is this actually a valid transition that's missing from
> > > > > the
> > > > > > assertion, or is something bad happening that I'm now
> > > > > ignoring?
> > > > > > 
> > > > > > Thanks!
> > > > > > Graham
> > > > > > 
> > > > > > _______________________________________________
> > > > > > nice mailing list
> > > > > > nice at lists.freedesktop.org
> > > > > > https://lists.freedesktop.org/mailman/listinfo/nice
> > > > > 
> > > 
> > > 
> 
> _______________________________________________
> nice mailing list
> nice at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/nice

-- 
Olivier Crête
olivier.crete at collabora.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/nice/attachments/20210720/0a61cec6/attachment.htm>