Question about implementing WebRTC Stats

[Olivier Crête]

Hi,
Answering with my libnice maintainer hat.
On Tue, 2020-10-27 at 19:31 +0000, Lin, Sherrill wrote:
> Hi Matthew,
> 
>  
> 
> Thanks a lot for your reply!
> 
>  
> 
> We’d like to elaborate on it a bit more. According to webrtc spec (
> https://w3c.github.io/webrtc-pc/#mandatory-to-implement-stats),
>  the mandatory fields for "RTCIceCandidatePairStats" include:
> transportId, localCandidateId, remoteCandidateId, state, nominated,
> bytesSent, bytesReceived, totalRoundTripTime, currentRoundTripTime
> 
> 
> 
> However, we find some of the fields not easy to implement:
> 
> 
> state, nominated, totalRoundTripTime, currentRoundTripTime
> These information are all related to STUN connectivity checks, which
> currently are not exposed in the Libnice public APIs. It would be
> very difficult to collect these data without modifying Libnice
>  APIs.

Libnice currently the state and nominated-ness of each pair in an
internal data structure. I think it would be relatively simple to just
add a function to retrieve a copy of those.
Libnice doesn't currently measures the RTT, but we could relatively
easily store the time when a STUN request is sent and do the
substraction when a reply is received to get this. And we could expose
it in the same API.
> 
> bytesSent, bytesReceived
> These represent the total number of payload bytes sent/received on
> the candidate pair. We think of some potential approaches:
> 
> 
> 
> Add buffer probes in dtlssrtpenc and dtlssrtpdemux to calculate the
> bytes sent/receivedCollect from srtpenc(
> https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/blob/master/ext/srtp/gstsrtpenc.c#L1205
> ),
>  dtlsenc(
> https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/blob/master/ext/dtls/gstdtlsenc.c#L519
> ), and dtlssrtpdemux(
> https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/blob/master/ext/dtls/gstdtlssrtpdemux.c#L105
> ).
>  And we could expose an property in dtlssrtpenc and dtlssrtpdemux for
> the collected data.Calculate from the bytes sent/received data in RTP
> stream stats (
> https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/blob/master/ext/webrtc/gstwebrtcstats.c#L91
> ).
>  We would also need to gather the data for data-channel because
> that's not covered in RTP stream stats.
> 
> Please let us know if there's a better way to do it.

I wonder if the best approach wouldn't be to calculate these in libnice
and expose them from there directly.
Doing this at the GStreamer level wouldn't give the accurate result per
candidate pair.
Olivier
>  
> We'd like to get more guidance on how to implement these fields. If
> there's no good approach to do it yet, can we leave them out for now?
>  
> Thanks,
> Sherrill
>  
> 
> From: Matthew Waters <ystreet00 at gmail.com>
> 
> Date: Friday, October 23, 2020 at 2:44 AM
> 
> To: Discussion of the development of and with GStreamer <
> gstreamer-devel at lists.freedesktop.org>, "Lin, Sherrill" <
> lshuying at amazon.com>
> 
> Cc: comms-media-process <comms-media-process at amazon.com>
> 
> Subject: RE: [EXTERNAL] Question about implementing WebRTC Stats
> 
> 
>  
> 
> Hi,
> 
> 
> 
> It would certainly be beneficial to Implement the mandatory stats
> required but it wouldn't quite be a hard rejection if they aren't
> implemented.  It would require a reasonable explanation as to why
> they aren't implemented for new structures though.
> 
> 
> 
> Cheers
> 
> -Matt
> 
> On 23/10/20 8:00 am, Lin, Sherrill wrote:
> 
> > Hi there,
> >  
> > Our team in Alexa Communications wants to add selected ice
> > candidates information to WebRTC stats. We find that in
> > gstwebrtcstats.c (
> > https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/blob/master/ext/webrtc/gstwebrtcstats.c
> > ),
> >  RTCIceCandidateStats, RTCIceCandidatePairStats and
> > RTCTransportStats are not constructed yet. We think these three
> > stats are the structures needed for hosting selected ice candidate
> > data, and we are interested in implementing them following the
> > w3.org spec
> >  (https://www.w3.org/TR/webrtc-stats/).
> >  
> > However, we notice there’s a list of mandatory fields for
> > implementing each stats structure:
> > https://w3c.github.io/webrtc-pc/#mandatory-to-implement-stats. Does
> > it mean, for Gstreamer community to accept the code changes, we
> > will have to implement all the mandatory fields for
> >  all three stats structures as defined in that doc?
> >  
> > Thanks,
> > Sherrill
> > 
> > 
> > 
> > 
> > 
> > _______________________________________________
> > gstreamer-devel mailing list
> > gstreamer-devel at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/gstreamer-devel
> 
> 
> 
> 
> 
> 
> 
> 
> 
> _______________________________________________gstreamer-devel
> mailing listgstreamer-devel at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/gstreamer-devel
-- 
Olivier Crête
olivier.crete at collabora.com


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/gstreamer-devel/attachments/20201027/8920f811/attachment.htm>