fix(codec-selection): Disabled VP9 for p2p on Firefox.
When VP9 is at the top of the codec list in the remote offer, Firefox omits it in the answer but decides to use VP9 for encoding media. Therefore the remote peer gets an answer with no VP9 in the list and when it updates its local description (with no VP9) after a renegotiation, it stops rendering the VP9 video received from Firefox.
fix(codec-selection) Pick only 1 VP9 codec for p2p negotiation.
More than one VP9 payload type is produced in the offer, more experimental ones if the local endpoint has no local video source. Munge the SDP to remove all codecs besides the one with profile-id=0 so that only one VP9 codec is negotiated with p2p peer.
fix(JingleSession) Modify encoding to stop/start outgoing media on p2p.
This is much faster than changing direction on the transceiver and doesn't need a renegotiation for the suspending/resuming media. Also fixes an issue with audio-only mode in p2p where media doesn't resume when audio-only mode is disabled if it was enabled while the jvb connection was active.
ref(JingleSession) Always call sRD->cA-sLD for renegotiation.
For renegotiations triggered by the application, it is ok to execute sRD->cA-sLD cycle on the p2p initiator instead of executing cO->sLD->sRD cycle. Chrome sometimes creates a third m-line for p2p when createOffer is called which is unexpected and therefore it breaks the client's renegotiation cycle. This has been done in other places already, i.e., when remote sources are signaled, etc. so making it uniform across all renegotiations.
Revert "Adjusted wrapping method to have RTCRtpSender setting parameters in sequential manner in order to solve race condition when transaction id can be reset at the end of videoSender setParameters and cause exception if another update was already in fly and before its check for transaction id presence."
This reverts commit 17ade96a70.
It causes setParameters failures which can cause media being leaved to
JVB when in a P2P connection.
* fix(codec-selection) Use a configurable preferred codecs list to select the codecs.
Allow asymmetric codecs to be configured on the endpoints. This means that Firefox and Safari which have bugs with VP9 encode will now encode VP8 but decode VP9 coming in from Chromium endpoints.
* fix(codec-selection) Add unit tests for the codec selection logic.
* feat(codec-selection): Introduce mobileCodecPreferenceOrder setting in config.js.
* fix(codec-selection) Avoid unnecessary renegotiations.
On participant join/leave, check if the new intersection of codecs are already configured to be the top n codecs.
* Address review comments
* fix: Strip the codecs that are not in the codec list from the video m-lines.
Also, ignore remote codecs published in presence for p2p connections.
* fix: Define default codec order for mobile and desktop.
Fix an issue where p2p between mobile and desktop was broken.
* fix: Add default codecs for both p2p and jvb on mobile devices.
Munge the initial offer sent out by RN clients since RN doesn't support RTCRtpTransceiver#setCodecPreferences.
fix(JingleSession) Reconfigure the stream encodings after p2p->jvb switch.
Fixes blurry screenshare in some cases. There is a higher probability of the higher layers getting suspended when all the stream encodings are enabled for low fps SS. Make sure only the highest spatial layer is sent for low fps SS after p2p->jvb switch. Make sure all the stream encodings are enabled for high fps SS after p2p->jvb switch.
Adjusted wrapping method to have RTCRtpSender setting parameters in sequential manner in order to solve race condition when transaction id can be reset at the end of videoSender setParameters and cause exception if another update was already in fly and before its check for transaction id presence.
fix(ssrc-rewriting) Update owners when a new SSRC is assigned to existing source.
Fixes a bug where the remote video freezes if the remote user does an ICE restart using session-terminate. In this case, a new SSRC is assigned to the remote source which will have the same source-name.
fix(stats) Obtain resolution/fps from 'outbound-rtp' stats. (#2265)
* fix(stats) Obtain resolution/fps from 'outbound-rtp' stats.
'Track' based stats were dropped in Chrome 112, therefore send resolution and fps for the simulcast case needs to be calculated based on the 'outbound-rtp' streams that are currently active.
* squash: remove an unwanted log
* Squash: Address review comments.
Use outbound-rtp stats for both Firefox and Safari.
Firefox - Ignore active field if not present in the stats and calc fps using 'framesSent'.
fix(TPC) Avoid applying constraints on desktop tracks if not needed. (#2246)
* fix(TPC) Avoid applying constraints on desktop tracks if not needed.
Ignore sender constraints if the client is already sending video of the requested resolution. For desktop tracks, max resolution will be the height of the window being captured irrespective of the height being requested by the peer. Therfore, check if the configured resolution is equal to the track height for all requested heights > 0.
Fixes an issue where the track addition fails because of setParameters failing on the video track. This is only seen the torture tests because this is very timing specific.
* squash: check correct scaleResolutionDownBy value for p2p conn.
fix(codec-selection) Apply codec preferences to initial offer/answer.
This fixes an issue where p2p clients (with different codec preferences) fail to decode video because the negotiated codecs are removed from the supported codecs list after the media session is established. The codec preferences will be applied when the first offer/answer is created.
fix(codec-selection) Check if disabled is already removed from sdp.
This should avoid unnecessary renegotiations if disabledCodec setting is present in config.js and should improve time to receive media when joining big calls.
Made updating video sender parameters in sequential manner in order to solve race condition when transaction id (which is generated during getParameters) can be reset at the end of videoSender setParameters and cause exception if another update was already in fly and before its check for transaction id presence.
fix(TPC) Disable media instead of changing dir for p2p->jvb switch. (#2226)
* fix(TPC) Disable media instead of changing dir for p2p->jvb switch.
Resume or suspend the media on the jvb peerconnection by changing the RTCRtpEncodingParamters.active state instead of changing the direction on the transceiver. This avoids the needs to start a O/A renegotiation cycle for these operations. The media direction will be changed only for p2p lastn=0 case since video needs to be disabled on both the sender and the peer for p2p lastn=0 case.
* Address review comments
* Disable media after adding source while media is suspended on the jvb connection. Default 'active' state for stream encodings after the source is added is 'true'.
* Wait for all the promises to be settled before returning
fix(BridgeChannel): Attempt conn retries when there are remote endpoints.
Currently, the client doesn't attempt to re-establish bridge WS when the connection is closed by the remote end with a code 1001. We have noticed that Cloudflare terminates the WS with the same error code when it recycles its proxies. When that happens, client ends up not having a bridge channel which can result quality issues.
Client now tries to re-establish connection when there are remote endpoints in the call.
fix(SignalingLayer) Update SSRC owners on leave. (#2184)
* fix(SignalingLayer) Update SSRC owners on leave.
Update the SSRC owners in the following cases:
1. When a remote endpoint leaves the call.
2. When a source-remove is received.
3. When a source is remapped (with ssrc-rewriting enabled).
Create the remote track even if presence is not yet received. The ssrc owner check prevents the client from creating a dummy track when the call switches over from p2p to jvb when the last remote endpoint leaves the call.
* ref(SignalingLayer) alpha sort methods.
Clean up unused methods, _findEndpointSourceInfoForMediaType is not used anymore.
* squash: Address review comments.
fix(qualitycontrol): Cleanup old receiver constraints.
Endpoint based receiver constraints and other endpoint based bridge signaling messages are no longer supported by latest JVB after the switch to source-name signaling.
Rename method names 'sendNewReceiverVideoConstraintsMessage'->'sendReceiverVideoConstraintsMessage', 'setNewReceiverVideoConstraints'->'setReceiverVideoConstraints'
1. Checks peer's preferred codec in p2p case. Mobile and web have different preferred codecs.
2. Log an error message when the preferred codec is not offered by JVB.
3. Clean up code related to deprecated config.js settings 'preferH264' and 'disableH264'.
4. Refactor the codec selection logic so that correct codec is picked.
Hristo Terezov
Jaya Allamsetty
Saúl Ibarra Corretgé
Jaya Allamsetty
mbondarenko
Jorge Oliveira
bgrozev
Daniel McAssey
Daniel McAssey
Saúl Ibarra Corretgé