v1.4.0
github-actions
released this
01 Apr 05:59
·
2 commits
to lukas/populate-reconnect-resp
since this release
In this release, we are introducing a major update to our connection quality algorithm. The improved algorithm boasts greater accuracy and quicker responsiveness to network conditions, providing users with early notifications about poor network connections.
We have also introduced a new internal transport between signal and RTC nodes for signal messages. Instead of relying on unreliable one-way pubsub messages, you can now opt for an RPC transport using psrpc. Note: Before setting signal_relay.enabled
, please ensure every instance in your cluster is running v1.4.0+; otherwise, instances will be unable to communicate with each other.
Added
- Added config to disable active RED encoding. Use NACK instead #1476 #1477
- Added option to skip TCP fallback if TCP RTT is high #1484
- psrpc based signaling between signal and RTC #1485
- Connection quality algorithm revamp #1490 #1491 #1493 #1496 #1497 #1500 #1505 #1507 #1509 #1516 #1520 #1521 #1527 #1528 #1536
- Support for topics in data channel messages #1489
- Added active filter to ListEgress #1517
- Handling for React Native and Rust SDK ClientInfo #1544
Fixed
- Fixed unsubscribed speakers stuck as speaking to clients #1475
- Do not include packet in RED if timestamp is too far back #1478
- Prevent PLI layer lock getting stuck #1481
- Fix a case of changing video quality not succeeding #1483
- Resync on pub muted for audio to avoid jump in sequence numbers on unmute #1487
- Fixed a case of data race #1492
- Inform reconnecting participant about recently disconnected users #1495
- Send room update that may be missed by reconnected participant #1499
- Fixed regression for AV1 forwarding #1538
- Ensure sequence number continuity #1539
- Give proper grace period when recorder is still in the room #1547
- Fix sequence number offset on packet drop #1556
- Fix signal client message buffer size #1561
Changed
- Reduce lock scope getting RTCP sender reports #1473
- Avoid duplicate queueReconcile in subscription manager #1474
- Do not log TURN errors with prefix "error when handling datagram" #1494
- Improvements to TCP fallback mode #1498
- Unify forwarder between dependency descriptor and no DD case. #1543
- Increase sequence number cache to handle high rate tracks #1560