Single Peer Connection by pblazej · Pull Request #919 · livekit/client-sdk-swift

pblazej · 2026-02-17T09:06:31Z

Type-safe transport management via TransportMode (mirrors JS naming) 🎉
- This is not purely additive, so requires some regression testing 💣
Benchmarks will come in a separate PR (once we have the methodology)
There's no separate test for this, just switching the default behavior
- This is debatable, but the question is what to duplicate here... (which cases)
- I think JS does the exact opposite
- Rust creates separate (basic) cases

github-actions · 2026-02-17T09:06:43Z

⚠️ This PR does not contain any files in the .changes directory.

…erty

Sources/LiveKit/Core/Room+Engine.swift

refactor: introduce TransportMode enum to consolidate transport state

Sources/LiveKit/Core/SignalClient.swift

xianshijing-lk · 2026-02-27T07:54:59Z

can you port these single / dual PC rust tests over to swift ?
https://github.com/livekit/rust-sdks/blob/c0b0868cef64a5d705d7f91def1d99f715730aba/livekit/tests/peer_connection_signaling_test.rs

They should be able to catch most of the issues I ran into when working on the single peer connection issues on Rust

Port V0 (dual PC) and V1 (single PC) signaling integration tests from rust-sdks peer_connection_signaling_test.rs. Adds 17 tests covering connect, two-participant discovery, audio track pub/sub, reconnect, data channels, node failure recovery, many-tracks publish, and double reconnect for both signaling modes. - Add PeerConnectionSignalingTests with SignalingMode enum and ReconnectWatcher helper - Parameterize singlePeerConnection in RoomTestingOptions - Extract shared TestAudioTrack to LiveKitTestSupport/Tracks.swift - Consolidate duplicate TestTrack from TrackTests.swift Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Verify 50 audio frames received (not just first frame) in audio track tests, matching Rust's NativeAudioStream frame count validation - Add subscriber-side track verification after reconnect, confirming remote participant still sees publisher's tracks post-reconnect - Add frame counting and expect(minimumFrames:) to AudioTrackWatcher Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

pblazej · 2026-02-27T12:56:51Z

rust tests

Done, there's some duplication with existing tests, plus I left true as the default pc mode for other tests (no hard feelings here).

The reason I did not add them is most of these scenarios are covered somewhere e.g. in data channel tests (reconnect, etc.).

pblazej · 2026-03-02T11:59:27Z

Getting some timeouts in tests, looks like general problem with the runner.

xianshijing-lk · 2026-03-02T13:20:55Z

Curiously, do you need the same DSP munging in Swift SDK like in Rust SDK ?

Munge SDP to change a=inactive to a=recvonly for RTP media m-lines in single PC mode.
This is needed because WebRTC can generate inactive direction even when transceivers
were configured as recvonly.

pblazej · 2026-03-02T14:16:29Z

DSP munging

Yes, seems like it's necessary for the same reasons ⬇️

WebRTC's sdp_offer_answer.cc can generate a=inactive instead of a=recvonly for recvOnly transceivers during renegotiation in single peer connection mode, breaking video track subscription. This matches the fix applied in the Rust SDK (commit 2827707f). - Add singlePCMode flag to Transport - Add mungeInactiveToRecvOnlyForMedia() to rewrite a=inactive in RTP m-sections while preserving data channel sections - Apply munging in _negotiateSequence() before setLocalDescription - Add 3 unit tests for SDP munging Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

pblazej · 2026-03-03T12:02:17Z

There's some flakiness in these stress tests - not sure if caused by invalid server state or something on the client yet.

- Retry room connect (3 attempts, 2s delay) to handle transient POSIX 57 - Graceful track cleanup: unpublishAll + disconnect in withRooms teardown - Reduce post-disconnect sleep from 5s to 1s - Set server departure_timeout to 1s for faster room cleanup - Use didPublishTrack delegate to fix fullReconnect race condition - Replace nodeFailure with fullReconnect in signaling tests - Update macos-26 to Xcode 26.3 and 26.2 simulators Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

pblazej · 2026-03-06T10:45:25Z

Tests/LiveKitTestSupport/Room.swift

I made some global changes here to reduce test flakiness, key findings:

publishManyTracks can poison other tests as the server is unable to listen for socket connections

it does not look like a bug in the server code (listener is a separate goroutine), but rather a restricted VM environment (3 cores here)

it works reasonably well for fewer tracks

we can consider moving to -xlarge runners for other "stress" tests, however it's paid

pblazej · 2026-03-12T08:23:26Z

@lukasIO @xianshijing-lk do you see any gaps in the logic? You've got the full context.

@hiroshihorie is the swift side acceptable?

xianshijing-lk

lgtm, two nits.

xianshijing-lk · 2026-02-27T07:21:45Z

Sources/LiveKit/Core/Room+Engine.swift

-                                           delegate: self)
+            let isSinglePC = singlePeerConnection
+            let isSubscriberPrimary = isSinglePC ? false : joinResponse.subscriberPrimary
+            log("subscriberPrimary: \(isSubscriberPrimary), singlePeerConnection: \(isSinglePC)")


is the log intentional ?

it's .debug and it existed before: log("subscriberPrimary: \(joinResponse.subscriberPrimary)")

xianshijing-lk · 2026-03-18T14:55:11Z

Sources/LiveKit/Core/Room+Engine.swift

            let publisher = try Transport(config: rtcConfiguration,
                                          target: .publisher,
-                                          primary: !isSubscriberPrimary,
+                                          primary: isSinglePC || !isSubscriberPrimary,


curiously, what does primary mean here ? why primary is true when isSinglePC is true ?

it maps the inverse relationship between the TransportMode and Transport, mostly in Room+TransportDelegate (transport wouldn't know if it's "the only one" in single PC mode):

func transport(_ transport: Transport, didUpdateState pcState: LKRTCPeerConnectionState) { log("target: \(transport.target), connectionState: \(pcState.description)") // primary connected if transport.isPrimary { if pcState.isConnected { primaryTransportConnectedCompleter.resume(returning: ())

Sources/LiveKit/Support/Utils.swift

xianshijing-lk · 2026-03-19T13:59:58Z

@lukasIO @xianshijing-lk do you see any gaps in the logic? You've got the full context.

@hiroshihorie is the swift side acceptable?

@pblazej , can you check if things are working with auto_subscribe is set to false ?
We just found out some bugs there in Rust.

# Conflicts: # Sources/LiveKit/Protos/livekit_metrics.pb.swift

pblazej · 2026-03-19T14:17:19Z

We just found out some bugs there in Rust.

@xianshijing-lk I think there's some regression here vs main: after unsubscribing a video track I'm left with 8x8 "empty" video - is it confirmed to be on the SFU side?

In single PC mode the receive transceiver is reused (direction changes) rather than removed, so peerConnection(didRemove:) never fires and the VideoView is left with a stale track producing 8x8 empty frames. Clear the track immediately like the Rust SDK does. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

pblazej · 2026-03-20T08:28:08Z

The above issue with (un)subscribe seems to be tangentially related to transceiver management again, as we were listening to a wrong signal: b64d296

Now it mimics Rust (see TODO) https://github.com/livekit/rust-sdks/blob/56669aecf4bbd1875889b7af4612cf37e86ed364/livekit/src/room/publication/remote.rs#L251

cc @xianshijing-lk anything else reproducible in Rust?

pblazej added 4 commits February 17, 2026 09:41

proto: update to 1.44

de1dae8

feat: add single peer connection support to Room and RoomOptions

4e909c6

feat: implement support for single peer connection mode

161f5d8

test: enable single peer connection in room options for test cases

9149d2e

pblazej added 4 commits February 17, 2026 10:11

refactor: extract JoinRequest construction into a separate method

10862f9

fix: select transport based on single peer connection state

8cffacc

refactor: simplify transport selection with new receiveTransport prop…

53d7c3f

…erty

proto: regenerate with newer version

bbb7d8b

pblazej commented Feb 18, 2026

View reviewed changes

Sources/LiveKit/Core/Room+Engine.swift Outdated Show resolved Hide resolved

pblazej added 5 commits February 18, 2026 14:44

refactor: unify transport state management using a structured enum

1b9d016

refactor: introduce TransportMode enum to consolidate transport state

change

48b0a0f

test: update memory leak checks to use transport property in RoomTests

8cd0335

refactor: move transport management logic into TransportMode enum

8735aae

Merge branch 'main' into blaze/single-pc

4810ec3

pblazej requested review from hiroshihorie and lukasIO February 18, 2026 14:44

pblazej marked this pull request as ready for review February 18, 2026 14:44

refactor: simplify transport management and TransportMode logic

a54ffb3

lukasIO reviewed Feb 19, 2026

View reviewed changes

Sources/LiveKit/Core/SignalClient.swift Show resolved Hide resolved

pblazej added 3 commits February 20, 2026 09:06

feat: implement v1 RTC path with automatic fallback to legacy path

2b98cf8

fix: use established signal path version during quick reconnect

1069328

refactor: remove validate and forceSecure options from URL construction

a55f7f5

pblazej requested a review from xianshijing-lk February 25, 2026 14:32

pblazej and others added 3 commits February 27, 2026 13:45

gitignore

a5aa3f5

test: make single pc false in legacy tests

dc828be

pblazej and others added 2 commits March 3, 2026 10:06

Merge branch 'main' into blaze/single-pc

7b2130f

pblazej force-pushed the blaze/single-pc branch from d4810d7 to 5dc133b Compare March 3, 2026 10:32

pblazej force-pushed the blaze/single-pc branch 9 times, most recently from 39340bc to f608281 Compare March 6, 2026 09:55

pblazej force-pushed the blaze/single-pc branch from f608281 to fb20e80 Compare March 6, 2026 10:19

pblazej commented Mar 6, 2026

View reviewed changes

pblazej mentioned this pull request Mar 10, 2026

chore: regenerate Swift protobuf files #939

Merged

xianshijing-lk approved these changes Mar 18, 2026

View reviewed changes

xianshijing-lk reviewed Mar 18, 2026

View reviewed changes

Sources/LiveKit/Support/Utils.swift Outdated Show resolved Hide resolved

pblazej added 2 commits March 19, 2026 15:06

refactor: url validation cmt

cecef05

Merge branch 'main' into blaze/single-pc

cc9f0b7

# Conflicts: # Sources/LiveKit/Protos/livekit_metrics.pb.swift

pblazej requested a review from xianshijing-lk March 20, 2026 08:30

Conversation

pblazej commented Feb 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Feb 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

xianshijing-lk commented Feb 27, 2026

Uh oh!

pblazej commented Feb 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pblazej commented Mar 2, 2026

Uh oh!

xianshijing-lk commented Mar 2, 2026

Uh oh!

pblazej commented Mar 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pblazej commented Mar 3, 2026

Uh oh!

pblazej Mar 6, 2026

Choose a reason for hiding this comment

Uh oh!

pblazej commented Mar 12, 2026

Uh oh!

xianshijing-lk left a comment

Choose a reason for hiding this comment

Uh oh!

xianshijing-lk Feb 27, 2026

Choose a reason for hiding this comment

Uh oh!

pblazej Mar 19, 2026

Choose a reason for hiding this comment

Uh oh!

xianshijing-lk Mar 18, 2026

Choose a reason for hiding this comment

Uh oh!

pblazej Mar 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

xianshijing-lk commented Mar 19, 2026

Uh oh!

pblazej commented Mar 19, 2026

Uh oh!

pblazej commented Mar 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

pblazej commented Feb 17, 2026 •

edited

Loading

github-actions bot commented Feb 17, 2026 •

edited

Loading

pblazej commented Feb 27, 2026 •

edited

Loading

pblazej commented Mar 2, 2026 •

edited

Loading

pblazej Mar 19, 2026 •

edited

Loading

pblazej commented Mar 20, 2026 •

edited

Loading