TEL-886: Media Port rework by alexlivekit · Pull Request #802 · livekit/sip

alexlivekit · 2026-08-20T19:20:39Z

This change is intended to:

Have little functional differences between it an main.
- the one known thing: Plug-in processor moved back to ingesting 48khz audio and returning 48khz audio.
Add interface for clarity.
Add machinery to support renegotiation, early media, hold, etc. All these are still disabled for now, but there.
Add tests.

alexlivekit · 2026-08-20T22:47:34Z

Huh, the race in CI is real, and is apparently a problem in media-sdk (yay zerocopy) that previous tests simply did not expose.
So... no regression compared to main.

alexfish8 · 2026-08-20T22:54:56Z

Huh, the race in CI is real, and is apparently a problem in media-sdk (yay zerocopy) that previous tests simply did not expose. So... no regression compared to main.

Will fix this in media-sdk.

dennwc · 2026-08-21T11:38:13Z

+	negotiated *sdp.MediaConfig
+
+	audioIn  *msdk.WriteCloserSwitch[msdk.PCM16Sample] // SIP RTP -> LK PCM
+	audioOut *msdk.WriteCloserSwitch[msdk.PCM16Sample] // LK PCM -> SIP RTP


Why these two are pointers, but the other two are not?

ha, nice catch.
We want to initialize the audio ones with msdk.NewWriteCloserSwitch, so that's why they're pointers.
Why the other ones are not? No reason. Fixed.

Aren't they safe to use as zero values?

dennwc · 2026-08-21T11:41:17Z

+		return errors.New("unexpected local address change")
+	}
+
+	audioToPort := p.audioOut.Swap(nil) // either nil or no-op closer


Maybe assert that it's the case during swap?

Not sure I follow. Are you talking about something like:

audioToPort := p.audioOut.Swap(nil) if audioToPort != nil { if _, ok := audioToPort.(msdk.writeCloser[x]); !ok { log.Errorw(...) } }

?

The only thin that swaps these is the configure function itself, so I'm not sure what the benefit would be there.
The comment is there to explain why we don't close the popped value.

dennwc · 2026-08-21T11:45:23Z

+
+	// WriteOutboundDTMFTo tells the room where to send DTMF to.
+	// Returns the previously-set writer (if one exists).
+	WriteOutboundDTMFTo(w msdk.WriteCloser[*livekit.SipDTMF]) msdk.WriteCloser[*livekit.SipDTMF]


Any specific reason to use pointers in livekit.SipDTMF? It's a tiny payload, might as well just copy it around.

I guess the thread safety is a good benefit of copying. Fair enough.

WriteSample passes lock by value: github.com/livekit/protocol/livekit.SipDTMF contains google.golang.org/protobuf/runtime/protoimpl.MessageState contains sync.Mutex I guess this is one,

Oh, right, it's a protobuf. Can we swap it for a non-proto struct? I remember we had races on proto structs before.

alexfish8 · 2026-08-25T04:01:48Z

Huh, the race in CI is real, and is apparently a problem in media-sdk (yay zerocopy) that previous tests simply did not expose. So... no regression compared to main.

Will fix this in media-sdk.

Fixed in media-sdk#81.

… workflow. (#808)

dennwc · 2026-08-27T14:31:43Z

+
+	// WriteOutboundDTMFTo tells the room where to send DTMF to.
+	// Returns the previously-set writer (if one exists).
+	WriteOutboundDTMFTo(w msdk.WriteCloser[*livekit.SipDTMF]) msdk.WriteCloser[*livekit.SipDTMF]


Oh, right, it's a protobuf. Can we swap it for a non-proto struct? I remember we had races on proto structs before.

dennwc · 2026-08-27T14:34:43Z

+	negotiated *sdp.MediaConfig
+
+	audioIn  *msdk.WriteCloserSwitch[msdk.PCM16Sample] // SIP RTP -> LK PCM
+	audioOut *msdk.WriteCloserSwitch[msdk.PCM16Sample] // LK PCM -> SIP RTP


Aren't they safe to use as zero values?

alexfish8 · 2026-08-28T20:20:31Z

 	if s.cli != nil { // Process reinvite for existing outbound calls
 		oc := s.cli.getActiveCall(cc.ID())
 		newCSeq := cc.InviteCSeq()
 		if oc != nil && oc.cc != nil && oc.cc.InviteCSeq() < newCSeq {


This is pre-existing, behavior, but, in a follow-up PR, we should reply with some kind of error if the new sequence number is lower than the existing one (per spec, I think this should be internal server error). Adding a todo.

devin-ai-integration

Devin Review found 1 new potential issue.

3 flags not posted on this PR by your GitHub settings — view them in Devin Review. (Configure)

devin-ai-integration · 2026-08-28T21:23:50Z

GetOutboundAudioWriter

+		if old := c.lkRoom.WriteOutboundAudioTo(nil); old != nil {
+			old.Close()
+		}


🟡 Dial-tone transfer closes the reused outbound audio writer

When a transfer plays dial-tone, WriteOutboundAudioTo(nil) returns the media port's own outbound writer and old.Close() closes it. That same writer (GetOutboundAudioWriter) is then used to play ringback and is re-attached to the room if the transfer fails, so ringback can be silent and a failed transfer can leave the caller with no room audio.

Suggested change

if old := c.lkRoom.WriteOutboundAudioTo(nil); old != nil {

old.Close()

}

c.lkRoom.WriteOutboundAudioTo(nil)

Was this helpful? React with 👍 or 👎 to provide feedback.

alexlivekit requested a review from a team as a code owner August 20, 2026 19:20

alexlivekit force-pushed the tel-886/media-port branch from 11f4b21 to dd06f29 Compare August 20, 2026 19:28

This comment was marked as resolved.

Sign in to view

alexfish8 mentioned this pull request Aug 21, 2026

Fix double close of inbound audio writer. #804

Merged

alexlivekit force-pushed the tel-886/media-port branch from 6a063e6 to dce2728 Compare August 21, 2026 07:52

This comment was marked as resolved.

Sign in to view

alexlivekit mentioned this pull request Aug 21, 2026

TEL-886: Self review #805

Merged

dennwc reviewed Aug 21, 2026

View reviewed changes

Base automatically changed from tel-886/prep to main August 21, 2026 16:42

alexlivekit force-pushed the tel-886/media-port branch 3 times, most recently from 47b61bd to 5ad9b22 Compare August 21, 2026 20:59

alexfish8 mentioned this pull request Aug 21, 2026

Remove struct field inboundCall.audioOut and mirror existing transfer workflow. #808

Merged

This comment was marked as resolved.

Sign in to view

alexlivekit and others added 5 commits August 26, 2026 06:12

Squashing work thus far

96a51dc

Fix double close of inbound audio writer (#804)

6970903

Remove struct field inboundCall.audioOut and mirror existing transfer…

b98805c

… workflow. (#808)

Merge main into tel-886/media-port.

a5f41a9

Check for nil audio writer when transferring calls.

470911b

alexlivekit force-pushed the tel-886/media-port branch from 4c60a45 to 470911b Compare August 26, 2026 13:19

alexlivekit added 4 commits August 26, 2026 10:07

TEL-886: Drop intermediate switch, use room directly (#814)

33cf505

Bumping media-sdk to main

cd59d1d

paralelizing tests

7e08d03

TEL-886: Fix SSRC race to codec (#815)

b6d87ed

dennwc reviewed Aug 27, 2026

View reviewed changes

alexfish8 added 2 commits August 28, 2026 13:21

Remove nil checks oc.media and oc.cc.

189556e

Prefer ctx to rctx.

4829a25

alexfish8 added 2 commits August 28, 2026 14:01

Rename oldRoomAudioOut to roomAudioOutput.

c6aa003

Remove outboundCall.audioOut (connect media and room directly).

b6f7b9a

alexfish8 reviewed Aug 28, 2026

View reviewed changes

devin-ai-integration Bot reviewed Aug 28, 2026

View reviewed changes

Uh oh!

Conversation

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

Uh oh!

Uh oh!

This comment was marked as resolved.

Uh oh!

alexlivekit commented Aug 20, 2026

Uh oh!

alexfish8 commented Aug 20, 2026

Uh oh!

This comment was marked as resolved.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

alexfish8 commented Aug 25, 2026

Uh oh!

This comment was marked as resolved.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

devin-ai-integration Bot left a comment

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Aug 28, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

alexlivekit commented Aug 20, 2026 •

edited

Loading