field name didn't match with frontend request#5
Closed
yeoleobun wants to merge 1 commit into
Closed
Conversation
Contributor
|
Should modify the field name in agent.js to -> |
yeoleobun
pushed a commit
to yeoleobun/rustpbx
that referenced
this pull request
Mar 16, 2026
fix(tts): timestamp gap between tts track, and clock rate for g722
anujarkitekt
added a commit
to parvbhullar/media-gateway
that referenced
this pull request
May 24, 2026
…apacity, CDR, leak fix, metrics Six related changes that move the SIP↔WebRTC bridge from "smoke-tested end-to-end" to "ready for real customers". 1. Codec passthrough on the SIP leg (restsend#1) - The SIP-side transcoder cap was hardcoded to `AudioCapability::pcmu()` regardless of what the carrier offered. Any carrier negotiating G.722, PCMA, or Opus-on-SIP would have its audio silently corrupted because the bridge would feed packets to a PCMU decoder. - New `sip_side_audio_offer()` advertises opus, g722, pcmu, pcma on the inbound RTP PC. `build_inbound_rtp_pc` now also returns the negotiated `AudioCapability` (intersect carrier-offered set with our supported set in carrier-preference order — first match wins). Rejects with a clear error if the carrier offered nothing we support (was previously a silent fall-through to PCMU). - G.729 deliberately omitted (patent-encumbered; no real encoder/decoder in rustrtc's default build). - `dispatch_webrtc` plumbs the negotiated codec into `BridgePeer::setup_bridge_with_codecs` as the RTP-side cap. 4 tests: PCMA-only, opus-on-SIP, multi-codec carrier preference, unsupported-only rejection. 2. DTMF pass-through (RFC 2833 / RFC 4733) — SIP → WebRTC (restsend#2) - `sip_side_audio_offer()` now also includes `AudioCapability::telephone_event()`. WebRTC-side capability list in `build_outbound_webrtc_pc` adds telephone-event alongside the voice codec. - The voice-codec intersection logic filters out telephone-event (it's signaling, not voice — would crash the transcoder if picked). Telephone-event PT is captured separately from the carrier's offer and used to install `BridgePeer::set_dtmf_sink(BridgeEndpoint::Rtp, dtmf_pt, ...)`. The bridge data plane already had the skip-transcoding-for-DTMF check (`media/bridge.rs:1328-1337`); it was no-op only because no sink was ever installed. With the sink in place, RFC 2833 telephone-event RTP packets flow verbatim from the SIP carrier to the WebRTC bot. - Reverse direction (WebRTC bot → SIP carrier DTMF) is intentionally deferred — same sink-install pattern but keyed on the bot's negotiated PT from the answer SDP. Defer until needed; most use cases are SIP → bot. - 3 tests: voice-vs-DTMF separation, non-standard DTMF PT respect, no-DTMF graceful fallback (carrier didn't offer it → sink not installed, debug log notes DTMF is in-band-only). 3. Capacity caps applied to webrtc dispatch (restsend#6) - `TrunkCapacityState` + `TrunkCapacityGate` were Phase 5/T artifacts that had no live caller in the SIP path; the gate was never wired to anything. WebRTC dispatch was equally unmetered. - New `lookup_webrtc_capacity_limits` helper in webrtc_route_dispatch reads `max_concurrent` + `max_cps` from the trunk row. - `dispatch_webrtc_bridge` (call.rs) now consults the gate at the start of dispatch: * Both limits None → unlimited, skip gate. * `CallsExhausted` → 503 + log + CDR. * `CpsExhausted` → 503 + log + CDR. * `Ok(permit)` → stash on `WebRtcBridgeSession._permit`; RAII Drop releases the slot at BYE-time teardown. 4. CDR rows for bridge calls (restsend#5) - Bridge calls used to be invisible to the CDR pipeline — billing and audit blind. New `proxy::webrtc_bridge_sessions::emit_bridge_call_record` + the `BridgeCallRecordInfo` shape grouping the constructor args. - `WebRtcBridgeSession` gains 8 CDR fields (call_id, caller_uri, callee_uri, from_number, to_number, trunk_name, trunk_id, start_time) populated at dispatch time. - Failure paths in dispatch_webrtc_bridge emit failed-call CDRs (capacity exhausted, dispatch error, reply_with failure, missing SDP, bad UTF-8, no DB connection). Success path emits the final CDR at BYE-time teardown with status_code=200 and hangup_reason=ByCaller (or Failed on adapter.close error). - Reuses existing `CallRecord` schema; bridge calls are tagged via `metadata.call_type="webrtc_bridge"` so reports can filter them. CDR write failures log warn and never block the call. - 2 unit tests for the emit helper (completed + failed dispositions). 5. Session-leak fix on dispatch failure (restsend#9) - Race condition in dispatch_webrtc: after the remote bot accepts the offer (signaling adapter returns NegotiateOutcome with a SessionHandle), any subsequent failure — SDP parse, set_remote_ description, audio_capability_for, setup_bridge_with_codecs — would return Err without invoking `adapter.close(ctx, &session)`. The bot's session leaked until its own idle timeout (often minutes). Same problem on the SIP side: `tx.reply_with` failure after the bot accepted left the session orphaned. - Fix: post-negotiate steps in dispatch_webrtc are wrapped in a closure that calls `adapter.close()` first on any error, then propagates. `DispatchOutcome` now carries the `SignalingContext` so the caller (proxy::call) can drive `adapter.close` on its own reply_with failure without re-deriving endpoint_url/auth_header from the DB. - Unit test injects a probe adapter that returns garbage answer SDP; asserts close fires exactly once with the right session handle before the function returns Err. 6. Bridge Prometheus metrics (restsend#8) - 4 new metrics under the `rustpbx_bridge_*` namespace: * `rustpbx_bridge_sessions_active` (gauge) — inc on WebRtcBridgeSessions::insert, dec on remove. * `rustpbx_bridge_dispatch_total{outcome}` (counter) — labels: success / signaling_error / rtp_setup_error / reply_error. Outcomes mutually exclusive: success emits from call.rs only after reply_with succeeds. * `rustpbx_bridge_signaling_latency_seconds{adapter}` (histogram) — time spent in `adapter.negotiate`, labeled by adapter name. * `rustpbx_bridge_bye_total{outcome}` (counter) — ok / teardown_error, emitted from teardown_webrtc_bridge_if_present. - Uses the same `metrics::` macro API as the rest of the codebase (no new deps). Observability addon exposes them at `/metrics` Prometheus endpoint once enabled. cargo check --release: clean. cargo test --release --lib proxy::bridge::webrtc: 14/14 pass (11 with this commit, + 3 from the codec passthrough commit that landed in the previous PR). cargo test --release --lib proxy::webrtc_bridge_sessions: 3/3. Open follow-ups (none blocking): - Bot → SIP DTMF (reverse of restsend#2): symmetric `set_dtmf_sink` on BridgeEndpoint::WebRtc keyed on the bot-negotiated telephone-event PT. - Re-INVITE handling (hold/resume/mid-call codec change): SIP dialog state machine extension; needs softphone-driven verification. - Real-audio end-to-end validation with actual Pipecat (not just the mock-offer test server).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The front end use field name
model, but back end usemodel_type.rustpbx/static/js/agent.js
Line 804 in f2623f6
They are not matching, so back end will use default model
16_zh_enand ignore the model parameter in request.