fix(helper): gate audio observer on fixture ack

cai

2026-08-11 949be8e5f1a75286b8b517dacda76ed27646e77b

 src/main.rs

@@ -59,6 +59,123 @@
const STREAM_TIMING_FIRST_CHUNK: u64 = 1;
const STREAM_TIMING_MAX_ELAPSED_MS: u64 = 5_000;
const STREAM_TIMING_SOURCE: &str = "stream_anchor_monotonic";
const CONTROLLED_FIXTURE_PROBE_TOPIC: &str = "controlled_fixture_attribute_probe";
const CONTROLLED_FIXTURE_ACK_TOPIC: &str = "controlled_fixture_attribute_ack";
const CONTROLLED_FIXTURE_PROTOCOL_VERSION: u64 = 1;
const CONTROLLED_FIXTURE_GENERATION: u64 = 1;
const CONTROLLED_FIXTURE_PROBE_RECHECKS: usize = 3;
const CONTROLLED_FIXTURE_PROBE_RECHECK_DELAY: Duration = Duration::from_millis(25);
const CONTROLLED_FIXTURE_PROBE_TTL: Duration = Duration::from_millis(250);

#[derive(Debug, Deserialize)]
#[serde(rename_all = "camelCase")]
struct ControlledFixtureAttributeProbe {
    #[serde(rename = "type")]
    message_type: String,
    protocol_version: u64,
    call_id_hash: String,
    call_trace_id_hash: String,
    generation: u64,
    client_fixture_sequence: String,
}

#[derive(Debug, Serialize)]
#[serde(rename_all = "camelCase")]
struct ControlledFixtureAttributeAck {
    #[serde(rename = "type")]
    message_type: &'static str,
    protocol_version: u64,
    call_id_hash: String,
    call_trace_id_hash: String,
    generation: u64,
    client_fixture_sequence: String,
    result: &'static str,
    #[serde(skip_serializing_if = "Option::is_none")]
    input_source_category: Option<&'static str>,
    #[serde(skip_serializing_if = "Option::is_none")]
    reject_reason: Option<&'static str>,
}

#[derive(Debug)]
struct PendingControlledFixtureProbe {
    sender: ParticipantIdentity,
    sequence: String,
    expires_at: Instant,
}

fn sha256_hex(value: &str) -> String {
    let mut digest = Sha256::new();
    digest.update(value.as_bytes());
    digest
        .finalize()
        .iter()
        .map(|byte| format!("{byte:02x}"))
        .collect()
}

fn controlled_fixture_probe(
    payload: &[u8],
    call_id: &str,
    trace_id: &str,
    sender: &ParticipantIdentity,
    expected_participant: Option<&str>,
) -> Option<PendingControlledFixtureProbe> {
    if !is_bound_user_participant(sender.as_str(), expected_participant) {
        return None;
    }
    let probe: ControlledFixtureAttributeProbe = serde_json::from_slice(payload).ok()?;
    if probe.message_type != CONTROLLED_FIXTURE_PROBE_TOPIC
        || probe.protocol_version != CONTROLLED_FIXTURE_PROTOCOL_VERSION
        || probe.generation != CONTROLLED_FIXTURE_GENERATION
        || probe.call_id_hash != sha256_hex(call_id)
        || probe.call_trace_id_hash != sha256_hex(trace_id)
        || probe.client_fixture_sequence.trim().is_empty()
    {
        return None;
    }
    Some(PendingControlledFixtureProbe {
        sender: sender.clone(),
        sequence: probe.client_fixture_sequence,
        expires_at: Instant::now() + CONTROLLED_FIXTURE_PROBE_TTL,
    })
}

fn classify_controlled_fixture_attributes(
    actual_participant: &str,
    expected_participant: Option<&str>,
    attributes: &std::collections::HashMap<String, String>,
    requested_sequence: &str,
) -> Result<(), &'static str> {
    if !is_bound_user_participant(actual_participant, expected_participant) {
        return Err("wrong_participant");
    }
    let source = attributes
        .get("inputSourceCategory")
        .map(String::as_str)
        .unwrap_or("");
    if source != "controlled_fixture" {
        return Err(if source.is_empty() {
            "missing_attributes"
        } else {
            "wrong_source"
        });
    }
    let sequence = attributes
        .get("clientFixtureSequence")
        .map(String::as_str)
        .unwrap_or("");
    if sequence.trim().is_empty() {
        return Err("missing_sequence");
    }
    if sequence != requested_sequence {
        return Err("wrong_sequence");
    }
    Ok(())
}

fn controlled_fixture_observer_gate(pending_probe: bool, probe_result: Option<bool>) -> bool {
    !pending_probe || probe_result == Some(true)
}

#[tokio::main(flavor = "multi_thread")]
async fn main() -> Result<()> {
@@ -768,6 +885,10 @@
        "runtime helper user_track_subscribe_requested"
    );

    let mut current_user_participant: Option<RemoteParticipant> = None;
    let mut pending_probe: Option<PendingControlledFixtureProbe> = None;
    let mut acknowledged_probe_sequences = HashSet::new();

    while let Some(event) = events.recv().await {
        match event {
            RoomEvent::TrackSubscribed {
@@ -797,6 +918,26 @@
                    track_source = %track_source,
                    "runtime helper user_track_subscribed"
                );
                let pending_sequence = pending_probe.as_ref().map(|probe| probe.sequence.clone());
                let participant_for_probe = participant.clone();
                let probe_result = process_controlled_fixture_probe(
                    &mut pending_probe,
                    &mut acknowledged_probe_sequences,
                    Some(&participant_for_probe),
                    &sink,
                    &call_id,
                    &trace_id,
                    user_participant_identity.as_deref(),
                )
                .await;
                if !controlled_fixture_observer_gate(pending_sequence.is_some(), probe_result) {
                    warn!(
                        call_id = %call_id,
                        trace_id = %trace_id,
                        "runtime helper withheld audio observer until controlled fixture ACK"
                    );
                    continue;
                }
                spawn_user_audio_frame_observer(
                    track,
                    call_id.clone(),
@@ -812,6 +953,38 @@
                    user_participant_identity.clone(),
                    participant,
                );
                current_user_participant = Some(participant_for_probe);
            }
            RoomEvent::DataReceived {
                payload,
                topic: Some(topic),
                participant: Some(sender),
                ..
            } if topic == CONTROLLED_FIXTURE_PROBE_TOPIC => {
                if let Ok(probe) =
                    serde_json::from_slice::<ControlledFixtureAttributeProbe>(&payload)
                {
                    if acknowledged_probe_sequences.contains(&probe.client_fixture_sequence) {
                        continue;
                    }
                }
                pending_probe = controlled_fixture_probe(
                    &payload,
                    &call_id,
                    &trace_id,
                    &sender.identity(),
                    user_participant_identity.as_deref(),
                );
                process_controlled_fixture_probe(
                    &mut pending_probe,
                    &mut acknowledged_probe_sequences,
                    current_user_participant.as_ref(),
                    &sink,
                    &call_id,
                    &trace_id,
                    user_participant_identity.as_deref(),
                )
                .await;
            }
            RoomEvent::TrackSubscribed {
                track: RemoteTrack::Video(track),
@@ -852,6 +1025,79 @@
            _ => {}
        }
    }
}

async fn process_controlled_fixture_probe(
    pending_probe: &mut Option<PendingControlledFixtureProbe>,
    acknowledged_probe_sequences: &mut HashSet<String>,
    participant: Option<&RemoteParticipant>,
    sink: &BotAudioOutputSink,
    call_id: &str,
    trace_id: &str,
    expected_participant: Option<&str>,
) -> Option<bool> {
    let Some(probe) = pending_probe.take() else {
        return None;
    };
    if acknowledged_probe_sequences.contains(&probe.sequence) || Instant::now() > probe.expires_at {
        return Some(false);
    }
    let Some(participant) = participant else {
        *pending_probe = Some(probe);
        return None;
    };
    if participant.identity() != probe.sender {
        return Some(false);
    }
    let mut decision = Err("timeout");
    for attempt in 0..CONTROLLED_FIXTURE_PROBE_RECHECKS {
        if Instant::now() > probe.expires_at {
            break;
        }
        let attributes = participant.attributes();
        decision = classify_controlled_fixture_attributes(
            &participant.identity().to_string(),
            expected_participant,
            &attributes,
            &probe.sequence,
        );
        if decision.is_ok() || !matches!(decision, Err("missing_attributes")) {
            break;
        }
        if attempt + 1 < CONTROLLED_FIXTURE_PROBE_RECHECKS {
            sleep(CONTROLLED_FIXTURE_PROBE_RECHECK_DELAY).await;
        }
    }
    let (result, input_source_category, reject_reason) = match decision {
        Ok(()) => ("observed", Some("controlled_fixture"), None),
        Err(reason) => ("rejected", None, Some(reason)),
    };
    let ack = ControlledFixtureAttributeAck {
        message_type: CONTROLLED_FIXTURE_ACK_TOPIC,
        protocol_version: CONTROLLED_FIXTURE_PROTOCOL_VERSION,
        call_id_hash: sha256_hex(call_id),
        call_trace_id_hash: sha256_hex(trace_id),
        generation: CONTROLLED_FIXTURE_GENERATION,
        client_fixture_sequence: probe.sequence.clone(),
        result,
        input_source_category,
        reject_reason,
    };
    let payload = match serde_json::to_vec(&ack) {
        Ok(payload) => payload,
        Err(_) => return Some(false),
    };
    let local_participant = sink.room.local_participant();
    let publish = local_participant.publish_data(DataPacket {
        payload,
        topic: Some(CONTROLLED_FIXTURE_ACK_TOPIC.to_string()),
        reliable: true,
        destination_identities: vec![probe.sender],
    });
    if publish.await.is_ok() {
        acknowledged_probe_sequences.insert(probe.sequence);
    }
    Some(result == "observed")
}

async fn handle_finished_turn(
@@ -3243,6 +3489,10 @@
    F: FnOnce() -> std::collections::HashMap<String, String>,
{
    if !is_bound_user_participant(participant_identity, expected_participant) {
        warn!(
            "audioIngressOriginStatus" = "wrong_participant_or_track",
            "runtime helper rejected audio participant before VAD/session"
        );
        return (vad.in_speech, vad.in_speech, None);
    }
    observe_frame_and_start_session(
@@ -3281,7 +3531,10 @@
        Ok(metadata) => metadata,
        Err(reason) => {
            warn!(call_id = %call_id, trace_id = %trace_id, turn_id = %turn_id,
                reason, origin_status, "runtime helper asr_realtime_metadata_rejected");
                reason, audioIngressOriginStatus = %AudioIngressMetadata::rejected_origin_status(
                    reason, &attributes
                ),
                "runtime helper asr_realtime_metadata_rejected");
            return;
        }
    };
@@ -3291,6 +3544,7 @@
            &metadata.client_fixture_sequence,
        ) {
            warn!(call_id = %call_id, trace_id = %trace_id, turn_id = %turn_id,
                audioIngressOriginStatus = "sequence_replayed_or_regressed",
                "runtime helper asr_realtime_metadata_sequence_rejected");
            return;
        }
@@ -4089,11 +4343,7 @@

#[cfg(test)]
mod tests {
    use super::{
        ReplyChunkMarker, ReplyChunkMarkerState, RuntimeTurnDeviceOutput, RuntimeTurnStreamEvent,
        RuntimeTurnStreamState, RuntimeTurnStreamTimingPhase, runtime_session_nonce_hash,
        should_publish_device_output,
    };
    use super::*;
    use std::{
        collections::HashSet,
        io::{Read, Write},
@@ -4330,15 +4580,13 @@
        let second_request = request_rx
            .recv_timeout(Duration::from_secs(2))
            .expect("second session request");
        assert!(first_request.contains("\\\"clientFixtureSequence\\\":\\\"fixture-01\\\""));
        assert!(second_request.contains("\\\"clientFixtureSequence\\\":\\\"fixture-02\\\""));
        assert!(first_request.contains("\"clientFixtureSequence\":\"fixture-01\""));
        assert!(second_request.contains("\"clientFixtureSequence\":\"fixture-02\""));
        assert!(
            first_request
                .contains("\\\"audioIngressOriginStatus\\\":\\\"controlled_fixture_bound\\\"")
            first_request.contains("\"audioIngressOriginStatus\":\"controlled_fixture_bound\"")
        );
        assert!(
            second_request
                .contains("\\\"audioIngressOriginStatus\\\":\\\"controlled_fixture_bound\\\"")
            second_request.contains("\"audioIngressOriginStatus\":\"controlled_fixture_bound\"")
        );
        // The same production boundary rejects a wrong participant before VAD/session creation.
        assert!(!is_bound_user_participant(
@@ -4432,6 +4680,38 @@
            true,
        );
        assert!(upload.is_none());
        vad.reset_current_turn();
        attrs.remove("clientFixtureSequence");
        let (_, _, missing_sequence_turn) = observe_bound_participant_frame(
            "participant-user",
            Some("participant-user"),
            &mut vad,
            "call-001",
            "trace-001",
            "participant-user",
            "track-001",
            6,
            6_000,
            &frame,
            Client::new(),
            RealtimeAsrConfig {
                enabled: true,
                url: Some(format!("http://{address}/runtime/asr/realtime")),
                runtime_token: Some("test".to_string()),
                runtime_session_nonce: Some("test".to_string()),
                chunk_duration_ms: 200,
            },
            || attrs.clone(),
            &mut upload,
            &mut last_fixture_sequence,
            true,
        );
        assert!(missing_sequence_turn.is_none() && upload.is_none());
        assert_eq!(
            "sequence_absent",
            AudioIngressMetadata::rejected_origin_status("incomplete_metadata", &attrs)
        );
        assert_eq!(2, captured_count.load(Ordering::SeqCst));
        vad.reset_current_turn();
        attrs.insert("inputSourceCategory".to_string(), "other".to_string());
        let mut invalid_upload = None;
@@ -4594,4 +4874,150 @@
            assert!(published.is_empty());
        }
    }

    #[test]
    fn controlled_fixture_probe_requires_bound_hashes_and_protocol() {
        let call_id = "call-ack-1";
        let trace_id = "trace-ack-1";
        let payload = serde_json::to_vec(&json!({
            "type": CONTROLLED_FIXTURE_PROBE_TOPIC,
            "protocolVersion": CONTROLLED_FIXTURE_PROTOCOL_VERSION,
            "callIdHash": sha256_hex(call_id),
            "callTraceIdHash": sha256_hex(trace_id),
            "generation": CONTROLLED_FIXTURE_GENERATION,
            "clientFixtureSequence": "fixture-01"
        }))
        .unwrap();
        let sender = ParticipantIdentity("user-1".to_string());
        let pending =
            controlled_fixture_probe(&payload, call_id, trace_id, &sender, Some("user-1"))
                .expect("valid probe");
        assert_eq!(pending.sequence, "fixture-01");
        assert!(
            controlled_fixture_probe(&payload, call_id, "other-trace", &sender, Some("user-1"),)
                .is_none()
        );
        assert!(
            controlled_fixture_probe(
                &payload,
                call_id,
                trace_id,
                &ParticipantIdentity("other-user".to_string()),
                Some("user-1"),
            )
            .is_none()
        );
    }

    #[test]
    fn controlled_fixture_attributes_ack_only_on_exact_current_sequence() {
        let mut attributes = std::collections::HashMap::from([
            (
                "inputSourceCategory".to_string(),
                "controlled_fixture".to_string(),
            ),
            (
                "clientFixtureSequence".to_string(),
                "fixture-01".to_string(),
            ),
        ]);
        assert!(
            classify_controlled_fixture_attributes(
                "user-1",
                Some("user-1"),
                &attributes,
                "fixture-01"
            )
            .is_ok()
        );
        attributes.insert(
            "clientFixtureSequence".to_string(),
            "fixture-02".to_string(),
        );
        assert_eq!(
            classify_controlled_fixture_attributes(
                "user-1",
                Some("user-1"),
                &attributes,
                "fixture-01"
            ),
            Err("wrong_sequence")
        );
        attributes.remove("clientFixtureSequence");
        assert_eq!(
            classify_controlled_fixture_attributes(
                "user-1",
                Some("user-1"),
                &attributes,
                "fixture-01"
            ),
            Err("missing_sequence")
        );
        assert_eq!(
            classify_controlled_fixture_attributes(
                "other-user",
                Some("user-1"),
                &attributes,
                "fixture-01"
            ),
            Err("wrong_participant")
        );
    }

    #[test]
    fn controlled_fixture_ack_payload_is_reliable_and_redacted() {
        let ack = ControlledFixtureAttributeAck {
            message_type: CONTROLLED_FIXTURE_ACK_TOPIC,
            protocol_version: CONTROLLED_FIXTURE_PROTOCOL_VERSION,
            call_id_hash: sha256_hex("call-1"),
            call_trace_id_hash: sha256_hex("trace-1"),
            generation: CONTROLLED_FIXTURE_GENERATION,
            client_fixture_sequence: "fixture-01".to_string(),
            result: "observed",
            input_source_category: Some("controlled_fixture"),
            reject_reason: None,
        };
        let encoded = serde_json::to_vec(&ack).unwrap();
        let decoded: serde_json::Value = serde_json::from_slice(&encoded).unwrap();
        assert_eq!(decoded["type"], CONTROLLED_FIXTURE_ACK_TOPIC);
        assert_eq!(
            decoded["protocolVersion"],
            CONTROLLED_FIXTURE_PROTOCOL_VERSION
        );
        assert_eq!(decoded["result"], "observed");
        assert!(decoded.get("callId").is_none());
        assert!(decoded.get("traceId").is_none());
        assert!(decoded.get("participantIdentity").is_none());
        assert!(decoded.get("audio").is_none());
    }

    #[test]
    fn controlled_fixture_probe_does_not_create_session_or_audio_side_effects() {
        let attributes = std::collections::HashMap::new();
        assert_eq!(
            classify_controlled_fixture_attributes(
                "user-1",
                Some("user-1"),
                &attributes,
                "fixture-01"
            ),
            Err("missing_attributes")
        );
        assert_eq!(
            CONTROLLED_FIXTURE_PROBE_TOPIC,
            "controlled_fixture_attribute_probe"
        );
        assert_eq!(
            CONTROLLED_FIXTURE_ACK_TOPIC,
            "controlled_fixture_attribute_ack"
        );
    }

    #[test]
    fn controlled_fixture_probe_must_be_observed_before_audio_observer() {
        assert!(controlled_fixture_observer_gate(false, None));
        assert!(controlled_fixture_observer_gate(true, Some(true)));
        assert!(!controlled_fixture_observer_gate(true, Some(false)));
        assert!(!controlled_fixture_observer_gate(true, None));
    }
}