From caeed9acdc676aa10f788e81edf45892b8cdcc8e Mon Sep 17 00:00:00 2001
From: cai <cai@nbcai.cc>
Date: Tue, 25 Aug 2026 12:48:13 +0800
Subject: [PATCH] fix(helper): bind ASR ingress to fixture generation
---
src/main.rs | 332 +++++++++++++++++++++++++++++++++++++++++++++---------
1 files changed, 275 insertions(+), 57 deletions(-)
diff --git a/src/main.rs b/src/main.rs
index af61418..b18a534 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -70,7 +70,7 @@
const CONTROLLED_FIXTURE_ACK_RESULTS: [&str; 4] =
["observed", "rejected", "timeout", "publish_failed"];
-const CONTROLLED_FIXTURE_REJECT_REASONS: [&str; 10] = [
+const CONTROLLED_FIXTURE_REJECT_REASONS: [&str; 16] = [
"missing_attributes",
"wrong_source",
"missing_sequence",
@@ -80,6 +80,12 @@
"duplicate_or_old_sequence",
"no_current_participant",
"ack_publish_failed",
+ "missing_generation",
+ "invalid_generation",
+ "wrong_generation",
+ "invalid_language",
+ "incomplete_metadata",
+ "invalid_source_or_sequence",
"unknown",
];
@@ -264,7 +270,7 @@
trace_id_hash: &str,
generation: u64,
sequence: &str,
- acknowledged_probe_sequences: &mut HashSet<String>,
+ acknowledged_probe_sequences: &mut HashSet<(u64, String)>,
) -> ControlledFixtureAckPublishOutcome
where
F: Future<Output = Result<(), E>>,
@@ -282,7 +288,7 @@
Some(ack_result),
reject_reason,
);
- acknowledged_probe_sequences.insert(sequence.to_string());
+ acknowledged_probe_sequences.insert((generation, sequence.to_string()));
ControlledFixtureAckPublishOutcome {
observed,
published: true,
@@ -338,7 +344,7 @@
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
+ || !valid_input_generation(probe.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()
@@ -355,6 +361,25 @@
received_at,
expires_at: received_at + CONTROLLED_FIXTURE_PROBE_TTL,
})
+}
+
+fn controlled_fixture_probe_binding_decision(
+ probe_sender: &str,
+ current_audio_participant: Option<&str>,
+ expected_participant: Option<&str>,
+) -> Result<(), &'static str> {
+ if !is_bound_user_participant(probe_sender, expected_participant) {
+ return Err("wrong_participant");
+ }
+ let Some(current_audio_participant) = current_audio_participant else {
+ return Err("no_current_participant");
+ };
+ if current_audio_participant != probe_sender
+ || !is_bound_user_participant(current_audio_participant, expected_participant)
+ {
+ return Err("wrong_participant");
+ }
+ Ok(())
}
fn controlled_fixture_visibility_bucket(elapsed: Duration) -> &'static str {
@@ -546,6 +571,58 @@
return Err("wrong_sequence");
}
Ok(())
+}
+
+fn controlled_fixture_probe_attribute_decision(
+ actual_participant: &str,
+ expected_participant: Option<&str>,
+ attributes: &std::collections::HashMap<String, String>,
+ probe: &PendingControlledFixtureProbe,
+) -> Result<(), &'static str> {
+ if !is_bound_user_participant(actual_participant, expected_participant) {
+ return Err("wrong_participant");
+ }
+ let metadata = match AudioIngressMetadata::from_participant(attributes) {
+ Ok(Some(metadata)) => metadata,
+ Ok(None) => return Err("missing_attributes"),
+ Err(reason) => return Err(reason),
+ };
+ if metadata.client_fixture_sequence != probe.sequence {
+ return Err("wrong_sequence");
+ }
+ if metadata.input_generation != probe.generation {
+ return Err("wrong_generation");
+ }
+ Ok(())
+}
+
+async fn observe_controlled_fixture_probe_attributes<F>(
+ expires_at: Instant,
+ actual_participant: &str,
+ expected_participant: Option<&str>,
+ probe: &PendingControlledFixtureProbe,
+ mut read_attributes: F,
+) -> Result<(), &'static str>
+where
+ F: FnMut() -> std::collections::HashMap<String, String>,
+{
+ loop {
+ if Instant::now() > expires_at {
+ return Err("timeout");
+ }
+ match controlled_fixture_probe_attribute_decision(
+ actual_participant,
+ expected_participant,
+ &read_attributes(),
+ probe,
+ ) {
+ Ok(()) => return Ok(()),
+ Err("missing_attributes") | Err("missing_generation") => {
+ sleep(CONTROLLED_FIXTURE_PROBE_RECHECK_DELAY).await;
+ }
+ Err(reason) => return Err(reason),
+ }
+ }
}
async fn observe_controlled_fixture_attributes<F>(
@@ -1282,6 +1359,10 @@
expected.is_none_or(|value| identity == value)
}
+fn valid_input_generation(generation: u64) -> bool {
+ generation > 0
+}
+
async fn observe_user_audio_events(
mut events: UnboundedReceiver<RoomEvent>,
call_id: String,
@@ -1410,6 +1491,18 @@
}
current_user_participant = Some(participant_for_probe);
}
+ RoomEvent::TrackUnsubscribed {
+ track: RemoteTrack::Audio(_),
+ publication: _,
+ participant,
+ } => {
+ if current_user_participant
+ .as_ref()
+ .is_some_and(|current| current.identity() == participant.identity())
+ {
+ current_user_participant = None;
+ }
+ }
RoomEvent::DataReceived {
payload,
topic: Some(topic),
@@ -1419,7 +1512,9 @@
if let Ok(probe) =
serde_json::from_slice::<ControlledFixtureAttributeProbe>(&payload)
{
- if acknowledged_probe_sequences.contains(&probe.client_fixture_sequence) {
+ if acknowledged_probe_sequences
+ .contains(&(probe.generation, probe.client_fixture_sequence.clone()))
+ {
record_controlled_fixture_probe_event(
&call_id,
&trace_id,
@@ -1512,7 +1607,7 @@
async fn process_controlled_fixture_probe(
pending_probe: &mut Option<PendingControlledFixtureProbe>,
- acknowledged_probe_sequences: &mut HashSet<String>,
+ acknowledged_probe_sequences: &mut HashSet<(u64, String)>,
participant: Option<&RemoteParticipant>,
sink: &BotAudioOutputSink,
expected_participant: Option<&str>,
@@ -1523,7 +1618,7 @@
let Some(probe) = pending_probe.take() else {
return None;
};
- if acknowledged_probe_sequences.contains(&probe.sequence) {
+ if acknowledged_probe_sequences.contains(&(probe.generation, probe.sequence.clone())) {
record_controlled_fixture_probe_event(
runtime_call_id,
runtime_trace_id,
@@ -1584,16 +1679,28 @@
);
return Some(false);
}
- let decision = observe_controlled_fixture_attributes(
+ let decision = observe_controlled_fixture_probe_attributes(
probe.expires_at,
- &participant.identity().to_string(),
+ participant.identity().as_str(),
expected_participant,
- &probe.sequence,
+ &probe,
|| participant.attributes(),
)
.await;
+ if let Err(reason) = decision {
+ record_controlled_fixture_attribute_decision(
+ Err(reason),
+ runtime_call_id,
+ runtime_trace_id,
+ &probe.call_id_hash,
+ &probe.call_trace_id_hash,
+ probe.generation,
+ &probe.sequence,
+ );
+ return Some(false);
+ }
let (ack_result, reject_reason, observed) = record_controlled_fixture_attribute_decision(
- decision,
+ Ok(()),
runtime_call_id,
runtime_trace_id,
&probe.call_id_hash,
@@ -3802,6 +3909,7 @@
};
let mut realtime_asr_upload: Option<RealtimeAsrUpload> = None;
let mut last_fixture_sequence: Option<String> = None;
+ let mut last_fixture_generation: Option<u64> = None;
while let Some(drained) = frame_rx.recv().await {
let frame = drained.frame;
@@ -3852,6 +3960,7 @@
|| participant.attributes(),
&mut realtime_asr_upload,
&mut last_fixture_sequence,
+ &mut last_fixture_generation,
turn_bridge_config.asr_realtime_enabled,
);
@@ -3988,6 +4097,7 @@
read_attributes: F,
upload_slot: &mut Option<RealtimeAsrUpload>,
last_fixture_sequence: &mut Option<String>,
+ last_fixture_generation: &mut Option<u64>,
realtime_enabled: bool,
) -> (bool, bool, Option<FinishedSpeechTurn>)
where
@@ -4014,6 +4124,7 @@
read_attributes,
upload_slot,
last_fixture_sequence,
+ last_fixture_generation,
realtime_enabled,
);
}
@@ -4036,6 +4147,7 @@
read_attributes: F,
upload_slot: &mut Option<RealtimeAsrUpload>,
last_fixture_sequence: &mut Option<String>,
+ last_fixture_generation: &mut Option<u64>,
realtime_enabled: bool,
) -> (bool, bool, Option<FinishedSpeechTurn>)
where
@@ -4062,6 +4174,7 @@
read_attributes,
upload_slot,
last_fixture_sequence,
+ last_fixture_generation,
realtime_enabled,
)
}
@@ -4075,6 +4188,7 @@
read_attributes: impl FnOnce() -> std::collections::HashMap<String, String>,
upload_slot: &mut Option<RealtimeAsrUpload>,
last_fixture_sequence: &mut Option<String>,
+ last_fixture_generation: &mut Option<u64>,
realtime_enabled: bool,
) {
let turn_id = format!("turn-{:04}", vad.turn_index);
@@ -4092,9 +4206,11 @@
}
};
if let Some(metadata) = metadata.as_ref() {
- if !fixture_sequence_is_new(
+ if !fixture_binding_is_new(
last_fixture_sequence.as_deref(),
+ *last_fixture_generation,
&metadata.client_fixture_sequence,
+ metadata.input_generation,
) {
warn!(call_id = %call_id, trace_id = %trace_id, turn_id = %turn_id,
audioIngressOriginStatus = "sequence_replayed_or_regressed",
@@ -4114,6 +4230,7 @@
Ok(upload) => {
if let Some(metadata) = metadata {
*last_fixture_sequence = Some(metadata.client_fixture_sequence);
+ *last_fixture_generation = Some(metadata.input_generation);
}
info!(call_id = %call_id, trace_id = %trace_id, turn_id = %turn_id,
origin_status, "runtime helper asr_realtime_session_started");
@@ -4141,6 +4258,23 @@
match (previous_number, current_number) {
(Some(previous), Some(current)) => current > previous,
_ => previous != current,
+ }
+}
+
+fn fixture_binding_is_new(
+ previous_sequence: Option<&str>,
+ previous_generation: Option<u64>,
+ current_sequence: &str,
+ current_generation: u64,
+) -> bool {
+ if !valid_input_generation(current_generation) {
+ return false;
+ }
+ match previous_generation {
+ Some(previous) if current_generation < previous => false,
+ Some(previous) if current_generation > previous => true,
+ Some(_) => fixture_sequence_is_new(previous_sequence, current_sequence),
+ None => true,
}
}
@@ -4912,35 +5046,42 @@
#[derive(Debug)]
enum PreAudioOrderEvent {
- DataReceived {
- sender: String,
- sequence: String,
- },
- TrackSubscribed {
- participant: String,
- attributes: HashMap<String, String>,
- },
+ DataReceived { sender: String, sequence: String },
+ TrackSubscribed { participant: String },
}
fn drive_pre_audio_order_test_seam(events: &[PreAudioOrderEvent]) -> Vec<&'static str> {
- let mut pending_sequence = None;
+ let call_id = "production-order-call";
+ let trace_id = "production-order-trace";
+ let mut pending_probe = None;
let mut effects = Vec::new();
for event in events {
match event {
- PreAudioOrderEvent::DataReceived { sender, sequence } if sender == "user-1" => {
- pending_sequence = Some(sequence.as_str());
+ PreAudioOrderEvent::DataReceived { sender, sequence } => {
+ 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": sequence,
+ }))
+ .expect("production probe payload");
+ pending_probe = controlled_fixture_probe(
+ &payload,
+ call_id,
+ trace_id,
+ &ParticipantIdentity(sender.clone()),
+ Some("user-1"),
+ );
}
- PreAudioOrderEvent::TrackSubscribed {
- participant,
- attributes,
- } => {
- let pending = pending_sequence.is_some();
- let probe_result = pending_sequence.map(|sequence| {
- classify_controlled_fixture_attributes(
- participant,
+ PreAudioOrderEvent::TrackSubscribed { participant } => {
+ let pending = pending_probe.is_some();
+ let probe_result = pending_probe.as_ref().map(|probe| {
+ controlled_fixture_probe_binding_decision(
+ probe.sender.as_str(),
+ Some(participant),
Some("user-1"),
- attributes,
- sequence,
)
.is_ok()
});
@@ -4958,9 +5099,8 @@
if observer_started {
effects.push("observer_started");
}
- pending_sequence = None;
+ pending_probe = None;
}
- PreAudioOrderEvent::DataReceived { .. } => {}
}
}
effects
@@ -4994,6 +5134,8 @@
"clientFixtureSequence".to_string(),
"fixture-01".to_string(),
),
+ ("inputGeneration".to_string(), "1".to_string()),
+ ("language".to_string(), "ja-JP".to_string()),
]);
let mut starts = Vec::new();
for (session_index, sequence) in [(1, "fixture-01"), (2, "fixture-02")] {
@@ -5019,6 +5161,7 @@
let is_in_speech = vad.in_speech;
assert!(!was_in_speech && is_in_speech);
attributes.insert("clientFixtureSequence".to_string(), sequence.to_string());
+ attributes.insert("language".to_string(), "ja-JP".to_string());
let metadata = AudioIngressMetadata::from_participant(&attributes)
.expect("valid participant attributes")
.expect("controlled fixture metadata");
@@ -5033,6 +5176,7 @@
let session_json: serde_json::Value =
serde_json::from_slice(&session_line).expect("session start json");
assert_eq!(sequence, session_json["clientFixtureSequence"]);
+ assert_eq!("ja-JP", session_json["language"]);
starts.push(metadata.client_fixture_sequence);
vad.reset_current_turn();
}
@@ -5125,9 +5269,12 @@
"clientFixtureSequence".to_string(),
"fixture-01".to_string(),
),
+ ("inputGeneration".to_string(), "1".to_string()),
+ ("language".to_string(), "ja-JP".to_string()),
]);
let mut upload = None;
let mut last_fixture_sequence = None;
+ let mut last_fixture_generation = None;
let config = RealtimeAsrConfig {
enabled: true,
url: Some(format!("http://{address}/runtime/asr/realtime")),
@@ -5151,6 +5298,7 @@
|| attrs.clone(),
&mut upload,
&mut last_fixture_sequence,
+ &mut last_fixture_generation,
true,
);
assert!(!was && is && turn.is_none());
@@ -5161,6 +5309,8 @@
"clientFixtureSequence".to_string(),
"fixture-02".to_string(),
);
+ attrs.insert("inputGeneration".to_string(), "2".to_string());
+ attrs.insert("language".to_string(), "zh-CN".to_string());
let (was, is, turn) = observe_bound_participant_frame(
"participant-user",
Some("participant-user"),
@@ -5173,10 +5323,11 @@
2_000,
&frame,
Client::new(),
- config,
+ config.clone(),
|| attrs.clone(),
&mut upload,
&mut last_fixture_sequence,
+ &mut last_fixture_generation,
true,
);
assert!(!was && is && turn.is_none());
@@ -5191,6 +5342,16 @@
.expect("second session request");
assert!(first_request.contains("\"clientFixtureSequence\":\"fixture-01\""));
assert!(second_request.contains("\"clientFixtureSequence\":\"fixture-02\""));
+ assert!(first_request.contains("\"callId\":\"call-001\""));
+ assert!(first_request.contains("\"traceId\":\"trace-001\""));
+ assert!(second_request.contains("\"callId\":\"call-001\""));
+ assert!(second_request.contains("\"traceId\":\"trace-001\""));
+ assert!(first_request.contains("\"inputSourceCategory\":\"controlled_fixture\""));
+ assert!(second_request.contains("\"inputSourceCategory\":\"controlled_fixture\""));
+ assert!(first_request.contains("\"inputGeneration\":1"));
+ assert!(second_request.contains("\"inputGeneration\":2"));
+ assert!(first_request.contains("\"language\":\"ja-JP\""));
+ assert!(second_request.contains("\"language\":\"zh-CN\""));
assert!(
first_request.contains("\"audioIngressOriginStatus\":\"controlled_fixture_bound\"")
);
@@ -5207,6 +5368,32 @@
attrs.insert(
"clientFixtureSequence".to_string(),
"fixture-03".to_string(),
+ );
+ attrs.insert("inputGeneration".to_string(), "1".to_string());
+ let (_, is_old_generation, old_generation_turn) = observe_bound_participant_frame(
+ "participant-user",
+ Some("participant-user"),
+ &mut vad,
+ "call-001",
+ "trace-001",
+ "participant-user",
+ "track-001",
+ 3,
+ 3_000,
+ &frame,
+ Client::new(),
+ config.clone(),
+ || attrs.clone(),
+ &mut upload,
+ &mut last_fixture_sequence,
+ &mut last_fixture_generation,
+ true,
+ );
+ assert!(is_old_generation && old_generation_turn.is_none() && upload.is_none());
+ vad.reset_current_turn();
+ attrs.insert(
+ "clientFixtureSequence".to_string(),
+ "fixture-04".to_string(),
);
let (_, is_wrong, wrong_turn) = observe_bound_participant_frame(
"participant-other",
@@ -5230,6 +5417,7 @@
|| attrs.clone(),
&mut upload,
&mut last_fixture_sequence,
+ &mut last_fixture_generation,
true,
);
assert!(!is_wrong && wrong_turn.is_none() && upload.is_none());
@@ -5260,6 +5448,7 @@
|| attrs.clone(),
&mut upload,
&mut last_fixture_sequence,
+ &mut last_fixture_generation,
true,
);
assert!(upload.is_none());
@@ -5286,6 +5475,7 @@
|| attrs.clone(),
&mut upload,
&mut last_fixture_sequence,
+ &mut last_fixture_generation,
true,
);
assert!(upload.is_none());
@@ -5313,6 +5503,7 @@
|| attrs.clone(),
&mut upload,
&mut last_fixture_sequence,
+ &mut last_fixture_generation,
true,
);
assert!(missing_sequence_turn.is_none() && upload.is_none());
@@ -5346,6 +5537,7 @@
|| attrs,
&mut invalid_upload,
&mut last_fixture_sequence,
+ &mut last_fixture_generation,
true,
);
assert!(invalid_turn.is_none());
@@ -5788,7 +5980,7 @@
published: true,
}
);
- assert!(acknowledged.contains("fixture-02"));
+ assert!(acknowledged.contains(&(CONTROLLED_FIXTURE_GENERATION, "fixture-02".to_string())));
assert_eq!(successful_observer_starts, 1);
}
@@ -5822,6 +6014,49 @@
.await;
assert_eq!(result, Ok(()));
assert_eq!(reads, 4);
+ }
+
+ #[tokio::test]
+ async fn production_probe_waits_for_generation_attribute_before_binding() {
+ let received_at = Instant::now();
+ let probe = PendingControlledFixtureProbe {
+ sender: ParticipantIdentity("user-1".to_string()),
+ call_id_hash: "call-hash".to_string(),
+ call_trace_id_hash: "trace-hash".to_string(),
+ generation: 2,
+ sequence: "fixture-02".to_string(),
+ received_at,
+ expires_at: received_at + CONTROLLED_FIXTURE_PROBE_TTL,
+ };
+ let expected = HashMap::from([
+ (
+ "inputSourceCategory".to_string(),
+ "controlled_fixture".to_string(),
+ ),
+ (
+ "clientFixtureSequence".to_string(),
+ "fixture-02".to_string(),
+ ),
+ ("inputGeneration".to_string(), "2".to_string()),
+ ]);
+ let mut reads = 0;
+ let result = observe_controlled_fixture_probe_attributes(
+ probe.expires_at,
+ "user-1",
+ Some("user-1"),
+ &probe,
+ || {
+ reads += 1;
+ if reads == 1 {
+ HashMap::new()
+ } else {
+ expected.clone()
+ }
+ },
+ )
+ .await;
+ assert_eq!(result, Ok(()));
+ assert_eq!(reads, 2);
}
#[tokio::test]
@@ -6174,24 +6409,13 @@
#[test]
fn production_event_order_probe_then_track_publishes_ack_before_observer() {
- let attributes = HashMap::from([
- (
- "inputSourceCategory".to_string(),
- "controlled_fixture".to_string(),
- ),
- (
- "clientFixtureSequence".to_string(),
- "fixture-01".to_string(),
- ),
- ]);
let effects = drive_pre_audio_order_test_seam(&[
PreAudioOrderEvent::DataReceived {
sender: "user-1".to_string(),
- sequence: "fixture-01".to_string(),
+ sequence: "1".to_string(),
},
PreAudioOrderEvent::TrackSubscribed {
participant: "user-1".to_string(),
- attributes,
},
]);
assert_eq!(effects, ["ack_observed", "observer_started"]);
@@ -6199,19 +6423,13 @@
#[test]
fn production_event_order_negative_probe_has_no_observer_or_session_effect() {
- let mut invalid = HashMap::new();
- invalid.insert(
- "inputSourceCategory".to_string(),
- "ordinary_mic".to_string(),
- );
let effects = drive_pre_audio_order_test_seam(&[
PreAudioOrderEvent::DataReceived {
sender: "user-1".to_string(),
- sequence: "fixture-01".to_string(),
+ sequence: "1".to_string(),
},
PreAudioOrderEvent::TrackSubscribed {
- participant: "user-1".to_string(),
- attributes: invalid,
+ participant: "cross-call-user".to_string(),
},
]);
assert!(effects.is_empty());
--
Gitblit v1.9.3