Detach realtime sideband observers without ending the session (#57817)

## Why

Stopping a sideband observer should close its WebSocket while preserving the frontend-owned realtime session. The existing frameless session close also sends `session.close`, so observer shutdown needs a transport-only close.

## What changed

- Add `close_transport()` to send a normal WebSocket close without `session.close`.
- Close the sideband transport on cancellation or when the call becomes inactive, including an existing connection awaiting adoption by the input task. Bound the close attempt to five seconds.
- Thread an optional sideband base URL through realtime start processing and allow nested thread ID fields in request serialization scopes.

## Testing

Add WebSocket and core conversation regression tests covering observer detachment for both realtime parsers and conversation versions. Verify that explicit frameless session close still sends `session.close`.

GitOrigin-RevId: 874a8980b0fcd45058e976b00c0d968f5b4b2404
This commit is contained in:
Jiahao
2026-10-28 20:10:15 +00:00
committed by copyberry
parent c92700d299
commit ef186f812a
8 changed files with 253 additions and 14 deletions
@@ -55,6 +55,7 @@ use tracing::error;
use tracing::info;
use tracing::trace;
use tracing::warn;
use tungstenite::protocol::CloseFrame;
use tungstenite::protocol::WebSocketConfig;
use tungstenite::protocol::frame::coding::CloseCode;
use url::Url;
@@ -76,6 +77,7 @@ enum WsCommand {
tx_result: oneshot::Sender<Result<(), WsError>>,
},
Close {
frame: Option<CloseFrame>,
tx_result: oneshot::Sender<Result<(), WsError>>,
},
}
@@ -116,9 +118,9 @@ impl WsStream {
break;
}
}
WsCommand::Close { tx_result } => {
WsCommand::Close { frame, tx_result } => {
info!("realtime websocket sending close");
let result = inner.send(Message::Close(None)).await;
let result = inner.send(Message::Close(frame)).await;
if let Err(err) = &result {
error!("realtime websocket close failed: {err}");
report_write_policy_error(err);
@@ -226,8 +228,10 @@ impl WsStream {
.await
}
async fn close(&self) -> Result<(), WsError> {
self.request(|tx_result| WsCommand::Close { tx_result })
/// Ask the socket pump to send a close frame to the connected peer and
/// return the write result.
async fn close(&self, frame: Option<CloseFrame>) -> Result<(), WsError> {
self.request(|tx_result| WsCommand::Close { frame, tx_result })
.await
}
}
@@ -493,7 +497,7 @@ impl RealtimeWebsocketWriter {
)));
}
}
if let Err(err) = self.stream.close().await
if let Err(err) = self.stream.close(/*frame*/ None).await
&& !matches!(err, WsError::ConnectionClosed | WsError::AlreadyClosed)
{
return Err(ApiError::Stream(format!(
@@ -503,6 +507,20 @@ impl RealtimeWebsocketWriter {
Ok(())
}
/// Detach an observer without ending the frontend-owned realtime session.
pub async fn close_transport(&self) -> Result<(), ApiError> {
if self.is_closed.swap(true, Ordering::SeqCst) {
return Ok(());
}
self.stream
.close(Some(CloseFrame {
code: CloseCode::Normal,
reason: "".into(),
}))
.await
.map_err(|err| ApiError::Stream(format!("failed to close websocket transport: {err}")))
}
async fn send_json(&self, message: &RealtimeOutboundMessage) -> Result<(), ApiError> {
match message {
RealtimeOutboundMessage::DelegationContextAppend {
@@ -25,6 +25,7 @@ use futures::SinkExt;
use futures::StreamExt;
use http::HeaderMap;
use http::StatusCode;
use pretty_assertions::assert_eq;
use serde_json::Value;
use serde_json::json;
use tokio::net::TcpListener;
@@ -33,6 +34,8 @@ use tokio_tungstenite::accept_hdr_async;
use tokio_tungstenite::tungstenite::Message;
use tokio_tungstenite::tungstenite::handshake::server::Request;
use tokio_tungstenite::tungstenite::handshake::server::Response;
use tokio_tungstenite::tungstenite::protocol::CloseFrame;
use tokio_tungstenite::tungstenite::protocol::frame::coding::CloseCode;
type RealtimeWsStream = tokio_tungstenite::WebSocketStream<tokio::net::TcpStream>;
@@ -372,6 +375,108 @@ async fn realtime_ws_existing_call_retries_activation_404_with_a_bounded_budget(
}
}
#[tokio::test]
async fn realtime_ws_existing_call_transport_close_preserves_session() {
for event_parser in [
RealtimeEventParser::RealtimeV2,
RealtimeEventParser::FramelessBidi,
] {
let (addr, server) = spawn_realtime_ws_server(|mut ws| async move {
assert_eq!(
ws.next().await.expect("close frame").expect("valid close"),
Message::Close(Some(CloseFrame {
code: CloseCode::Normal,
reason: "".into(),
})),
"observer detach must not send an application session.close"
);
})
.await;
let client = test_client(test_provider(format!("http://{addr}")))
.with_webrtc_sideband_base_url(format!("http://{addr}"));
let connection = client
.connect_existing_call_sideband(
RealtimeSessionConfig {
instructions: String::new(),
initial_items: Vec::new(),
delegation_ack_filler: None,
model: None,
session_id: None,
event_parser,
session_mode: RealtimeSessionMode::Conversational,
output_modality: RealtimeOutputModality::Audio,
voice: RealtimeVoice::Marin,
},
"rtc_frontend_owned",
HeaderMap::new(),
HeaderMap::new(),
RealtimeTranscriptState::default(),
)
.await
.expect("attach observer");
connection
.writer()
.close_transport()
.await
.expect("detach transport");
connection.close().await.expect("already detached");
tokio::time::timeout(Duration::from_secs(5), server)
.await
.expect("close was flushed before dropping connection")
.expect("server task");
}
}
#[tokio::test]
async fn realtime_ws_frameless_session_close_still_ends_session() {
let (addr, server) = spawn_realtime_ws_server(|mut ws| async move {
let session_close = ws
.next()
.await
.expect("session close")
.expect("valid frame");
assert_eq!(
serde_json::from_str::<Value>(session_close.to_text().expect("text")).expect("json"),
json!({"type": "session.close"})
);
assert_eq!(
ws.next()
.await
.expect("websocket close")
.expect("valid close"),
Message::Close(None)
);
})
.await;
let client = test_client(test_provider(format!("http://{addr}")))
.with_webrtc_sideband_base_url(format!("http://{addr}"));
let connection = client
.connect_existing_call_sideband(
RealtimeSessionConfig {
instructions: String::new(),
initial_items: Vec::new(),
delegation_ack_filler: None,
model: None,
session_id: None,
event_parser: RealtimeEventParser::FramelessBidi,
session_mode: RealtimeSessionMode::Conversational,
output_modality: RealtimeOutputModality::Audio,
voice: RealtimeVoice::Marin,
},
"rtc_session_owner",
HeaderMap::new(),
HeaderMap::new(),
RealtimeTranscriptState::default(),
)
.await
.expect("connect session");
connection.close().await.expect("close session");
tokio::time::timeout(Duration::from_secs(5), server)
.await
.expect("close flushed")
.expect("server task");
}
#[tokio::test]
async fn realtime_ws_e2e_send_while_next_event_waits() {
let (addr, server) = spawn_realtime_ws_server(|mut ws: RealtimeWsStream| async move {