Remove WebSocket transport from code-mode hosts (#40692)

## What changed

- Accept only `http://` and `https://` gRPC endpoints for app server
  `--code-mode-host` connections.
- Limit the standalone code-mode host to stdio and gRPC listeners.
- Remove the WebSocket session provider, dual-WebSocket negotiation, transport
  lanes, and related dependencies and tests.
- Keep coverage for shared and prewarmed gRPC hosts and reject WebSocket host
  URLs in CLI parsing tests.

GitOrigin-RevId: 8792604e79ffd54ce776074a4dc3f7d55e63752a
This commit is contained in:
Ruslan Nigmatullin
2026-08-25 20:37:15 +00:00
committed by copyberry
parent 9be8d6e1c3
commit 725b3a44f8
31 changed files with 153 additions and 2766 deletions
-1
View File
@@ -6,4 +6,3 @@ pub use grpc_session::GrpcCodeModeSessionProvider;
pub use remote_session::DisabledCodeModeSessionProvider;
pub use remote_session::ProcessOwnedCodeModeSession;
pub use remote_session::ProcessOwnedCodeModeSessionProvider;
pub use remote_session::WebSocketCodeModeSessionProvider;
+5 -85
View File
@@ -18,8 +18,6 @@ use codex_code_mode_protocol::StartedCell;
use codex_code_mode_protocol::WaitOutcome;
use codex_code_mode_protocol::WaitRequest;
use codex_code_mode_protocol::host::SessionId;
use codex_http_client::HttpClientFactory;
use codex_http_client::OutboundProxyPolicy;
use codex_install_context::InstallContext;
use tokio::sync::Semaphore;
use tokio::sync::watch;
@@ -43,11 +41,6 @@ pub struct ProcessOwnedCodeModeSessionProvider {
#[derive(Default)]
pub struct DisabledCodeModeSessionProvider;
/// Creates code-mode sessions backed by one shared remote WebSocket connection.
pub struct WebSocketCodeModeSessionProvider {
host: Arc<OwnedCodeModeHost>,
}
impl ProcessOwnedCodeModeSessionProvider {
pub fn with_host_program(host_program: PathBuf) -> Self {
Self {
@@ -68,9 +61,7 @@ impl Default for ProcessOwnedCodeModeSessionProvider {
impl CodeModeSessionProvider for ProcessOwnedCodeModeSessionProvider {
fn availability(&self) -> Result<(), String> {
let HostEndpoint::Process(host_program) = &self.host.endpoint else {
unreachable!("a process-owned provider always has a process endpoint");
};
let host_program = &self.host.host_program;
if host_program.is_file() {
Ok(())
} else {
@@ -119,49 +110,6 @@ impl CodeModeSessionProvider for DisabledCodeModeSessionProvider {
}
}
impl WebSocketCodeModeSessionProvider {
pub fn new(websocket_url: String) -> Self {
Self::with_http_client_factory(
websocket_url,
HttpClientFactory::new(OutboundProxyPolicy::ReqwestDefault),
)
}
/// Creates a remote host using the application's effective proxy and TLS policy.
pub fn with_http_client_factory(
websocket_url: String,
http_client_factory: HttpClientFactory,
) -> Self {
Self {
host: Arc::new(OwnedCodeModeHost::websocket(
websocket_url,
http_client_factory,
)),
}
}
}
impl CodeModeSessionProvider for WebSocketCodeModeSessionProvider {
fn create_session<'a>(
&'a self,
delegate: Arc<dyn CodeModeSessionDelegate>,
) -> CodeModeSessionProviderFuture<'a> {
self.create_session_with_limits(delegate, CodeModeSessionCellExecutionLimits::default())
}
fn create_session_with_limits<'a>(
&'a self,
delegate: Arc<dyn CodeModeSessionDelegate>,
limits: CodeModeSessionCellExecutionLimits,
) -> CodeModeSessionProviderFuture<'a> {
Box::pin(create_host_session(
delegate,
Arc::clone(&self.host),
limits,
))
}
}
async fn create_host_session(
delegate: Arc<dyn CodeModeSessionDelegate>,
host: Arc<OwnedCodeModeHost>,
@@ -172,16 +120,8 @@ async fn create_host_session(
Ok(Arc::new(session))
}
enum HostEndpoint {
Process(PathBuf),
WebSocket {
websocket_url: String,
http_client_factory: HttpClientFactory,
},
}
struct OwnedCodeModeHost {
endpoint: HostEndpoint,
host_program: PathBuf,
connection: StdMutex<Option<Arc<Connection>>>,
connect_permit: Semaphore,
connection_generation: AtomicU64,
@@ -192,21 +132,7 @@ struct OwnedCodeModeHost {
impl OwnedCodeModeHost {
fn new(host_program: PathBuf) -> Self {
Self {
endpoint: HostEndpoint::Process(host_program),
connection: StdMutex::new(None),
connect_permit: Semaphore::new(/*permits*/ 1),
connection_generation: AtomicU64::new(0),
last_connection_error: StdMutex::new(None),
next_session_id: AtomicU64::new(1),
}
}
fn websocket(websocket_url: String, http_client_factory: HttpClientFactory) -> Self {
Self {
endpoint: HostEndpoint::WebSocket {
websocket_url,
http_client_factory,
},
host_program,
connection: StdMutex::new(None),
connect_permit: Semaphore::new(/*permits*/ 1),
connection_generation: AtomicU64::new(0),
@@ -238,13 +164,7 @@ impl OwnedCodeModeHost {
{
return Err(ConnectionError::Other(error.clone()));
}
let connection = match &self.endpoint {
HostEndpoint::Process(host_program) => Connection::spawn(host_program).await,
HostEndpoint::WebSocket {
websocket_url,
http_client_factory,
} => Connection::connect_websocket(websocket_url, http_client_factory).await,
};
let connection = Connection::spawn(&self.host_program).await;
let new_connection = match connection {
Ok(connection) => connection,
Err(error) => {
@@ -312,7 +232,7 @@ struct SessionInner {
retired_cleanups: StdMutex<Vec<SessionCleanup>>,
}
/// A logical code-mode session assigned to a process or WebSocket host.
/// A logical code-mode session assigned to a process host.
pub struct ProcessOwnedCodeModeSession {
inner: Arc<SessionInner>,
}
@@ -20,32 +20,24 @@ use codex_code_mode_protocol::host::Capability;
use codex_code_mode_protocol::host::CapabilitySet;
use codex_code_mode_protocol::host::ClientHello;
use codex_code_mode_protocol::host::ClientToHost;
use codex_code_mode_protocol::host::DUAL_WEBSOCKET_CAPABILITY;
use codex_code_mode_protocol::host::EncodedFrame;
use codex_code_mode_protocol::host::FramedReader;
use codex_code_mode_protocol::host::FramedWriter;
use codex_code_mode_protocol::host::HostToClient;
use codex_code_mode_protocol::host::MAX_FRAME_BYTES;
use codex_code_mode_protocol::host::MAX_PENDING_DELEGATE_CALLS;
use codex_code_mode_protocol::host::ProtocolVersion;
use codex_code_mode_protocol::host::RequestId;
use codex_code_mode_protocol::host::SESSION_RESOURCE_LIMITS_CAPABILITY;
use codex_code_mode_protocol::host::SupportedProtocolVersions;
use codex_code_mode_protocol::host::TransportLane;
use codex_http_client::HttpClientFactory;
use codex_protocol::shell_environment::scrub_non_inheritable_env_vars;
use codex_websocket_client::WebSocketConnector;
use futures::StreamExt;
use tokio::io::AsyncBufReadExt;
use tokio::io::BufReader;
use tokio::process::Child;
use tokio::process::ChildStdin;
use tokio::process::ChildStdout;
use tokio::process::Command;
use tokio::sync::mpsc;
use tokio::sync::oneshot;
use tokio::task::JoinHandle;
use tokio_tungstenite::tungstenite::client::IntoClientRequest;
use tokio_tungstenite::tungstenite::http::Uri;
use tokio_tungstenite::tungstenite::protocol::WebSocketConfig;
use tokio_util::sync::CancellationToken;
use tracing::debug;
use tracing::warn;
@@ -57,19 +49,14 @@ use self::driver::DriverLifecycle;
pub(super) use self::driver::RemoteSession;
pub(super) use self::driver::SessionCleanup;
use self::reader::drive_reader;
use self::transport::ConnectionReader;
use self::transport::ConnectionWriter;
mod driver;
mod reader;
mod transport;
const IPC_CHANNEL_CAPACITY: usize = 128;
const HOST_HANDSHAKE_TIMEOUT: Duration = Duration::from_secs(10);
const LOCAL_HOST_STARTUP_TIMEOUT: Duration = Duration::from_secs(30);
// TODO(anp) make this timeout configurable if 60 seconds is insufficient.
const DEFAULT_HOST_WAIT_TRANSPORT_TIMEOUT: Duration = Duration::from_secs(60);
const MAX_WEBSOCKET_FRAME_BYTES: usize = MAX_FRAME_BYTES + std::mem::size_of::<u32>();
// Host spawn errors become model-visible tool output. Bound configured paths
// while preserving the executable-bearing suffix needed to diagnose failures.
const MAX_DISPLAYED_HOST_PROGRAM_BYTES: usize = 512;
@@ -80,7 +67,6 @@ pub(super) enum ConnectionError {
host_program: PathBuf,
error: io::Error,
},
BulkConnectionUnavailable(String),
Other(String),
}
@@ -111,9 +97,7 @@ impl fmt::Display for ConnectionError {
&host_program[suffix_start..]
)
}
Self::BulkConnectionUnavailable(message) | Self::Other(message) => {
formatter.write_str(message)
}
Self::Other(message) => formatter.write_str(message),
}
}
}
@@ -133,7 +117,7 @@ struct CallerCancellation {
}
struct ConnectionSupervisor {
owner: ConnectionOwner,
child: Child,
event_tx: mpsc::Sender<DriverEvent>,
cancellation: CancellationToken,
alive: Arc<AtomicBool>,
@@ -143,50 +127,6 @@ struct ConnectionSupervisor {
writer_task: JoinHandle<Result<(), String>>,
}
enum ConnectionOwner {
Process(Box<Child>),
WebSocket,
}
struct BulkConnectionOptions {
websocket_url: String,
http_client_factory: HttpClientFactory,
}
impl BulkConnectionOptions {
async fn connect(
&self,
token: &str,
) -> Result<(ConnectionReader, ConnectionWriter), ConnectionError> {
let control_uri = self.websocket_url.parse::<Uri>().map_err(|error| {
ConnectionError::Other(format!(
"failed to build code-mode host bulk websocket URL: {error}"
))
})?;
let bulk_path = format!("{}/bulk/{token}", control_uri.path().trim_end_matches('/'));
let bulk_path_and_query = match control_uri.query() {
Some(query) => format!("{bulk_path}?{query}"),
None => bulk_path,
};
let mut bulk_uri_parts = control_uri.into_parts();
bulk_uri_parts.path_and_query = Some(bulk_path_and_query.parse().map_err(|error| {
ConnectionError::Other(format!(
"failed to build code-mode host bulk websocket path: {error}"
))
})?);
let bulk_url = Uri::from_parts(bulk_uri_parts)
.map_err(|error| {
ConnectionError::Other(format!(
"failed to build code-mode host bulk websocket URL: {error}"
))
})?
.to_string();
connect_websocket_transport(&bulk_url, &self.http_client_factory)
.await
.map_err(|error| ConnectionError::BulkConnectionUnavailable(error.to_string()))
}
}
impl CallerCancellation {
fn new() -> Self {
Self {
@@ -253,52 +193,19 @@ impl Connection {
.take()
.ok_or_else(|| ConnectionError::Other("spawned code-mode host has no stdout".into()))?;
Self::establish(
ConnectionReader::Stdio(FramedReader::new(stdout)),
ConnectionWriter::Stdio(FramedWriter::new(stdin)),
ConnectionOwner::Process(Box::new(child)),
/*bulk_connection_options*/ None,
)
.await
}
pub(super) async fn connect_websocket(
websocket_url: &str,
http_client_factory: &HttpClientFactory,
) -> Result<Self, ConnectionError> {
let (reader, writer) =
connect_websocket_transport(websocket_url, http_client_factory).await?;
Self::establish(
reader,
writer,
ConnectionOwner::WebSocket,
Some(BulkConnectionOptions {
websocket_url: websocket_url.to_string(),
http_client_factory: http_client_factory.clone(),
}),
)
.await
Self::establish(FramedReader::new(stdout), FramedWriter::new(stdin), child).await
}
async fn establish(
mut reader: ConnectionReader,
mut writer: ConnectionWriter,
mut owner: ConnectionOwner,
bulk_connection_options: Option<BulkConnectionOptions>,
mut reader: FramedReader<ChildStdout>,
mut writer: FramedWriter<ChildStdin>,
mut child: Child,
) -> Result<Self, ConnectionError> {
let handshake = async {
let dual_capability =
Capability::new(DUAL_WEBSOCKET_CAPABILITY).map_err(|error| error.to_string())?;
let session_limits_capability = Capability::new(SESSION_RESOURCE_LIMITS_CAPABILITY)
.map_err(|error| error.to_string())?;
let optional_capabilities = if bulk_connection_options.is_some() {
CapabilitySet::try_new([dual_capability.clone(), session_limits_capability])
.map_err(|error| error.to_string())?
} else {
CapabilitySet::try_new([session_limits_capability])
.map_err(|error| error.to_string())?
};
let optional_capabilities = CapabilitySet::try_new([session_limits_capability])
.map_err(|error| error.to_string())?;
let hello = ClientHello::new(
SupportedProtocolVersions::try_new([ProtocolVersion::V1])
.map_err(|err| err.to_string())?,
@@ -318,24 +225,7 @@ impl Connection {
Some(HostToClient::HostHello(hello))
if hello.selected_version() == ProtocolVersion::V1 =>
{
let capabilities = hello.capabilities().clone();
let bulk_token = if capabilities.contains(&dual_capability) {
hello
.bulk_connection_token()
.map(str::to_string)
.ok_or_else(|| {
"code-mode host advertised dual websockets without a pairing token"
.to_string()
})
.map(Some)?
} else if hello.bulk_connection_token().is_some() {
return Err(
"code-mode host returned an unexpected bulk pairing token".to_string()
);
} else {
None
};
Ok((capabilities, bulk_token))
Ok(hello.capabilities().clone())
}
Some(HostToClient::HandshakeRejected { reason }) => {
Err(format!("code-mode host rejected the handshake: {reason:?}"))
@@ -346,99 +236,42 @@ impl Connection {
None => Err("code-mode host exited during handshake".to_string()),
}
};
let handshake_timeout = match &owner {
ConnectionOwner::Process(_) => LOCAL_HOST_STARTUP_TIMEOUT,
ConnectionOwner::WebSocket => HOST_HANDSHAKE_TIMEOUT,
};
let handshake_result = match tokio::time::timeout(handshake_timeout, handshake).await {
Ok(result) => result,
Err(_) => {
let _ = writer.close().await;
owner.close().await;
return Err(ConnectionError::Other(
"timed out negotiating with the code-mode host".into(),
));
}
};
let (capabilities, bulk_token) = match handshake_result {
let handshake_result =
match tokio::time::timeout(LOCAL_HOST_STARTUP_TIMEOUT, handshake).await {
Ok(result) => result,
Err(_) => {
kill_and_reap(&mut child).await;
return Err(ConnectionError::Other(
"timed out negotiating with the code-mode host".into(),
));
}
};
let capabilities = match handshake_result {
Ok(negotiated) => negotiated,
Err(err) => {
let _ = writer.close().await;
owner.close().await;
kill_and_reap(&mut child).await;
return Err(ConnectionError::Other(err));
}
};
let (bulk_reader, bulk_writer) = if let Some(token) = bulk_token {
let Some(options) = bulk_connection_options else {
let _ = writer.close().await;
owner.close().await;
return Err(ConnectionError::Other(
"code-mode host negotiated an unsupported bulk websocket".to_string(),
));
};
match options.connect(&token).await {
Ok((reader, writer)) => (Some(reader), Some(writer)),
Err(error) => {
let _ = writer.close().await;
owner.close().await;
return Err(error);
}
}
} else {
(None, None)
};
let (command_tx, command_rx) = mpsc::channel(IPC_CHANNEL_CAPACITY);
let (event_tx, event_rx) = mpsc::channel(IPC_CHANNEL_CAPACITY);
let (outgoing_tx, outgoing_rx) = mpsc::channel::<EncodedFrame>(IPC_CHANNEL_CAPACITY);
let (bulk_tx, bulk_rx) = if bulk_writer.is_some() {
let (sender, receiver) = mpsc::channel::<EncodedFrame>(MAX_PENDING_DELEGATE_CALLS);
(Some(sender), Some(receiver))
} else {
(None, None)
};
let dual_websocket = bulk_writer.is_some();
let cancellation = CancellationToken::new();
let alive = Arc::new(AtomicBool::new(true));
let failure = Arc::new(std::sync::Mutex::new(None));
let writer_cancellation = cancellation.clone();
let writer_task = tokio::spawn(async move {
if let (Some(bulk_writer), Some(bulk_rx)) = (bulk_writer, bulk_rx) {
tokio::try_join!(
drive_writer(writer, outgoing_rx, writer_cancellation.clone()),
drive_writer(bulk_writer, bulk_rx, writer_cancellation)
)?;
Ok(())
} else {
drive_writer(writer, outgoing_rx, writer_cancellation).await
}
});
let writer_task =
tokio::spawn(
async move { drive_writer(writer, outgoing_rx, writer_cancellation).await },
);
let reader_events = event_tx.clone();
let reader_cancellation = cancellation.clone();
let reader_task = tokio::spawn(async move {
let lane = dual_websocket.then_some(TransportLane::Control);
if let Some(bulk_reader) = bulk_reader {
tokio::try_join!(
drive_reader(
reader,
reader_events.clone(),
reader_cancellation.clone(),
lane,
),
drive_reader(
bulk_reader,
reader_events,
reader_cancellation,
Some(TransportLane::Bulk),
)
)?;
Ok(())
} else {
drive_reader(reader, reader_events, reader_cancellation, lane).await
}
});
let reader_task =
tokio::spawn(
async move { drive_reader(reader, reader_events, reader_cancellation).await },
);
let (driver, execute_claim_tx) = ConnectionDriver::new(
command_rx,
@@ -451,14 +284,10 @@ impl Connection {
cancellation: cancellation.clone(),
},
);
let driver = match bulk_tx {
Some(sender) => driver.with_bulk_sender(sender),
None => driver,
};
let driver_task = tokio::spawn(driver.run());
tokio::spawn(
ConnectionSupervisor {
owner,
child,
event_tx,
cancellation: cancellation.clone(),
alive: Arc::clone(&alive),
@@ -653,65 +482,21 @@ impl Connection {
}
}
async fn connect_websocket_transport(
websocket_url: &str,
http_client_factory: &HttpClientFactory,
) -> Result<(ConnectionReader, ConnectionWriter), ConnectionError> {
let request = websocket_url.into_client_request().map_err(|error| {
ConnectionError::Other(format!(
"failed to build code-mode host websocket request: {error}"
))
})?;
let connector = WebSocketConnector::new(http_client_factory)
.map_err(|error| {
ConnectionError::Other(format!(
"failed to configure code-mode host websocket TLS: {error}"
))
})?
.with_tcp_nodelay();
let websocket_config = WebSocketConfig::default()
.max_frame_size(Some(MAX_WEBSOCKET_FRAME_BYTES))
.max_message_size(Some(MAX_WEBSOCKET_FRAME_BYTES));
let (websocket, _) = tokio::time::timeout(
HOST_HANDSHAKE_TIMEOUT,
connector.connect(request, websocket_config),
)
.await
.map_err(|_| {
ConnectionError::Other("timed out connecting to the code-mode host websocket".into())
})?
.map_err(|error| {
ConnectionError::Other(format!(
"failed to connect to the code-mode host websocket: {error}"
))
})?;
let (writer, reader) = websocket.split();
Ok((
ConnectionReader::WebSocket(reader),
ConnectionWriter::WebSocket(writer),
))
}
async fn drive_writer(
mut writer: ConnectionWriter,
mut writer: FramedWriter<ChildStdin>,
mut outgoing: mpsc::Receiver<EncodedFrame>,
cancellation: CancellationToken,
) -> Result<(), String> {
loop {
tokio::select! {
_ = cancellation.cancelled() => {
return writer
.close()
.await
.map_err(|error| format!("failed to close code-mode host connection: {error}"));
}
_ = cancellation.cancelled() => return Ok(()),
frame = outgoing.recv() => {
let Some(frame) = frame else {
return Err("code-mode host outgoing stream closed".to_string());
};
tokio::select! {
_ = cancellation.cancelled() => return Ok(()),
result = writer.write_frame(frame) => {
result = writer.write_frame(&frame) => {
result.map_err(|error| {
format!("failed to write code-mode host message: {error}")
})?;
@@ -735,7 +520,7 @@ impl Drop for Connection {
impl ConnectionSupervisor {
async fn run(mut self) {
let mut owner_exited = false;
let mut child_exited = false;
let reason = tokio::select! {
biased;
_ = self.cancellation.cancelled() => failure_message(&self.failure),
@@ -745,35 +530,19 @@ impl ConnectionSupervisor {
},
result = &mut self.reader_task => task_failure("reader", result),
result = &mut self.writer_task => task_failure("writer", result),
reason = self.owner.wait() => {
owner_exited = true;
reason
result = self.child.wait() => {
child_exited = true;
match result {
Ok(status) => format!("code-mode host exited with status {status}"),
Err(error) => format!("failed waiting for code-mode host: {error}"),
}
}
};
mark_connection_dead(&self.alive, &self.failure, reason.clone());
let _ = self.event_tx.try_send(DriverEvent::Failed(reason));
self.cancellation.cancel();
if !owner_exited {
self.owner.close().await;
}
}
}
impl ConnectionOwner {
async fn wait(&mut self) -> String {
match self {
Self::Process(child) => match child.wait().await {
Ok(status) => format!("code-mode host exited with status {status}"),
Err(error) => format!("failed waiting for code-mode host: {error}"),
},
Self::WebSocket => std::future::pending().await,
}
}
async fn close(&mut self) {
match self {
Self::Process(child) => kill_and_reap(child).await,
Self::WebSocket => {}
if !child_exited {
kill_and_reap(&mut self.child).await;
}
}
}
@@ -9,7 +9,6 @@ use codex_code_mode_protocol::CodeModeSessionDelegate;
use codex_code_mode_protocol::host::EncodedFrame;
use codex_code_mode_protocol::host::HostToClient;
use codex_code_mode_protocol::host::RequestId;
use codex_code_mode_protocol::host::TransportLane;
use tokio::sync::mpsc;
use tokio_util::sync::CancellationToken;
@@ -42,7 +41,6 @@ pub(super) struct ConnectionDriver {
event_tx: mpsc::Sender<DriverEvent>,
execute_claim_rx: mpsc::UnboundedReceiver<RequestId>,
outgoing_tx: mpsc::Sender<EncodedFrame>,
bulk_tx: Option<mpsc::Sender<EncodedFrame>>,
requests: RequestTracker,
deferred_host_messages: VecDeque<HostToClient>,
sessions: SessionRegistry,
@@ -69,7 +67,6 @@ impl ConnectionDriver {
event_tx: event_tx.clone(),
execute_claim_rx,
outgoing_tx,
bulk_tx: None,
requests: RequestTracker::new(),
deferred_host_messages: VecDeque::new(),
sessions: SessionRegistry::new(),
@@ -137,17 +134,8 @@ impl ConnectionDriver {
}
}
pub(super) fn with_bulk_sender(mut self, sender: mpsc::Sender<EncodedFrame>) -> Self {
self.bulk_tx = Some(sender);
self
}
fn queue_frame(&mut self, frame: EncodedFrame, lane: TransportLane) -> bool {
let sender = match lane {
TransportLane::Control => &self.outgoing_tx,
TransportLane::Bulk => self.bulk_tx.as_ref().unwrap_or(&self.outgoing_tx),
};
match sender.try_send(frame) {
fn queue_frame(&mut self, frame: EncodedFrame) -> bool {
match self.outgoing_tx.try_send(frame) {
Ok(()) => true,
Err(mpsc::error::TrySendError::Full(_)) => {
self.fail("code-mode host outgoing queue is full".to_string());
@@ -127,8 +127,7 @@ impl ConnectionDriver {
},
&self.event_tx,
);
let lane = message.transport_lane();
self.queue_frame(frame, lane)
self.queue_frame(frame)
}
fn execute(
@@ -187,8 +186,7 @@ impl ConnectionDriver {
},
&self.event_tx,
);
let lane = message.transport_lane();
self.queue_frame(frame, lane)
self.queue_frame(frame)
}
fn wait(
@@ -317,7 +315,6 @@ impl ConnectionDriver {
};
self.requests
.insert_pending(request_id, pending, &self.event_tx);
let lane = message.transport_lane();
self.queue_frame(frame, lane)
self.queue_frame(frame)
}
}
@@ -318,8 +318,7 @@ impl ConnectionDriver {
}
}
};
let lane = message.transport_lane();
self.queue_frame(frame, lane)
self.queue_frame(frame)
}
pub(super) fn close_cell(&mut self, session_id: SessionId, cell_id: WireCellId) -> bool {
@@ -374,8 +374,7 @@ impl ConnectionDriver {
return false;
}
};
let lane = message.transport_lane();
self.queue_frame(frame, lane)
self.queue_frame(frame)
}
fn shutdown_abandoned_session(&mut self, session: RemoteSession) -> bool {
@@ -17,7 +17,6 @@ use codex_code_mode_protocol::host::CapabilitySet;
use codex_code_mode_protocol::host::ClientToHost;
use codex_code_mode_protocol::host::DelegateRequest;
use codex_code_mode_protocol::host::DelegateRequestId;
use codex_code_mode_protocol::host::DelegateResponse;
use codex_code_mode_protocol::host::EncodedFrame;
use codex_code_mode_protocol::host::HostRequest;
use codex_code_mode_protocol::host::HostResponse;
@@ -248,11 +247,6 @@ struct RecordingDelegate {
struct PanickingDelegate;
struct LargeResultBurstDelegate {
started: AtomicUsize,
release: CancellationToken,
}
#[derive(Debug, Eq, PartialEq)]
enum HeldDelegateEvent {
Started,
@@ -342,35 +336,6 @@ impl CodeModeSessionDelegate for PanickingDelegate {
fn cell_closed(&self, _cell_id: &CellId) {}
}
impl CodeModeSessionDelegate for LargeResultBurstDelegate {
fn invoke_tool<'a>(
&'a self,
_invocation: CodeModeNestedToolCall,
cancellation_token: CancellationToken,
) -> ToolInvocationFuture<'a> {
self.started.fetch_add(1, Ordering::Release);
let release = self.release.clone();
Box::pin(async move {
tokio::select! {
_ = cancellation_token.cancelled() => Err("cancelled".to_string()),
_ = release.cancelled() => Ok("x".repeat(256 * 1024).into()),
}
})
}
fn notify<'a>(
&'a self,
_call_id: String,
_cell_id: CellId,
_text: String,
_cancellation_token: CancellationToken,
) -> NotificationFuture<'a> {
Box::pin(async { Ok(()) })
}
fn cell_closed(&self, _cell_id: &CellId) {}
}
impl CodeModeSessionDelegate for RecordingDelegate {
fn invoke_tool<'a>(
&'a self,
@@ -692,97 +657,6 @@ async fn delegate_cancel_is_best_effort_and_sends_no_late_response() {
assert_eq!(delegate.notifications.load(Ordering::Relaxed), 0);
}
#[tokio::test]
async fn concurrent_large_delegate_results_do_not_disconnect_a_backpressured_bulk_lane() {
const CONCURRENT_RESULTS: usize = 129;
let (command_tx, command_rx) = mpsc::channel(/*max_capacity*/ 16);
let (event_tx, event_rx) = mpsc::channel(/*max_capacity*/ 16);
let (outgoing_tx, outgoing_rx) = mpsc::channel(/*max_capacity*/ 16);
let (bulk_tx, mut bulk_rx) = mpsc::channel(MAX_PENDING_DELEGATE_CALLS);
let cancellation = CancellationToken::new();
let alive = Arc::new(AtomicBool::new(true));
let failure = Arc::new(StdMutex::new(None));
let (driver, execute_claim_tx) = ConnectionDriver::new(
command_rx,
event_rx,
event_tx.clone(),
outgoing_tx,
DriverLifecycle {
alive: Arc::clone(&alive),
failure: Arc::clone(&failure),
cancellation: cancellation.clone(),
},
);
let driver_task = tokio::spawn(driver.with_bulk_sender(bulk_tx).run());
let mut harness = DriverHarness {
command_tx,
event_tx,
execute_claim_tx,
outgoing_rx,
cancellation,
alive,
failure,
driver_task,
};
let session = remote_session();
let delegate = Arc::new(LargeResultBurstDelegate {
started: AtomicUsize::new(0),
release: CancellationToken::new(),
});
harness.open(session.clone(), delegate.clone()).await;
let _started = harness
.start_cell(session.clone(), /*request_id*/ 2, "1")
.await;
for value in 1..=CONCURRENT_RESULTS {
harness
.start_tool_delegate(&session, DelegateRequestId::new(value as i64))
.await;
}
tokio::time::timeout(Duration::from_secs(10), async {
while delegate.started.load(Ordering::Acquire) < CONCURRENT_RESULTS {
tokio::task::yield_now().await;
}
})
.await
.expect("concurrent delegate calls should all start");
delegate.release.cancel();
tokio::time::timeout(Duration::from_secs(10), async {
while bulk_rx.len() < CONCURRENT_RESULTS {
assert!(
harness.alive.load(Ordering::Acquire),
"bulk queue disconnected before accepting all concurrent tool results"
);
tokio::task::yield_now().await;
}
})
.await
.expect("concurrent large results should queue behind the blocked bulk writer");
let _unrelated = harness.start_cell(session, /*request_id*/ 3, "2").await;
assert!(harness.alive.load(Ordering::Acquire));
for _ in 0..CONCURRENT_RESULTS {
let frame = bulk_rx.recv().await.expect("queued bulk delegate result");
let message = EncodedFrame::decode_framed::<ClientToHost>(&frame.into_framed_bytes())
.expect("decode queued delegate result");
let ClientToHost::DelegateResponse {
result:
WireResult::Ok {
value: DelegateResponse::ToolResult { result },
},
..
} = message
else {
panic!("expected a successful large delegate result");
};
assert_eq!(result.as_str().map(str::len), Some(256 * 1024));
}
assert!(harness.alive.load(Ordering::Acquire));
}
#[tokio::test]
async fn delegate_limit_returns_an_error_without_disconnecting() {
let mut harness = DriverHarness::start();
@@ -1,15 +1,14 @@
use codex_code_mode_protocol::host::TransportLane;
use codex_code_mode_protocol::host::FramedReader;
use tokio::process::ChildStdout;
use tokio::sync::mpsc;
use tokio_util::sync::CancellationToken;
use super::driver::DriverEvent;
use super::transport::ConnectionReader;
pub(super) async fn drive_reader(
mut reader: ConnectionReader,
mut reader: FramedReader<ChildStdout>,
events: mpsc::Sender<DriverEvent>,
cancellation: CancellationToken,
lane: Option<TransportLane>,
) -> Result<(), String> {
loop {
let message = tokio::select! {
@@ -21,11 +20,6 @@ pub(super) async fn drive_reader(
Ok(None) => return Err("code-mode host closed its stdout".to_string()),
Err(err) => return Err(format!("failed to read code-mode host message: {err}")),
};
if let Some(lane) = lane
&& !message.allows_transport_lane(lane)
{
return Err("code-mode host sent a message on the wrong websocket lane".to_string());
}
events
.send(DriverEvent::HostMessage(message))
.await
@@ -1,103 +0,0 @@
use std::io;
use std::time::Duration;
use codex_code_mode_protocol::host::ClientToHost;
use codex_code_mode_protocol::host::EncodedFrame;
use codex_code_mode_protocol::host::FramedReader;
use codex_code_mode_protocol::host::FramedWriter;
use codex_code_mode_protocol::host::HostToClient;
use codex_websocket_client::WebSocketConnection;
use futures::SinkExt;
use futures::StreamExt;
use futures::stream::SplitSink;
use futures::stream::SplitStream;
use tokio::process::ChildStdin;
use tokio::process::ChildStdout;
use tokio_tungstenite::tungstenite::Message;
const WEBSOCKET_CLOSE_TIMEOUT: Duration = Duration::from_secs(5);
pub(super) enum ConnectionReader {
Stdio(FramedReader<ChildStdout>),
WebSocket(SplitStream<WebSocketConnection>),
}
pub(super) enum ConnectionWriter {
Stdio(FramedWriter<ChildStdin>),
WebSocket(SplitSink<WebSocketConnection, Message>),
}
impl ConnectionReader {
pub(super) async fn read(&mut self) -> io::Result<Option<HostToClient>> {
match self {
Self::Stdio(reader) => reader.read().await,
Self::WebSocket(reader) => loop {
match reader.next().await {
Some(Ok(Message::Binary(frame))) => {
return EncodedFrame::decode_framed(&frame).map(Some);
}
Some(Ok(Message::Ping(_) | Message::Pong(_))) => {}
Some(Ok(Message::Close(_))) | None => return Ok(None),
Some(Ok(Message::Text(_))) => {
return Err(io::Error::new(
io::ErrorKind::InvalidData,
"code-mode host websocket messages must be binary framed messages",
));
}
Some(Ok(Message::Frame(_))) => {
return Err(io::Error::new(
io::ErrorKind::InvalidData,
"code-mode host websocket returned an unexpected raw frame",
));
}
Some(Err(error)) => {
return Err(io::Error::other(format!(
"failed to read code-mode host websocket message: {error}"
)));
}
}
},
}
}
}
impl ConnectionWriter {
pub(super) async fn write(&mut self, message: &ClientToHost) -> io::Result<()> {
self.write_frame(EncodedFrame::encode(message)?).await
}
pub(super) async fn write_frame(&mut self, frame: EncodedFrame) -> io::Result<()> {
match self {
Self::Stdio(writer) => writer.write_frame(&frame).await,
Self::WebSocket(writer) => writer
.send(Message::Binary(frame.into_framed_bytes().into()))
.await
.map_err(|error| {
io::Error::other(format!(
"failed to write code-mode host websocket message: {error}"
))
}),
}
}
pub(super) async fn close(&mut self) -> io::Result<()> {
match self {
Self::Stdio(_) => Ok(()),
Self::WebSocket(writer) => {
tokio::time::timeout(WEBSOCKET_CLOSE_TIMEOUT, writer.close())
.await
.map_err(|_| {
io::Error::new(
io::ErrorKind::TimedOut,
"timed out closing code-mode host websocket connection",
)
})?
.map_err(|error| {
io::Error::other(format!(
"failed to close code-mode host websocket connection: {error}"
))
})
}
}
}
}
@@ -1,43 +1,12 @@
use std::io;
use std::path::PathBuf;
use std::sync::Arc;
use std::time::Duration;
use codex_code_mode_protocol::CodeModeSessionCellExecutionLimits;
use codex_code_mode_protocol::CodeModeSessionProvider;
use codex_code_mode_protocol::ExecuteRequest;
use codex_code_mode_protocol::FunctionCallOutputContentItem;
use codex_code_mode_protocol::RuntimeResponse;
use codex_code_mode_protocol::host::Capability;
use codex_code_mode_protocol::host::CapabilitySet;
use codex_code_mode_protocol::host::ClientToHost;
use codex_code_mode_protocol::host::DUAL_WEBSOCKET_CAPABILITY;
use codex_code_mode_protocol::host::EncodedFrame;
use codex_code_mode_protocol::host::HostHello;
use codex_code_mode_protocol::host::HostRequest;
use codex_code_mode_protocol::host::HostResponse;
use codex_code_mode_protocol::host::HostToClient;
use codex_code_mode_protocol::host::ProtocolVersion;
use codex_code_mode_protocol::host::SESSION_RESOURCE_LIMITS_CAPABILITY;
use codex_code_mode_protocol::host::WireCellId;
use codex_code_mode_protocol::host::WireContentItem;
use codex_code_mode_protocol::host::WireResult;
use codex_code_mode_protocol::host::WireRuntimeResponse;
use codex_http_client::HttpClientFactory;
use codex_http_client::OutboundProxyPolicy;
use futures::SinkExt;
use futures::StreamExt;
use pretty_assertions::assert_eq;
use tokio::io::AsyncReadExt;
use tokio::io::AsyncWriteExt;
use tokio::net::TcpListener;
use tokio::time::timeout;
use tokio_tungstenite::accept_async;
use tokio_tungstenite::tungstenite::Message;
use super::ProcessOwnedCodeModeSession;
use super::ProcessOwnedCodeModeSessionProvider;
use super::WebSocketCodeModeSessionProvider;
use super::connection::ConnectionError;
use crate::NoopCodeModeSessionDelegate;
@@ -86,274 +55,6 @@ fn missing_host_error_preserves_utf8_boundaries_when_truncating_the_path() {
assert!(displayed_path.len() <= 512);
}
#[tokio::test]
async fn websocket_provider_executes_over_shared_connector() {
let listener = TcpListener::bind("127.0.0.1:0")
.await
.expect("websocket test listener should bind");
let websocket_url = format!(
"ws://{}",
listener
.local_addr()
.expect("websocket test listener should have an address")
);
let server = tokio::spawn(async move {
let (stream, _) = listener
.accept()
.await
.expect("websocket test host should accept a connection");
let mut websocket = accept_async(stream)
.await
.expect("websocket test host should complete the HTTP handshake");
while let Some(message) = websocket.next().await {
let message = message.expect("websocket test host should receive a valid message");
let frame = match message {
Message::Binary(frame) => frame,
Message::Ping(_) | Message::Pong(_) => continue,
Message::Close(_) => break,
Message::Text(_) | Message::Frame(_) => {
panic!("websocket test host received an unexpected message: {message:?}");
}
};
let request = EncodedFrame::decode_framed::<ClientToHost>(&frame)
.expect("websocket test host should decode a framed protocol message");
let responses = match request {
ClientToHost::ClientHello(hello) => {
let capability = Capability::new(SESSION_RESOURCE_LIMITS_CAPABILITY)
.expect("session-limit capability");
assert!(hello.optional_capabilities().contains(&capability));
assert_eq!(hello.required_capabilities(), &CapabilitySet::empty());
vec![HostToClient::HostHello(HostHello::new(
ProtocolVersion::V1,
CapabilitySet::empty(),
))]
}
ClientToHost::Request {
id,
request: HostRequest::OpenSession { session_id, .. },
} => vec![HostToClient::Response {
id,
result: WireResult::Ok {
value: HostResponse::SessionReady { session_id },
},
}],
ClientToHost::Request {
id,
request: HostRequest::Execute { request, .. },
} => {
assert_eq!(request.source, "text('shared connector')");
let cell_id = WireCellId::new("1");
vec![
HostToClient::Response {
id,
result: WireResult::Ok {
value: HostResponse::ExecutionStarted {
cell_id: cell_id.clone(),
},
},
},
HostToClient::InitialResponse {
id,
result: WireResult::Ok {
value: WireRuntimeResponse::Result {
cell_id,
content_items: vec![WireContentItem::InputText {
text: "shared connector".to_string(),
}],
error_text: None,
},
},
},
]
}
ClientToHost::Request {
id,
request: HostRequest::ShutdownSession { session_id },
} => vec![HostToClient::Response {
id,
result: WireResult::Ok {
value: HostResponse::SessionClosed { session_id },
},
}],
request => {
panic!("websocket test host received an unexpected request: {request:?}")
}
};
for response in responses {
let frame = EncodedFrame::encode(&response)
.expect("websocket test host should encode a framed response");
websocket
.send(Message::Binary(frame.into_framed_bytes().into()))
.await
.expect("websocket test host should send its response");
}
}
});
let provider = WebSocketCodeModeSessionProvider::with_http_client_factory(
websocket_url,
HttpClientFactory::new(OutboundProxyPolicy::ReqwestDefault),
);
let session = provider
.create_session(Arc::new(NoopCodeModeSessionDelegate))
.await
.expect("shared websocket connector should open a code-mode session");
let error = provider
.create_session_with_limits(
Arc::new(NoopCodeModeSessionDelegate),
CodeModeSessionCellExecutionLimits {
max_yield_time_ms: Some(250),
max_heap_size_bytes: None,
},
)
.await
.err()
.expect("legacy host should reject a limited session");
assert_eq!(
error,
format!(
"code-mode host does not support session resource limits: missing `{SESSION_RESOURCE_LIMITS_CAPABILITY}` capability"
)
);
let second_session = provider
.create_session(Arc::new(NoopCodeModeSessionDelegate))
.await
.expect("rejecting limited sessions should preserve the shared legacy-host connection");
second_session
.shutdown()
.await
.expect("second unlimited session should shut down");
drop(second_session);
let response = session
.execute(ExecuteRequest {
tool_call_id: "shared-websocket".to_string(),
enabled_tools: Vec::new(),
source: "text('shared connector')".to_string(),
yield_time_ms: None,
max_output_tokens: None,
})
.await
.expect("shared websocket connector should start a cell")
.initial_response()
.await
.expect("shared websocket connector should return a cell result");
assert_eq!(
response,
RuntimeResponse::Result {
cell_id: codex_code_mode_protocol::CellId::new("1".to_string()),
content_items: vec![FunctionCallOutputContentItem::InputText {
text: "shared connector".to_string(),
}],
error_text: None,
}
);
session
.shutdown()
.await
.expect("shared websocket connector should shut down its session");
drop(session);
drop(provider);
timeout(Duration::from_secs(5), server)
.await
.expect("websocket test host should disconnect promptly")
.expect("websocket test host task should succeed");
}
#[tokio::test]
async fn websocket_provider_fails_when_a_negotiated_bulk_connection_is_unavailable() {
let listener = TcpListener::bind("127.0.0.1:0")
.await
.expect("websocket test listener should bind");
let websocket_url = format!(
"ws://{}/?access_token=shared-token",
listener
.local_addr()
.expect("websocket test listener should have an address")
);
let server = tokio::spawn(async move {
let (stream, _) = listener
.accept()
.await
.expect("websocket test host should accept the first connection");
let mut control = accept_async(stream)
.await
.expect("first control websocket should connect");
let frame = control
.next()
.await
.expect("first client hello")
.expect("first websocket frame")
.into_data();
let ClientToHost::ClientHello(hello) =
EncodedFrame::decode_framed(&frame).expect("decode first client hello")
else {
panic!("expected first client hello");
};
let capability =
Capability::new(DUAL_WEBSOCKET_CAPABILITY).expect("dual websocket capability");
assert!(hello.optional_capabilities().contains(&capability));
let session_limits_capability =
Capability::new(SESSION_RESOURCE_LIMITS_CAPABILITY).expect("session-limit capability");
assert!(
hello
.optional_capabilities()
.contains(&session_limits_capability)
);
let hello = HostToClient::HostHello(
HostHello::new(
ProtocolVersion::V1,
CapabilitySet::try_new([capability.clone()]).expect("host capabilities"),
)
.with_bulk_connection_token("fallback-token".to_string()),
);
let frame = EncodedFrame::encode(&hello).expect("encode dual host hello");
control
.send(Message::Binary(frame.into_framed_bytes().into()))
.await
.expect("send dual host hello");
let (mut bulk, _) = listener
.accept()
.await
.expect("websocket test host should accept the bulk connection");
let mut request = [0_u8; 1024];
let request_len = bulk
.read(&mut request)
.await
.expect("read bulk websocket handshake");
let request = std::str::from_utf8(&request[..request_len]).expect("bulk HTTP request");
assert!(
request.starts_with("GET /bulk/fallback-token?access_token=shared-token HTTP/1.1\r\n")
);
bulk.write_all(b"HTTP/1.1 404 Not Found\r\nContent-Length: 0\r\nConnection: close\r\n\r\n")
.await
.expect("reject unavailable bulk websocket");
drop(bulk);
drop(control);
});
let provider = WebSocketCodeModeSessionProvider::new(websocket_url);
let error = match provider
.create_session(Arc::new(NoopCodeModeSessionDelegate))
.await
{
Ok(_) => panic!("provider should reject an unavailable negotiated bulk websocket"),
Err(error) => error,
};
assert!(
error.contains("404"),
"unexpected negotiated bulk websocket error: {error}"
);
drop(provider);
timeout(Duration::from_secs(5), server)
.await
.expect("negotiated bulk websocket test host should disconnect promptly")
.expect("negotiated bulk websocket test host task should succeed");
}
#[tokio::test]
async fn provider_returns_missing_host_error() {
let provider = ProcessOwnedCodeModeSessionProvider::with_host_program(