mirror of
https://github.com/openai/codex.git
synced 2026-11-07 20:30:33 +00:00
Propagate trace context through gRPC code mode (#41017)
## What changed - Inject W3C `traceparent` metadata into code-mode session and execution requests. - Carry each execution's trace context on streamed tool callbacks so callback and nested-tool spans remain connected across the gRPC boundary. - Preserve the relevant spans when execution, callback, and dispatch work moves into spawned tasks. GitOrigin-RevId: 6482963cb9b2e2f0ea27f1345bba208b70a80d0b
This commit is contained in:
@@ -14,6 +14,7 @@ use codex_code_mode_protocol::CellId;
|
||||
use codex_code_mode_protocol::WaitRequest;
|
||||
use codex_code_mode_protocol::grpc as proto;
|
||||
use codex_code_mode_protocol::grpc::code_mode_host_server::CodeModeHost;
|
||||
use codex_protocol::protocol::W3cTraceContext;
|
||||
use futures::Stream;
|
||||
use futures::StreamExt;
|
||||
use tokio::sync::mpsc;
|
||||
@@ -21,6 +22,7 @@ use tokio_stream::wrappers::ReceiverStream;
|
||||
use tonic::Request;
|
||||
use tonic::Response;
|
||||
use tonic::Status;
|
||||
use tracing::Instrument;
|
||||
|
||||
use self::session::GrpcHostState;
|
||||
use self::session::GrpcSession;
|
||||
@@ -29,6 +31,17 @@ use self::waits::WaitRegistration;
|
||||
type GrpcStream<T> = Pin<Box<dyn Stream<Item = Result<T, Status>> + Send + 'static>>;
|
||||
type GrpcFuture<'a, T> = Pin<Box<dyn Future<Output = Result<Response<T>, Status>> + Send + 'a>>;
|
||||
|
||||
fn trace_context_from_request<T>(request: &Request<T>) -> Option<W3cTraceContext> {
|
||||
request
|
||||
.metadata()
|
||||
.get("traceparent")
|
||||
.and_then(|value| value.to_str().ok())
|
||||
.map(|traceparent| W3cTraceContext {
|
||||
traceparent: Some(traceparent.to_string()),
|
||||
tracestate: None,
|
||||
})
|
||||
}
|
||||
|
||||
/// Serves transport-independent, leased code-mode sessions over gRPC.
|
||||
#[derive(Clone)]
|
||||
pub struct GrpcCodeModeHost {
|
||||
@@ -43,12 +56,6 @@ impl GrpcCodeModeHost {
|
||||
}
|
||||
}
|
||||
|
||||
#[tracing::instrument(
|
||||
name = "code_mode_host.grpc.open_session",
|
||||
level = "info",
|
||||
skip_all,
|
||||
fields(otel.name = "code_mode_host.grpc.open_session")
|
||||
)]
|
||||
async fn open_session_request(
|
||||
&self,
|
||||
request: proto::OpenSessionRequest,
|
||||
@@ -119,20 +126,10 @@ impl GrpcCodeModeHost {
|
||||
Ok(Response::new(proto::AcknowledgeNotificationResponse {}))
|
||||
}
|
||||
|
||||
#[tracing::instrument(
|
||||
name = "code_mode_host.grpc.execute",
|
||||
level = "info",
|
||||
skip_all,
|
||||
fields(
|
||||
otel.name = "code_mode_host.grpc.execute",
|
||||
session.id = %request.session_id,
|
||||
execution.id = %request.execution_id,
|
||||
call_id = %request.tool_call_id,
|
||||
)
|
||||
)]
|
||||
async fn execute_request(
|
||||
&self,
|
||||
request: proto::ExecuteRequest,
|
||||
callback_traceparent: Option<String>,
|
||||
) -> Result<Response<GrpcStream<proto::ExecuteEvent>>, Status> {
|
||||
let session = self.state.session(&request.session_id)?;
|
||||
validation::identifier(&request.execution_id, "execution ID")?;
|
||||
@@ -154,7 +151,12 @@ impl GrpcCodeModeHost {
|
||||
}
|
||||
};
|
||||
let cell_id = started.cell_id.clone();
|
||||
session.admit_execution(execution_id.clone(), cell_id.to_string(), cell_permit)?;
|
||||
session.admit_execution(
|
||||
execution_id.clone(),
|
||||
cell_id.to_string(),
|
||||
cell_permit,
|
||||
callback_traceparent,
|
||||
)?;
|
||||
|
||||
let (sender, receiver) = mpsc::channel(/*buffer*/ 2);
|
||||
sender
|
||||
@@ -167,22 +169,26 @@ impl GrpcCodeModeHost {
|
||||
)),
|
||||
}))
|
||||
.map_err(|_| Status::internal("failed to publish code-mode execution admission"))?;
|
||||
tokio::spawn(async move {
|
||||
let _request_permit = request_permit;
|
||||
tokio::select! {
|
||||
biased;
|
||||
_ = sender.closed() => {}
|
||||
response = started.initial_response() => {
|
||||
let event = response.map(|response| proto::ExecuteEvent {
|
||||
event: Some(proto::execute_event::Event::Outcome(
|
||||
conversions::execution_outcome(response),
|
||||
)),
|
||||
}).map_err(Status::internal);
|
||||
let _ = sender.send(event).await;
|
||||
let outcome_span = tracing::Span::current();
|
||||
tokio::spawn(
|
||||
async move {
|
||||
let _request_permit = request_permit;
|
||||
tokio::select! {
|
||||
biased;
|
||||
_ = sender.closed() => {}
|
||||
response = started.initial_response() => {
|
||||
let event = response.map(|response| proto::ExecuteEvent {
|
||||
event: Some(proto::execute_event::Event::Outcome(
|
||||
conversions::execution_outcome(response),
|
||||
)),
|
||||
}).map_err(Status::internal);
|
||||
let _ = sender.send(event).await;
|
||||
}
|
||||
_ = session.closed.cancelled() => {}
|
||||
}
|
||||
_ = session.closed.cancelled() => {}
|
||||
}
|
||||
});
|
||||
.instrument(outcome_span),
|
||||
);
|
||||
|
||||
let stream = ReceiverStream::new(receiver).inspect(move |event| {
|
||||
if matches!(
|
||||
@@ -278,7 +284,16 @@ impl CodeModeHost for GrpcCodeModeHost {
|
||||
'a: 'async_trait,
|
||||
Self: 'async_trait,
|
||||
{
|
||||
Box::pin(self.open_session_request(request.into_inner()))
|
||||
let trace = trace_context_from_request(&request);
|
||||
let request = request.into_inner();
|
||||
let open_session_span = tracing::info_span!("code_mode_host.grpc.open_session");
|
||||
if let Some(trace) = trace.as_ref() {
|
||||
codex_otel::set_parent_from_w3c_trace_context(&open_session_span, trace);
|
||||
}
|
||||
Box::pin(
|
||||
self.open_session_request(request)
|
||||
.instrument(open_session_span),
|
||||
)
|
||||
}
|
||||
|
||||
fn close_session<'a, 'async_trait>(
|
||||
@@ -333,7 +348,24 @@ impl CodeModeHost for GrpcCodeModeHost {
|
||||
'a: 'async_trait,
|
||||
Self: 'async_trait,
|
||||
{
|
||||
Box::pin(self.execute_request(request.into_inner()))
|
||||
let trace = trace_context_from_request(&request);
|
||||
let request = request.into_inner();
|
||||
let execute_span = tracing::info_span!(
|
||||
"code_mode_host.grpc.execute",
|
||||
otel.name = "code_mode_host.grpc.execute",
|
||||
session.id = %request.session_id,
|
||||
execution.id = %request.execution_id,
|
||||
call_id = %request.tool_call_id,
|
||||
);
|
||||
if let Some(trace) = trace.as_ref() {
|
||||
codex_otel::set_parent_from_w3c_trace_context(&execute_span, trace);
|
||||
}
|
||||
let callback_traceparent =
|
||||
codex_otel::span_w3c_trace_context(&execute_span).and_then(|trace| trace.traceparent);
|
||||
Box::pin(
|
||||
self.execute_request(request, callback_traceparent)
|
||||
.instrument(execute_span),
|
||||
)
|
||||
}
|
||||
|
||||
fn wait<'a, 'async_trait>(
|
||||
|
||||
@@ -139,6 +139,7 @@ async fn dropping_execution_before_admission_releases_its_reservation() {
|
||||
execution_id,
|
||||
"cell".to_string(),
|
||||
host.state.cell_permit().expect("reserve cell permit"),
|
||||
/*traceparent*/ None,
|
||||
)
|
||||
.expect_err("abandoned execution must not admit a runtime cell");
|
||||
assert_eq!(error.code(), Code::Cancelled);
|
||||
|
||||
@@ -103,7 +103,7 @@ impl GrpcSession {
|
||||
let canonical_tool_name =
|
||||
ToolName::new(tool_name.namespace.clone(), tool_name.name.clone())
|
||||
.with_default_namespace();
|
||||
let (sequence, subscriptions) = {
|
||||
let (sequence, traceparent, subscriptions) = {
|
||||
let state = self.state.lock().unwrap_or_else(PoisonError::into_inner);
|
||||
let Some(execution) = state.cells.get(&cell_id) else {
|
||||
return Err("code-mode cell closed before dispatching its tool call".to_string());
|
||||
@@ -124,7 +124,7 @@ impl GrpcSession {
|
||||
})
|
||||
.map(|subscription| (subscription.id, subscription.sender.clone()))
|
||||
.collect::<Vec<_>>();
|
||||
(sequence, subscriptions)
|
||||
(sequence, execution.traceparent.clone(), subscriptions)
|
||||
};
|
||||
if subscriptions.is_empty() {
|
||||
return Err("no code-mode tool subscription matches the requested tool".to_string());
|
||||
@@ -139,6 +139,7 @@ impl GrpcSession {
|
||||
tool_kind: conversions::tool_kind(invocation.tool_kind),
|
||||
input_json,
|
||||
sequence,
|
||||
traceparent,
|
||||
};
|
||||
if message.encoded_len() > MAX_FRAME_BYTES {
|
||||
return Err(format!(
|
||||
|
||||
@@ -71,6 +71,7 @@ pub(super) struct SessionState {
|
||||
|
||||
pub(super) struct ExecutionState {
|
||||
pub(super) execution_id: String,
|
||||
pub(super) traceparent: Option<String>,
|
||||
pub(super) tool_call_sequence: u64,
|
||||
permit: OwnedSemaphorePermit,
|
||||
}
|
||||
@@ -296,6 +297,7 @@ impl GrpcSession {
|
||||
execution_id: String,
|
||||
cell_id: String,
|
||||
permit: OwnedSemaphorePermit,
|
||||
traceparent: Option<String>,
|
||||
) -> Result<(), Status> {
|
||||
let mut state = self.state.lock().unwrap_or_else(PoisonError::into_inner);
|
||||
if !state.pending_executions.remove(&execution_id) {
|
||||
@@ -308,6 +310,7 @@ impl GrpcSession {
|
||||
};
|
||||
entry.insert(ExecutionState {
|
||||
execution_id,
|
||||
traceparent,
|
||||
tool_call_sequence: 0,
|
||||
permit,
|
||||
});
|
||||
|
||||
Reference in New Issue
Block a user