Add optional process metrics sampling for code-mode hosts (#57841)

## What changed

- Add `GetProcessMetrics` and provider sampling APIs for process ID, start time, cumulative CPU time, and resident memory, exposed through `ThreadManager` without creating a thread or session.
- Distinguish unsupported collection, a locally absent host, and sampling failures. Process-owned providers never start or replace a host when sampling.
- Bound gRPC sampling to one second without retries or explicit connection resets. Run optional collectors on a detached worker with at most one collection in flight, so stalled collection cannot block the HTTP/2 driver or runtime shutdown.

## Testing

Add coverage for snapshots, collector errors and panics, deadlines, overload, and shutdown with a stuck collector. Extend lifecycle tests to check that sampling preserves session state and keepalive behavior without starting or restarting local hosts.

GitOrigin-RevId: eeafe275891d0464636f28a29345a87aed1a12dc
This commit is contained in:
aakash-oai
2026-10-28 21:34:25 +00:00
committed by copyberry
parent 7ce5fa7bdc
commit d373d64967
18 changed files with 559 additions and 15 deletions
+15 -1
View File
@@ -46,6 +46,7 @@ mod conversion;
mod deadline;
mod generation;
mod operations;
mod process_metrics;
mod reconnect;
mod state;
mod transport;
@@ -65,6 +66,11 @@ fn inject_span_traceparent<T>(request: &mut tonic::Request<T>, span: &tracing::S
}
/// Creates code-mode sessions over an HTTP/2 gRPC connection.
///
/// Metrics requests do not open sessions, retry RPCs, or explicitly reset the
/// connection. URL providers may initialize their client or establish a network
/// connection; caller-supplied channels and transports retain their reconnect
/// policy. Preventing child creation requires a process-owned transport.
#[derive(Clone)]
pub struct GrpcCodeModeSessionProvider {
transport: Arc<SharedTransport>,
@@ -94,7 +100,8 @@ impl GrpcCodeModeSessionProvider {
Self::from_transport(SharedTransport::with_channel(channel))
}
/// Uses a caller-owned transport to control connection establishment and recovery.
/// Uses a caller-owned transport, which can prevent observation RPCs from
/// starting or replacing process-owned connections.
pub fn with_transport<T>(transport: T) -> Self
where
T: tower::Service<
@@ -238,6 +245,13 @@ impl GrpcCodeModeSessionProvider {
}
impl CodeModeSessionProvider for GrpcCodeModeSessionProvider {
fn sample_process_metrics(
&self,
) -> CodeModeSessionResultFuture<'_, codex_code_mode_protocol::CodeModeProcessMetricsSample>
{
Box::pin(process_metrics::sample(self))
}
fn create_session(&self) -> CodeModeSessionProviderFuture<'_> {
self.create_session_with_limits(CodeModeSessionCellExecutionLimits::default())
}
@@ -0,0 +1,50 @@
//! A bounded, non-retrying host observation. No session binding or recovery is
//! involved: a failed RPC never explicitly retires the execution connection.
//! Connection establishment and reconnection remain the transport's policy.
use std::time::Duration;
use codex_code_mode_protocol::CodeModeProcessMetrics;
use codex_code_mode_protocol::CodeModeProcessMetricsSample;
use codex_code_mode_protocol::grpc;
use tonic::Code;
use super::GrpcCodeModeSessionProvider;
pub(super) async fn sample(
provider: &GrpcCodeModeSessionProvider,
) -> Result<CodeModeProcessMetricsSample, String> {
let deadline = Duration::from_secs(/*secs*/ 1);
tokio::time::timeout(deadline, async {
let mut client = provider.transport.client().await?;
let mut request = tonic::Request::new(grpc::GetProcessMetricsRequest {});
request.set_timeout(deadline);
match client.get_process_metrics(request).await {
Ok(response) => {
let snapshot = response.into_inner();
if snapshot.process_id == 0
|| !snapshot.process_start_time_seconds.is_finite()
|| snapshot.process_start_time_seconds <= 0.0
|| !snapshot.cpu_seconds_total.is_finite()
|| snapshot.cpu_seconds_total < 0.0
{
return Err("invalid code-mode process metrics snapshot".to_string());
}
Ok(CodeModeProcessMetricsSample::Snapshot(
CodeModeProcessMetrics {
process_id: snapshot.process_id,
process_start_time_seconds: snapshot.process_start_time_seconds,
cpu_seconds_total: snapshot.cpu_seconds_total,
resident_memory_bytes: snapshot.resident_memory_bytes,
},
))
}
Err(status) if status.code() == Code::Unimplemented => {
Ok(CodeModeProcessMetricsSample::Unsupported)
}
Err(status) => Err(format!("code-mode process metrics failed: {status}")),
}
})
.await
.map_err(|_| "code-mode process metrics timed out".to_string())?
}