Files
codex/codex-rs/code-mode-host/tests/grpc_timing.rs
Abhinav 3305c4f31d Scope code mode callback delegates to individual executions (#44865)
## Why

A delegate bound at session creation cannot provide different callbacks for cells sharing that session. Each execution needs to retain its own delegate across yields and release it when the cell is cleaned up.

## What changed

- Pass `CodeModeSessionDelegate` to `execute` instead of session creation.
- Route tool calls, notifications, and cell closure callbacks through the execution's delegate in the in-process runtime and the gRPC and stdio transports.
- Retain delegates with pending executions and live cells, releasing them through closure and cancellation cleanup.

## Testing

Add coverage for distinct delegates across yielded cells, gRPC callbacks before cell admission, and delegate release after completion or abandoned execution cleanup. Update transport tests to verify callbacks reach the owning cell's delegate.

GitOrigin-RevId: 7469f52104b993e790a40c65fb800ad02b7fd606
2026-09-11 17:00:07 +00:00

293 lines
11 KiB
Rust

//! Checks that remote timing measures each observation, not client delays or cell lifetime.
use std::sync::Arc;
use std::time::Duration;
use std::time::Instant;
use anyhow::Context;
use anyhow::Result;
use codex_code_mode::CodeModeSessionProvider;
use codex_code_mode::ExecuteRequest;
use codex_code_mode::GrpcCodeModeSessionProvider;
use codex_code_mode::NoopCodeModeSessionDelegate;
use codex_code_mode::ProcessOwnedCodeModeSessionProvider;
use codex_code_mode::RuntimeResponse;
use codex_code_mode::WaitOutcome;
use codex_code_mode::WaitRequest;
use codex_code_mode_protocol::grpc;
use codex_code_mode_protocol::grpc::code_mode_host_client::CodeModeHostClient;
use pretty_assertions::assert_eq;
use tokio::time::sleep;
use tokio::time::timeout;
#[path = "support/host.rs"]
mod host;
#[expect(dead_code, reason = "Other suites use the shared cell_id helper.")]
#[path = "support/recording_delegate.rs"]
mod recording_delegate;
use host::HostHarness;
use recording_delegate::RecordingDelegate;
const TEST_TIMEOUT: Duration = Duration::from_secs(/*secs*/ 20);
/// Successful and failed JavaScript include computation and timers in their host timing.
/// Waiting to read a finished execution must not increase that measurement.
#[tokio::test]
async fn execution_timing_includes_javascript_but_excludes_delayed_reads() -> Result<()> {
let host = HostHarness::start("grpc://127.0.0.1:0").await?;
let mut client = CodeModeHostClient::connect(host.endpoint).await?;
let mut events = client
.open_session(grpc::OpenSessionRequest {
cell_execution_limits: None,
})
.await?
.into_inner();
let opened = timeout(TEST_TIMEOUT, events.message())
.await??
.context("session ended")?;
let Some(grpc::session_event::Event::Opened(opened)) = opened.event else {
anyhow::bail!("expected session opening event");
};
let endings = ["", "throw new Error('timed failure');"];
for (index, suffix) in endings.into_iter().enumerate() {
let observed = Instant::now();
let mut execution = client
.execute(grpc::ExecuteRequest {
session_id: opened.session_id.clone(),
execution_id: format!("execution-{index}"),
tool_call_id: format!("call-{index}"),
source: format!(
"const until = Date.now() + 150; while (Date.now() < until) {{}} \
await new Promise(resolve => setTimeout(resolve, 150)); {suffix}"
),
enabled_tools: Vec::new(),
yield_time_ms: Some(/*value*/ 5_000),
max_output_tokens: Some(/*value*/ 1_000),
})
.await?
.into_inner();
let started = timeout(TEST_TIMEOUT, execution.message())
.await??
.context("execution ended")?;
let Some(grpc::execute_event::Event::Started(started)) = started.event else {
anyhow::bail!("expected execution admission");
};
let closed = timeout(TEST_TIMEOUT, events.message())
.await??
.context("session ended")?;
assert!(
matches!(closed.event, Some(grpc::session_event::Event::CellClosed(closed))
if closed.cell_id == started.cell_id)
);
// CellClosed confirms the JS finished before introducing a client-only delay.
sleep(Duration::from_millis(/*millis*/ 600)).await;
let result = timeout(TEST_TIMEOUT, execution.message())
.await??
.context("execution ended")?;
let Some(grpc::execute_event::Event::Outcome(outcome)) = result.event else {
anyhow::bail!("expected execution outcome");
};
let duration = Duration::from_nanos(outcome.code_mode_host_duration_ns);
assert!(duration >= Duration::from_millis(/*millis*/ 250));
assert!(duration + Duration::from_millis(/*millis*/ 300) <= observed.elapsed());
let Some(grpc::execution_outcome::Outcome::Completed(completed)) = outcome.outcome else {
anyhow::bail!("expected completed execution");
};
if suffix.is_empty() {
assert_eq!(completed.error_text, None);
} else {
assert!(
completed
.error_text
.context("expected JS failure")?
.contains("timed failure")
);
}
}
Ok(())
}
/// The process-owned host freezes successful and failed execution timing before
/// the client reads its initial response, including JavaScript computation and timers.
#[tokio::test]
async fn stdio_execution_timing_includes_javascript_but_excludes_delayed_reads() -> Result<()> {
let provider = ProcessOwnedCodeModeSessionProvider::with_host_program(
codex_utils_cargo_bin::cargo_bin("codex-code-mode-host")?,
);
let delegate = Arc::new(RecordingDelegate::default());
let session = provider
.create_session()
.await
.map_err(anyhow::Error::msg)?;
let endings = ["", "throw new Error('timed failure');"];
for (index, suffix) in endings.into_iter().enumerate() {
let observed = Instant::now();
let started = session
.execute(
ExecuteRequest {
tool_call_id: format!("call-{index}"),
source: format!(
"const until = Date.now() + 150; while (Date.now() < until) {{}} \
await new Promise(resolve => setTimeout(resolve, 150)); \
notify('finished'); {suffix}"
),
enabled_tools: Vec::new(),
yield_time_ms: Some(/*value*/ 5_000),
max_output_tokens: Some(/*value*/ 1_000),
},
delegate.clone(),
)
.await
.map_err(anyhow::Error::msg)?;
timeout(TEST_TIMEOUT, delegate.notification_delivered.notified())
.await
.context("JS did not report completion of its timed work")?;
// Delay only after the host has finished the computation and timer.
sleep(Duration::from_millis(/*millis*/ 600)).await;
let response = timeout(TEST_TIMEOUT, started.initial_response())
.await?
.map_err(anyhow::Error::msg)?;
let duration = response
.code_mode_host_duration()
.context("missing stdio exec timing")?;
assert!(duration >= Duration::from_millis(/*millis*/ 250));
assert!(duration + Duration::from_millis(/*millis*/ 300) <= observed.elapsed());
let RuntimeResponse::Result { error_text, .. } = response else {
anyhow::bail!("expected completed execution");
};
if suffix.is_empty() {
assert_eq!(error_text, None);
} else {
assert!(
error_text
.context("expected JS failure")?
.contains("timed failure")
);
}
}
session.shutdown().await.map_err(anyhow::Error::msg)?;
Ok(())
}
/// Every wait and termination measures its own request, including missing-cell outcomes.
/// Earlier execution and background time must not be charged again by later observations.
#[tokio::test]
async fn observation_timing_excludes_previous_requests_and_background_time() -> Result<()> {
let host = HostHarness::start("grpc://127.0.0.1:0").await?;
let providers: [Arc<dyn CodeModeSessionProvider>; 2] = [
Arc::new(ProcessOwnedCodeModeSessionProvider::with_host_program(
codex_utils_cargo_bin::cargo_bin("codex-code-mode-host")?,
)),
Arc::new(GrpcCodeModeSessionProvider::new(host.endpoint)),
];
for provider in providers {
let session = provider
.create_session()
.await
.map_err(anyhow::Error::msg)?;
let started = session
.execute(
ExecuteRequest {
tool_call_id: "call-1".to_string(),
source: "await new Promise(() => {});".to_string(),
enabled_tools: Vec::new(),
yield_time_ms: Some(/*value*/ 200),
max_output_tokens: Some(/*value*/ 1_000),
},
Arc::new(NoopCodeModeSessionDelegate),
)
.await
.map_err(anyhow::Error::msg)?;
let cell_id = started.cell_id.clone();
let initial = started
.initial_response()
.await
.map_err(anyhow::Error::msg)?;
let duration = initial
.code_mode_host_duration()
.context("missing exec timing")?;
assert!(duration >= Duration::from_millis(/*millis*/ 150));
let yielded = RuntimeResponse::Yielded {
cell_id: cell_id.clone(),
content_items: Vec::new(),
code_mode_host_duration: Some(duration),
};
assert_eq!(initial, yielded);
for _ in 0..2 {
sleep(Duration::from_millis(/*millis*/ 300)).await;
let observed = Instant::now();
let outcome = session
.wait(WaitRequest {
cell_id: cell_id.clone(),
yield_time_ms: 50,
})
.await
.map_err(anyhow::Error::msg)?;
let duration = outcome
.code_mode_host_duration()
.context("missing wait timing")?;
assert!(duration >= Duration::from_millis(/*millis*/ 40));
assert!(
duration <= observed.elapsed(),
"wait included time before its request: {duration:?}"
);
assert_eq!(
outcome,
WaitOutcome::LiveCell(RuntimeResponse::Yielded {
cell_id: cell_id.clone(),
content_items: Vec::new(),
code_mode_host_duration: Some(duration),
})
);
}
let observed = Instant::now();
let terminated = session
.terminate(cell_id.clone())
.await
.map_err(anyhow::Error::msg)?;
let duration = terminated
.code_mode_host_duration()
.context("missing termination timing")?;
assert!(duration <= observed.elapsed());
assert_eq!(
terminated,
WaitOutcome::LiveCell(RuntimeResponse::Terminated {
cell_id: cell_id.clone(),
content_items: Vec::new(),
code_mode_host_duration: Some(duration),
})
);
let observed = Instant::now();
let missing = session
.wait(WaitRequest {
cell_id: cell_id.clone(),
yield_time_ms: 50,
})
.await
.map_err(anyhow::Error::msg)?;
let duration = missing
.code_mode_host_duration()
.context("missing absent-cell timing")?;
assert!(duration <= observed.elapsed());
assert_eq!(
missing,
WaitOutcome::MissingCell(RuntimeResponse::Result {
error_text: Some(format!("exec cell {cell_id} not found")),
cell_id,
content_items: Vec::new(),
code_mode_host_duration: Some(duration),
})
);
session.shutdown().await.map_err(anyhow::Error::msg)?;
}
Ok(())
}