File size: 11,583 Bytes
afa0cbf | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 | //! 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(())
}
|