| 1 | //! Real-process acceptance for `persist:true` background services on the |
| 2 | //! headless `codewhale exec` host. |
| 3 | //! |
| 4 | //! Three black-box contracts against the actual binary and real child |
| 5 | //! processes, with a `wiremock` OpenAI-compatible provider: |
| 6 | //! |
| 7 | //! - a successful exec releases the explicitly persisted service: the exec |
| 8 | //! process exits 0, emits a `service_released` receipt, and the service |
| 9 | //! process is still alive afterwards; |
| 10 | //! - a failed exec (incomplete non-limit stop) kills the pending service and |
| 11 | //! exits nonzero; |
| 12 | //! - a terminating signal mid-turn kills the pending service and exits |
| 13 | //! nonzero. |
| 14 | |
| 15 | #![cfg(unix)] |
| 16 | |
| 17 | use std::io::Read; |
| 18 | use std::path::Path; |
| 19 | use std::process::{Command, Stdio}; |
| 20 | use std::sync::atomic::{AtomicUsize, Ordering}; |
| 21 | use std::sync::{Arc, Mutex, OnceLock, mpsc}; |
| 22 | use std::time::{Duration, Instant}; |
| 23 | |
| 24 | async fn serialize_persistent_service_tests() -> tokio::sync::MutexGuard<'static, ()> { |
| 25 | static LOCK: OnceLock<tokio::sync::Mutex<()>> = OnceLock::new(); |
| 26 | LOCK.get_or_init(|| tokio::sync::Mutex::new(())) |
| 27 | .lock() |
| 28 | .await |
| 29 | } |
| 30 | |
| 31 | use serde_json::{Value, json}; |
| 32 | use tempfile::TempDir; |
| 33 | use wait_timeout::ChildExt; |
| 34 | use wiremock::matchers::{method, path}; |
| 35 | use wiremock::{Mock, MockServer, Request, Respond, ResponseTemplate}; |
| 36 | |
| 37 | const TEST_MODEL: &str = "persist-service-model"; |
| 38 | const RUN_TIMEOUT: Duration = Duration::from_secs(120); |
| 39 | |
| 40 | fn sse_chunk(value: Value) -> String { |
| 41 | format!( |
| 42 | "data: {}\n\n", |
| 43 | serde_json::to_string(&value).expect("SSE JSON") |
| 44 | ) |
| 45 | } |
| 46 | |
| 47 | /// First model turn: one Bash tool call staging the persistent service. |
| 48 | fn stage_service_sse(command: &str) -> String { |
| 49 | let arguments = serde_json::to_string(&json!({ |
| 50 | "command": command, |
| 51 | "background": true, |
| 52 | "persist": true, |
| 53 | })) |
| 54 | .expect("tool arguments JSON"); |
| 55 | [ |
| 56 | sse_chunk(json!({ |
| 57 | "id": "chatcmpl-stage", |
| 58 | "object": "chat.completion.chunk", |
| 59 | "model": TEST_MODEL, |
| 60 | "choices": [{"index": 0, "delta": {"tool_calls": [{"index": 0, "id": "call_persist", "type": "function", "function": {"name": "Bash", "arguments": arguments}}]}, "finish_reason": null}] |
| 61 | })), |
| 62 | sse_chunk(json!({ |
| 63 | "id": "chatcmpl-stage", |
| 64 | "object": "chat.completion.chunk", |
| 65 | "model": TEST_MODEL, |
| 66 | "choices": [{"index": 0, "delta": {}, "finish_reason": "tool_calls"}], |
| 67 | "usage": {"prompt_tokens": 12, "completion_tokens": 4, "total_tokens": 16} |
| 68 | })), |
| 69 | "data: [DONE]\n\n".to_string(), |
| 70 | ] |
| 71 | .join("") |
| 72 | } |
| 73 | |
| 74 | /// Second model turn: an ordinary completed final answer. |
| 75 | fn final_answer_sse() -> String { |
| 76 | [ |
| 77 | sse_chunk(json!({ |
| 78 | "id": "chatcmpl-final", |
| 79 | "object": "chat.completion.chunk", |
| 80 | "model": TEST_MODEL, |
| 81 | "choices": [{"index": 0, "delta": {"content": "service is up"}, "finish_reason": null}] |
| 82 | })), |
| 83 | sse_chunk(json!({ |
| 84 | "id": "chatcmpl-final", |
| 85 | "object": "chat.completion.chunk", |
| 86 | "model": TEST_MODEL, |
| 87 | "choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}], |
| 88 | "usage": {"prompt_tokens": 30, "completion_tokens": 3, "total_tokens": 33} |
| 89 | })), |
| 90 | "data: [DONE]\n\n".to_string(), |
| 91 | ] |
| 92 | .join("") |
| 93 | } |
| 94 | |
| 95 | /// Second model turn: incomplete non-limit stop. `length` now degrades and |
| 96 | /// continues the headless loop (no default max-turns), so this fixture uses |
| 97 | /// `content_filter` to force a failed exec and prove pending services die. |
| 98 | fn incomplete_answer_sse() -> String { |
| 99 | [ |
| 100 | sse_chunk(json!({ |
| 101 | "id": "chatcmpl-incomplete", |
| 102 | "object": "chat.completion.chunk", |
| 103 | "model": TEST_MODEL, |
| 104 | "choices": [{"index": 0, "delta": {"content": "partial"}, "finish_reason": null}] |
| 105 | })), |
| 106 | sse_chunk(json!({ |
| 107 | "id": "chatcmpl-incomplete", |
| 108 | "object": "chat.completion.chunk", |
| 109 | "model": TEST_MODEL, |
| 110 | "choices": [{"index": 0, "delta": {}, "finish_reason": "content_filter"}], |
| 111 | "usage": {"prompt_tokens": 30, "completion_tokens": 2, "total_tokens": 32} |
| 112 | })), |
| 113 | "data: [DONE]\n\n".to_string(), |
| 114 | ] |
| 115 | .join("") |
| 116 | } |
| 117 | |
| 118 | fn sse_response(body: String) -> ResponseTemplate { |
| 119 | ResponseTemplate::new(200) |
| 120 | .insert_header("content-type", "text/event-stream") |
| 121 | .insert_header("cache-control", "no-cache") |
| 122 | .set_body_string(body) |
| 123 | } |
| 124 | |
| 125 | fn json_response(value: Value) -> ResponseTemplate { |
| 126 | ResponseTemplate::new(200) |
| 127 | .insert_header("content-type", "application/json") |
| 128 | .set_body_json(value) |
| 129 | } |
| 130 | |
| 131 | /// Sequential provider: first POST stages the service; later POSTs get the |
| 132 | /// scenario's second turn. The failure scenario waits for explicit service |
| 133 | /// readiness; an optional delay holds the exec mid-turn for the signal case. |
| 134 | struct SequentialTurns { |
| 135 | requests: Arc<AtomicUsize>, |
| 136 | stage_command: String, |
| 137 | second_turn: String, |
| 138 | second_turn_delay: Option<Duration>, |
| 139 | second_turn_ready: Option<Mutex<mpsc::Receiver<()>>>, |
| 140 | } |
| 141 | |
| 142 | impl Respond for SequentialTurns { |
| 143 | fn respond(&self, _request: &Request) -> ResponseTemplate { |
| 144 | let call = self.requests.fetch_add(1, Ordering::SeqCst); |
| 145 | if call == 0 { |
| 146 | sse_response(stage_service_sse(&self.stage_command)) |
| 147 | } else { |
| 148 | if call == 1 |
| 149 | && let Some(ready) = &self.second_turn_ready |
| 150 | && ready |
| 151 | .lock() |
| 152 | .expect("service readiness receiver") |
| 153 | .recv_timeout(RUN_TIMEOUT) |
| 154 | .is_err() |
| 155 | { |
| 156 | return ResponseTemplate::new(504) |
| 157 | .set_body_string("test did not confirm service readiness"); |
| 158 | } |
| 159 | let response = sse_response(self.second_turn.clone()); |
| 160 | match self.second_turn_delay { |
| 161 | Some(delay) => response.set_delay(delay), |
| 162 | None => response, |
| 163 | } |
| 164 | } |
| 165 | } |
| 166 | } |
| 167 | |
| 168 | async fn start_mock_llm( |
| 169 | stage_command: &str, |
| 170 | second_turn: String, |
| 171 | second_turn_delay: Option<Duration>, |
| 172 | second_turn_ready: Option<mpsc::Receiver<()>>, |
| 173 | ) -> MockServer { |
| 174 | let server = MockServer::start().await; |
| 175 | |
| 176 | Mock::given(method("GET")) |
| 177 | .and(path("/v1/models")) |
| 178 | .respond_with(json_response(json!({ |
| 179 | "object": "list", |
| 180 | "data": [{ "id": TEST_MODEL, "object": "model" }] |
| 181 | }))) |
| 182 | .mount(&server) |
| 183 | .await; |
| 184 | |
| 185 | Mock::given(method("POST")) |
| 186 | .and(path("/v1/chat/completions")) |
| 187 | .respond_with(SequentialTurns { |
| 188 | requests: Arc::new(AtomicUsize::new(0)), |
| 189 | stage_command: stage_command.to_string(), |
| 190 | second_turn, |
| 191 | second_turn_delay, |
| 192 | second_turn_ready: second_turn_ready.map(Mutex::new), |
| 193 | }) |
| 194 | .mount(&server) |
| 195 | .await; |
| 196 | |
| 197 | server |
| 198 | } |
| 199 | |
| 200 | fn preserve_host_env(command: &mut Command) { |
| 201 | command.env_clear(); |
| 202 | for key in [ |
| 203 | "PATH", |
| 204 | "SHELL", |
| 205 | "TEMP", |
| 206 | "TMP", |
| 207 | "TERM", |
| 208 | "COLORTERM", |
| 209 | "LANG", |
| 210 | "LC_ALL", |
| 211 | ] { |
| 212 | if let Some(value) = std::env::var_os(key) { |
| 213 | command.env(key, value); |
| 214 | } |
| 215 | } |
| 216 | } |
| 217 | |
| 218 | fn exec_command(server: &MockServer, workspace: &Path, home: &Path) -> Command { |
| 219 | let mut command = Command::new(crate::binary::codewhale()); |
| 220 | preserve_host_env(&mut command); |
| 221 | command |
| 222 | .current_dir(workspace) |
| 223 | .arg("--workspace") |
| 224 | .arg(workspace) |
| 225 | .arg("--no-project-config") |
| 226 | .arg("exec") |
| 227 | .arg("--auto") |
| 228 | .arg("--sandbox") |
| 229 | .arg("danger-full-access") |
| 230 | .arg("--model") |
| 231 | .arg(TEST_MODEL) |
| 232 | .arg("--output-format") |
| 233 | .arg("stream-json") |
| 234 | .arg("start the service, then confirm") |
| 235 | .env("HOME", home) |
| 236 | .env("USERPROFILE", home) |
| 237 | .env("XDG_CONFIG_HOME", home.join(".config")) |
| 238 | .env("XDG_DATA_HOME", home.join(".local").join("share")) |
| 239 | .env("XDG_CACHE_HOME", home.join(".cache")) |
| 240 | .env( |
| 241 | "CODEWHALE_CONFIG_PATH", |
| 242 | home.join(".codewhale").join("config.toml"), |
| 243 | ) |
| 244 | .env( |
| 245 | "DEEPSEEK_CONFIG_PATH", |
| 246 | home.join(".deepseek").join("config.toml"), |
| 247 | ) |
| 248 | .env("DEEPSEEK_API_KEY", "ci-test-key-not-real") |
| 249 | .env("DEEPSEEK_BASE_URL", server.uri()) |
| 250 | .env("CODEWHALE_BASE_URL", server.uri()) |
| 251 | .env("DEEPSEEK_MODEL", TEST_MODEL) |
| 252 | .env("CODEWHALE_MODEL", TEST_MODEL) |
| 253 | .env("RUST_LOG", "warn") |
| 254 | .stdout(Stdio::piped()) |
| 255 | .stderr(Stdio::piped()); |
| 256 | std::fs::create_dir_all(home.join(".codewhale")).expect("create codewhale config dir"); |
| 257 | std::fs::create_dir_all(home.join(".deepseek")).expect("create deepseek config dir"); |
| 258 | command |
| 259 | } |
| 260 | |
| 261 | fn read_pipe_in_background<R>(mut reader: R) -> std::thread::JoinHandle<std::io::Result<Vec<u8>>> |
| 262 | where |
| 263 | R: Read + Send + 'static, |
| 264 | { |
| 265 | std::thread::spawn(move || { |
| 266 | let mut bytes = Vec::new(); |
| 267 | reader.read_to_end(&mut bytes)?; |
| 268 | Ok(bytes) |
| 269 | }) |
| 270 | } |
| 271 | |
| 272 | fn join_pipe(handle: std::thread::JoinHandle<std::io::Result<Vec<u8>>>, label: &str) -> String { |
| 273 | let bytes = handle |
| 274 | .join() |
| 275 | .unwrap_or_else(|_| panic!("{label} reader thread panicked")) |
| 276 | .unwrap_or_else(|error| panic!("{label} read failed: {error}")); |
| 277 | String::from_utf8_lossy(&bytes).into_owned() |
| 278 | } |
| 279 | |
| 280 | fn stream_events(stdout: &str) -> Vec<Value> { |
| 281 | stdout |
| 282 | .lines() |
| 283 | .filter_map(|line| serde_json::from_str::<Value>(line.trim()).ok()) |
| 284 | .collect() |
| 285 | } |
| 286 | |
| 287 | fn pid_is_alive(pid: i32) -> bool { |
| 288 | // SAFETY: signal 0 performs only an existence/permission check. |
| 289 | unsafe { libc::kill(pid, 0) == 0 } |
| 290 | } |
| 291 | |
| 292 | fn kill_process_group(pid: i32) { |
| 293 | // SAFETY: the pid came from this test's own staged service; the negative |
| 294 | // pid targets its process group only. |
| 295 | unsafe { |
| 296 | libc::kill(-pid, libc::SIGKILL); |
| 297 | } |
| 298 | } |
| 299 | |
| 300 | fn wait_for_pid_file(path: &Path) -> Result<i32, String> { |
| 301 | let deadline = Instant::now() + RUN_TIMEOUT; |
| 302 | loop { |
| 303 | if let Ok(contents) = std::fs::read_to_string(path) |
| 304 | && let Ok(pid) = contents.trim().parse::<i32>() |
| 305 | { |
| 306 | return Ok(pid); |
| 307 | } |
| 308 | if Instant::now() >= deadline { |
| 309 | return Err(format!( |
| 310 | "service pid file never appeared at {}", |
| 311 | path.display() |
| 312 | )); |
| 313 | } |
| 314 | std::thread::sleep(Duration::from_millis(50)); |
| 315 | } |
| 316 | } |
| 317 | |
| 318 | fn wait_for_pid_death(pid: i32) { |
| 319 | let deadline = Instant::now() + Duration::from_secs(15); |
| 320 | while pid_is_alive(pid) { |
| 321 | assert!( |
| 322 | Instant::now() < deadline, |
| 323 | "pending persistent service (pid {pid}) must be killed" |
| 324 | ); |
| 325 | std::thread::sleep(Duration::from_millis(50)); |
| 326 | } |
| 327 | } |
| 328 | |
| 329 | /// The staged service records its own pid, then stays alive. |
| 330 | const SERVICE_COMMAND: &str = "echo $$ > service.pid; exec sleep 600"; |
| 331 | |
| 332 | #[tokio::test(flavor = "multi_thread")] |
| 333 | async fn successful_exec_releases_persisted_service() { |
| 334 | let _serial = serialize_persistent_service_tests().await; |
| 335 | let server = start_mock_llm(SERVICE_COMMAND, final_answer_sse(), None, None).await; |
| 336 | let workspace = TempDir::new().expect("workspace tempdir"); |
| 337 | let home = TempDir::new().expect("home tempdir"); |
| 338 | |
| 339 | let mut child = exec_command(&server, workspace.path(), home.path()) |
| 340 | .spawn() |
| 341 | .expect("spawn codewhale exec"); |
| 342 | let stdout_reader = read_pipe_in_background(child.stdout.take().expect("stdout pipe")); |
| 343 | let stderr_reader = read_pipe_in_background(child.stderr.take().expect("stderr pipe")); |
| 344 | let status = child |
| 345 | .wait_timeout(RUN_TIMEOUT) |
| 346 | .expect("wait for exec") |
| 347 | .unwrap_or_else(|| { |
| 348 | let _ = child.kill(); |
| 349 | let _ = child.wait(); |
| 350 | panic!("exec timed out"); |
| 351 | }); |
| 352 | let stdout = join_pipe(stdout_reader, "stdout"); |
| 353 | let stderr = join_pipe(stderr_reader, "stderr"); |
| 354 | |
| 355 | let service_pid = wait_for_pid_file(&workspace.path().join("service.pid")) |
| 356 | .unwrap_or_else(|error| panic!("{error}\nstdout:\n{stdout}\nstderr:\n{stderr}")); |
| 357 | let events = stream_events(&stdout); |
| 358 | let released = events |
| 359 | .iter() |
| 360 | .find(|event| event.get("type").and_then(Value::as_str) == Some("service_released")) |
| 361 | .unwrap_or_else(|| { |
| 362 | panic!("missing service_released event\nstdout:\n{stdout}\nstderr:\n{stderr}") |
| 363 | }); |
| 364 | |
| 365 | assert!( |
| 366 | status.success(), |
| 367 | "successful exec must exit 0 (got {status:?})\nstderr:\n{stderr}" |
| 368 | ); |
| 369 | assert_eq!( |
| 370 | released.get("pid").and_then(Value::as_u64), |
| 371 | Some(u64::try_from(service_pid).expect("pid fits u64")), |
| 372 | "release receipt must carry the real service pid" |
| 373 | ); |
| 374 | assert_eq!( |
| 375 | released.get("ownership").and_then(Value::as_str), |
| 376 | Some("external") |
| 377 | ); |
| 378 | assert!( |
| 379 | pid_is_alive(service_pid), |
| 380 | "explicitly persisted service must survive successful headless exit" |
| 381 | ); |
| 382 | |
| 383 | kill_process_group(service_pid); |
| 384 | } |
| 385 | |
| 386 | #[tokio::test(flavor = "multi_thread")] |
| 387 | async fn failed_exec_kills_pending_service_and_exits_nonzero() { |
| 388 | let _serial = serialize_persistent_service_tests().await; |
| 389 | let (service_ready, service_ready_receiver) = mpsc::channel(); |
| 390 | let server = start_mock_llm( |
| 391 | SERVICE_COMMAND, |
| 392 | incomplete_answer_sse(), |
| 393 | None, |
| 394 | Some(service_ready_receiver), |
| 395 | ) |
| 396 | .await; |
| 397 | let workspace = TempDir::new().expect("workspace tempdir"); |
| 398 | let home = TempDir::new().expect("home tempdir"); |
| 399 | |
| 400 | let mut child = exec_command(&server, workspace.path(), home.path()) |
| 401 | .spawn() |
| 402 | .expect("spawn codewhale exec"); |
| 403 | let stdout_reader = read_pipe_in_background(child.stdout.take().expect("stdout pipe")); |
| 404 | let stderr_reader = read_pipe_in_background(child.stderr.take().expect("stderr pipe")); |
| 405 | |
| 406 | // Spawning a background process does not mean its first instruction ran. |
| 407 | // Hold the deliberate model failure until the real service is ready, so |
| 408 | // cancellation cannot kill it before it writes the PID we need to check. |
| 409 | let service_pid = match wait_for_pid_file(&workspace.path().join("service.pid")) { |
| 410 | Ok(pid) => pid, |
| 411 | Err(error) => { |
| 412 | drop(service_ready); |
| 413 | // Let the host clean up its managed services before forcing exit. |
| 414 | // SAFETY: direct child of this test. |
| 415 | unsafe { |
| 416 | libc::kill( |
| 417 | i32::try_from(child.id()).expect("child pid fits i32"), |
| 418 | libc::SIGTERM, |
| 419 | ); |
| 420 | } |
| 421 | if child |
| 422 | .wait_timeout(Duration::from_secs(5)) |
| 423 | .ok() |
| 424 | .flatten() |
| 425 | .is_none() |
| 426 | { |
| 427 | let _ = child.kill(); |
| 428 | let _ = child.wait(); |
| 429 | } |
| 430 | let stdout = join_pipe(stdout_reader, "stdout"); |
| 431 | let stderr = join_pipe(stderr_reader, "stderr"); |
| 432 | panic!("{error}\nstdout:\n{stdout}\nstderr:\n{stderr}"); |
| 433 | } |
| 434 | }; |
| 435 | assert!( |
| 436 | pid_is_alive(service_pid), |
| 437 | "service must be alive before the model fails" |
| 438 | ); |
| 439 | service_ready |
| 440 | .send(()) |
| 441 | .expect("release the deliberate model failure"); |
| 442 | |
| 443 | let status = match child.wait_timeout(RUN_TIMEOUT).expect("wait for exec") { |
| 444 | Some(status) => status, |
| 445 | None => { |
| 446 | let _ = child.kill(); |
| 447 | let _ = child.wait(); |
| 448 | let stdout = join_pipe(stdout_reader, "stdout"); |
| 449 | let stderr = join_pipe(stderr_reader, "stderr"); |
| 450 | panic!("exec timed out\nstdout:\n{stdout}\nstderr:\n{stderr}"); |
| 451 | } |
| 452 | }; |
| 453 | let stdout = join_pipe(stdout_reader, "stdout"); |
| 454 | let stderr = join_pipe(stderr_reader, "stderr"); |
| 455 | assert!( |
| 456 | !status.success(), |
| 457 | "provider incomplete stop must fail the exec\nstdout:\n{stdout}\nstderr:\n{stderr}" |
| 458 | ); |
| 459 | assert!( |
| 460 | !stream_events(&stdout) |
| 461 | .iter() |
| 462 | .any(|event| event.get("type").and_then(Value::as_str) == Some("service_released")), |
| 463 | "a failed exec must never release a pending service" |
| 464 | ); |
| 465 | wait_for_pid_death(service_pid); |
| 466 | } |
| 467 | |
| 468 | #[tokio::test(flavor = "multi_thread")] |
| 469 | async fn terminating_signal_kills_pending_service_and_exits_nonzero() { |
| 470 | let _serial = serialize_persistent_service_tests().await; |
| 471 | // Hold the second model turn open long past the signal. |
| 472 | let server = start_mock_llm( |
| 473 | SERVICE_COMMAND, |
| 474 | final_answer_sse(), |
| 475 | Some(Duration::from_secs(300)), |
| 476 | None, |
| 477 | ) |
| 478 | .await; |
| 479 | let workspace = TempDir::new().expect("workspace tempdir"); |
| 480 | let home = TempDir::new().expect("home tempdir"); |
| 481 | |
| 482 | let mut child = exec_command(&server, workspace.path(), home.path()) |
| 483 | .spawn() |
| 484 | .expect("spawn codewhale exec"); |
| 485 | let stdout_reader = read_pipe_in_background(child.stdout.take().expect("stdout pipe")); |
| 486 | let stderr_reader = read_pipe_in_background(child.stderr.take().expect("stderr pipe")); |
| 487 | |
| 488 | // The pid file proves the service was staged before the signal. |
| 489 | let service_pid = wait_for_pid_file(&workspace.path().join("service.pid")) |
| 490 | .unwrap_or_else(|error| panic!("{error}")); |
| 491 | assert!(pid_is_alive(service_pid)); |
| 492 | |
| 493 | // SAFETY: direct child of this test. |
| 494 | unsafe { |
| 495 | libc::kill( |
| 496 | i32::try_from(child.id()).expect("child pid fits i32"), |
| 497 | libc::SIGTERM, |
| 498 | ); |
| 499 | } |
| 500 | let status = child |
| 501 | .wait_timeout(Duration::from_secs(30)) |
| 502 | .expect("wait for signalled exec") |
| 503 | .unwrap_or_else(|| { |
| 504 | let _ = child.kill(); |
| 505 | let _ = child.wait(); |
| 506 | panic!("signalled exec did not exit"); |
| 507 | }); |
| 508 | let _ = join_pipe(stdout_reader, "stdout"); |
| 509 | let _ = join_pipe(stderr_reader, "stderr"); |
| 510 | |
| 511 | assert!( |
| 512 | !status.success(), |
| 513 | "a signalled exec must exit nonzero (got {status:?})" |
| 514 | ); |
| 515 | wait_for_pid_death(service_pid); |
| 516 | } |
| 517 |