| 1 | //! Runtime backends: tmux durability, inline, vm/ci stubs (#4176). |
| 2 | //! |
| 3 | //! Runtime owns process/session lifecycle and stream-json log capture. |
| 4 | //! Fleet modules must not import this module. |
| 5 | |
| 6 | use std::fs::{self, OpenOptions}; |
| 7 | use std::io::{BufRead, BufReader, Write}; |
| 8 | use std::path::{Path, PathBuf}; |
| 9 | use std::process::{Command, ExitStatus, Stdio}; |
| 10 | use std::thread; |
| 11 | |
| 12 | use anyhow::{Context, Result, bail}; |
| 13 | use serde::{Deserialize, Serialize}; |
| 14 | |
| 15 | use crate::registry::{LaneRecord, LaneRegistry, LaneStatus, TerminalTransition}; |
| 16 | use crate::worktree::{WorktreeProvision, provision_worktree, remove_worktree_if_expired}; |
| 17 | |
| 18 | /// Execution backend for a lane. |
| 19 | #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)] |
| 20 | #[serde(rename_all = "snake_case")] |
| 21 | pub enum RuntimeBackendKind { |
| 22 | Tmux, |
| 23 | Inline, |
| 24 | Vm, |
| 25 | Ci, |
| 26 | } |
| 27 | |
| 28 | impl RuntimeBackendKind { |
| 29 | pub fn as_str(self) -> &'static str { |
| 30 | match self { |
| 31 | Self::Tmux => "tmux", |
| 32 | Self::Inline => "inline", |
| 33 | Self::Vm => "vm", |
| 34 | Self::Ci => "ci", |
| 35 | } |
| 36 | } |
| 37 | |
| 38 | pub fn parse(raw: &str) -> Result<Self> { |
| 39 | match raw.trim().to_ascii_lowercase().as_str() { |
| 40 | "tmux" => Ok(Self::Tmux), |
| 41 | "inline" => Ok(Self::Inline), |
| 42 | "vm" => Ok(Self::Vm), |
| 43 | "ci" => Ok(Self::Ci), |
| 44 | other => bail!("unknown runtime backend `{other}` (use tmux|inline|vm|ci)"), |
| 45 | } |
| 46 | } |
| 47 | } |
| 48 | |
| 49 | /// Inputs for starting a lane under a runtime backend. |
| 50 | #[derive(Clone)] |
| 51 | pub struct LaneStartSpec { |
| 52 | /// Command argv to run inside the backend (e.g. `codewhale exec …`). |
| 53 | pub command: Vec<String>, |
| 54 | /// Working directory for the command (defaults to worktree or cwd). |
| 55 | pub cwd: Option<PathBuf>, |
| 56 | /// Process-local runtime overrides. Values are never written into the |
| 57 | /// Lane record or command argv; tmux bridges them through a private 0600 |
| 58 | /// environment file that the detached shell removes before execution. |
| 59 | pub environment: Vec<(String, String)>, |
| 60 | /// Executable that exposes Codewhale's hidden `lane-log-proxy` command. |
| 61 | /// Required by tmux so arbitrary/binary child output is framed as valid |
| 62 | /// NDJSON without trusting a shell pipeline. |
| 63 | pub log_proxy: Option<PathBuf>, |
| 64 | /// When set, provision an isolated git worktree + branch under this repo. |
| 65 | pub worktree: Option<WorktreeProvision>, |
| 66 | } |
| 67 | |
| 68 | impl std::fmt::Debug for LaneStartSpec { |
| 69 | fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { |
| 70 | f.debug_struct("LaneStartSpec") |
| 71 | .field("command", &self.command) |
| 72 | .field("cwd", &self.cwd) |
| 73 | .field( |
| 74 | "environment_keys", |
| 75 | &self |
| 76 | .environment |
| 77 | .iter() |
| 78 | .map(|(key, _)| key) |
| 79 | .collect::<Vec<_>>(), |
| 80 | ) |
| 81 | .field("log_proxy", &self.log_proxy) |
| 82 | .field("worktree", &self.worktree) |
| 83 | .finish() |
| 84 | } |
| 85 | } |
| 86 | |
| 87 | /// Runtime adapter contract. |
| 88 | pub trait RuntimeBackend { |
| 89 | fn kind(&self) -> RuntimeBackendKind; |
| 90 | |
| 91 | /// Start the lane process/session; mutates record with attach/log metadata. |
| 92 | fn start( |
| 93 | &self, |
| 94 | registry: &LaneRegistry, |
| 95 | record: &mut LaneRecord, |
| 96 | spec: &LaneStartSpec, |
| 97 | ) -> Result<()>; |
| 98 | |
| 99 | /// Human attach command, if any (tmux). |
| 100 | fn attach_command(&self, record: &LaneRecord) -> Option<String>; |
| 101 | |
| 102 | /// Stop the running session/process. |
| 103 | /// |
| 104 | /// `fence` pins the durable lifecycle generation the caller observed. It is |
| 105 | /// evaluated under the registry's per-Lane lock, so a record that moved on |
| 106 | /// between the caller's read and this call is refused rather than stopped. |
| 107 | /// The return value distinguishes "this call stopped it" from "it was |
| 108 | /// already terminal" and "the fence did not match", so a caller can report |
| 109 | /// a truthful outcome instead of claiming a transition someone else made. |
| 110 | fn stop( |
| 111 | &self, |
| 112 | registry: &LaneRegistry, |
| 113 | record: &mut LaneRecord, |
| 114 | fence: Option<u64>, |
| 115 | ) -> Result<TerminalTransition>; |
| 116 | |
| 117 | /// Reconcile durable backend state into the Lane record before display. |
| 118 | /// Most backends update synchronously and need no refresh; tmux records |
| 119 | /// the detached process exit in a private sidecar and folds it in on the |
| 120 | /// next read. |
| 121 | /// |
| 122 | /// Returns `true` when reconciliation *changed durable state*. Reads are |
| 123 | /// allowed to fold a finished Runtime exit into the record, but a surface |
| 124 | /// that calls this must be able to say so rather than reporting a pure |
| 125 | /// observation (#4022). |
| 126 | fn reconcile(&self, _registry: &LaneRegistry, _record: &mut LaneRecord) -> Result<bool> { |
| 127 | Ok(false) |
| 128 | } |
| 129 | |
| 130 | /// Optional worktree TTL cleanup after stop. |
| 131 | fn cleanup_worktree(&self, record: &LaneRecord) -> Result<()> { |
| 132 | if let Some(path) = record.worktree_path.as_ref() { |
| 133 | remove_worktree_if_expired( |
| 134 | path, |
| 135 | record.worktree_ttl_secs, |
| 136 | record.stopped_at.as_deref(), |
| 137 | )?; |
| 138 | } |
| 139 | Ok(()) |
| 140 | } |
| 141 | } |
| 142 | |
| 143 | pub fn resolve_backend(kind: RuntimeBackendKind) -> Box<dyn RuntimeBackend> { |
| 144 | match kind { |
| 145 | RuntimeBackendKind::Tmux => Box::new(TmuxRuntime), |
| 146 | RuntimeBackendKind::Inline => Box::new(InlineRuntime), |
| 147 | RuntimeBackendKind::Vm => Box::new(StubRuntime { |
| 148 | kind: RuntimeBackendKind::Vm, |
| 149 | }), |
| 150 | RuntimeBackendKind::Ci => Box::new(StubRuntime { |
| 151 | kind: RuntimeBackendKind::Ci, |
| 152 | }), |
| 153 | } |
| 154 | } |
| 155 | |
| 156 | pub fn backend_for(record: &LaneRecord) -> Box<dyn RuntimeBackend> { |
| 157 | resolve_backend(record.runtime) |
| 158 | } |
| 159 | |
| 160 | fn append_log_event(log_path: &Path, event: serde_json::Value) -> Result<()> { |
| 161 | let mut file = OpenOptions::new() |
| 162 | .create(true) |
| 163 | .append(true) |
| 164 | .open(log_path) |
| 165 | .with_context(|| format!("open lane log {}", log_path.display()))?; |
| 166 | let mut encoded = serde_json::to_vec(&event).context("serialize lane log event")?; |
| 167 | encoded.push(b'\n'); |
| 168 | file.write_all(&encoded) |
| 169 | .with_context(|| format!("write lane log {}", log_path.display()))?; |
| 170 | Ok(()) |
| 171 | } |
| 172 | |
| 173 | const MAX_EXIT_RECEIPT_BYTES: u64 = 4 * 1024; |
| 174 | const MAX_ENVIRONMENT_BYTES: u64 = 1024 * 1024; |
| 175 | const MAX_CHILD_LOG_FRAME_BYTES: usize = 64 * 1024; |
| 176 | const LANE_PROXY_FAILURE_EXIT_CODE: i32 = 125; |
| 177 | |
| 178 | #[derive(Debug, Deserialize, Serialize)] |
| 179 | struct LaneExitReceipt { |
| 180 | lane_id: String, |
| 181 | exit_code: i32, |
| 182 | } |
| 183 | |
| 184 | /// Inputs to the hidden Rust log proxy used by detached runtimes. |
| 185 | /// |
| 186 | /// The proxy is deliberately exposed by the Lane crate so the thin CLI |
| 187 | /// facade can invoke it before loading user configuration. Secrets travel in |
| 188 | /// the private environment file, never in argv or the Lane record. |
| 189 | #[derive(Debug, Clone)] |
| 190 | pub struct LaneLogProxySpec { |
| 191 | pub command: Vec<String>, |
| 192 | pub log_path: PathBuf, |
| 193 | pub receipt_path: PathBuf, |
| 194 | pub receipt_tmp_path: PathBuf, |
| 195 | pub environment_path: Option<PathBuf>, |
| 196 | pub lane_id: String, |
| 197 | } |
| 198 | |
| 199 | fn lane_exit_receipt_path(log_path: &Path) -> PathBuf { |
| 200 | log_path.with_extension("exit.json") |
| 201 | } |
| 202 | |
| 203 | fn lane_exit_receipt_tmp_path(log_path: &Path) -> PathBuf { |
| 204 | log_path.with_extension("exit.json.tmp") |
| 205 | } |
| 206 | |
| 207 | fn lane_environment_path(log_path: &Path) -> PathBuf { |
| 208 | log_path.with_extension("env.json") |
| 209 | } |
| 210 | |
| 211 | fn lane_environment_tmp_path(path: &Path) -> PathBuf { |
| 212 | path.with_extension("json.tmp") |
| 213 | } |
| 214 | |
| 215 | fn remove_file_if_present(path: &Path) -> Result<()> { |
| 216 | match std::fs::remove_file(path) { |
| 217 | Ok(()) => Ok(()), |
| 218 | Err(err) if err.kind() == std::io::ErrorKind::NotFound => Ok(()), |
| 219 | Err(err) => Err(err).with_context(|| format!("remove {}", path.display())), |
| 220 | } |
| 221 | } |
| 222 | |
| 223 | fn valid_environment_key(key: &str) -> bool { |
| 224 | let mut chars = key.chars(); |
| 225 | chars |
| 226 | .next() |
| 227 | .is_some_and(|ch| ch == '_' || ch.is_ascii_alphabetic()) |
| 228 | && chars.all(|ch| ch == '_' || ch.is_ascii_alphanumeric()) |
| 229 | } |
| 230 | |
| 231 | fn write_lane_environment(path: &Path, environment: &[(String, String)]) -> Result<()> { |
| 232 | for (key, _) in environment { |
| 233 | if !valid_environment_key(key) { |
| 234 | bail!("invalid lane environment key {key:?}"); |
| 235 | } |
| 236 | } |
| 237 | let encoded = serde_json::to_vec(environment).context("serialize private lane environment")?; |
| 238 | if encoded.len() as u64 > MAX_ENVIRONMENT_BYTES { |
| 239 | bail!( |
| 240 | "private lane environment exceeds {} bytes", |
| 241 | MAX_ENVIRONMENT_BYTES |
| 242 | ); |
| 243 | } |
| 244 | |
| 245 | let tmp_path = lane_environment_tmp_path(path); |
| 246 | remove_file_if_present(path)?; |
| 247 | remove_file_if_present(&tmp_path)?; |
| 248 | let mut options = OpenOptions::new(); |
| 249 | options.create_new(true).write(true); |
| 250 | #[cfg(unix)] |
| 251 | { |
| 252 | use std::os::unix::fs::OpenOptionsExt; |
| 253 | options.mode(0o600); |
| 254 | } |
| 255 | let result = (|| { |
| 256 | let mut file = options |
| 257 | .open(&tmp_path) |
| 258 | .with_context(|| format!("create private lane environment {}", tmp_path.display()))?; |
| 259 | file.write_all(&encoded) |
| 260 | .with_context(|| format!("write private lane environment {}", tmp_path.display()))?; |
| 261 | file.sync_all() |
| 262 | .with_context(|| format!("sync private lane environment {}", tmp_path.display()))?; |
| 263 | fs::rename(&tmp_path, path) |
| 264 | .with_context(|| format!("publish private lane environment {}", path.display()))?; |
| 265 | Ok(()) |
| 266 | })(); |
| 267 | if result.is_err() { |
| 268 | let _ = remove_file_if_present(&tmp_path); |
| 269 | let _ = remove_file_if_present(path); |
| 270 | } |
| 271 | result |
| 272 | } |
| 273 | |
| 274 | fn append_child_output(log_path: &Path, stream: &str, bytes: &[u8]) -> Result<()> { |
| 275 | let mut line = bytes; |
| 276 | if let Some(stripped) = line.strip_suffix(b"\n") { |
| 277 | line = stripped; |
| 278 | } |
| 279 | if let Some(stripped) = line.strip_suffix(b"\r") { |
| 280 | line = stripped; |
| 281 | } |
| 282 | if line.is_empty() { |
| 283 | return Ok(()); |
| 284 | } |
| 285 | if let Ok(event) = serde_json::from_slice::<serde_json::Value>(line) { |
| 286 | append_log_event(log_path, event) |
| 287 | } else { |
| 288 | append_log_event( |
| 289 | log_path, |
| 290 | serde_json::json!({ |
| 291 | "type": "lane_log", |
| 292 | "stream": stream, |
| 293 | "message": String::from_utf8_lossy(line), |
| 294 | }), |
| 295 | ) |
| 296 | } |
| 297 | } |
| 298 | |
| 299 | fn stream_child_output( |
| 300 | reader: impl std::io::Read, |
| 301 | log_path: PathBuf, |
| 302 | stream: &'static str, |
| 303 | ) -> Result<()> { |
| 304 | let mut reader = BufReader::new(reader); |
| 305 | let mut line = Vec::new(); |
| 306 | loop { |
| 307 | line.clear(); |
| 308 | let mut reached_eof = false; |
| 309 | while line.len() < MAX_CHILD_LOG_FRAME_BYTES { |
| 310 | let buffer = reader |
| 311 | .fill_buf() |
| 312 | .with_context(|| format!("read lane child {stream}"))?; |
| 313 | if buffer.is_empty() { |
| 314 | reached_eof = true; |
| 315 | break; |
| 316 | } |
| 317 | let available = buffer |
| 318 | .iter() |
| 319 | .position(|byte| *byte == b'\n') |
| 320 | .map_or(buffer.len(), |position| position + 1); |
| 321 | let take = available.min(MAX_CHILD_LOG_FRAME_BYTES - line.len()); |
| 322 | let ended_line = buffer.get(take.saturating_sub(1)) == Some(&b'\n'); |
| 323 | line.extend_from_slice(&buffer[..take]); |
| 324 | reader.consume(take); |
| 325 | if ended_line { |
| 326 | break; |
| 327 | } |
| 328 | } |
| 329 | if line.is_empty() && reached_eof { |
| 330 | return Ok(()); |
| 331 | } |
| 332 | append_child_output(&log_path, stream, &line)?; |
| 333 | if reached_eof { |
| 334 | return Ok(()); |
| 335 | } |
| 336 | } |
| 337 | } |
| 338 | |
| 339 | fn read_lane_environment(path: &Path) -> Result<Vec<(String, String)>> { |
| 340 | let metadata = fs::metadata(path).with_context(|| format!("stat {}", path.display()))?; |
| 341 | if metadata.len() > MAX_ENVIRONMENT_BYTES { |
| 342 | bail!( |
| 343 | "private lane environment {} exceeds {} bytes", |
| 344 | path.display(), |
| 345 | MAX_ENVIRONMENT_BYTES |
| 346 | ); |
| 347 | } |
| 348 | let bytes = fs::read(path).with_context(|| format!("read {}", path.display()))?; |
| 349 | if bytes.len() as u64 > MAX_ENVIRONMENT_BYTES { |
| 350 | bail!( |
| 351 | "private lane environment {} exceeds {} bytes", |
| 352 | path.display(), |
| 353 | MAX_ENVIRONMENT_BYTES |
| 354 | ); |
| 355 | } |
| 356 | let environment: Vec<(String, String)> = |
| 357 | serde_json::from_slice(&bytes).with_context(|| format!("parse {}", path.display()))?; |
| 358 | for (key, _) in &environment { |
| 359 | if !valid_environment_key(key) { |
| 360 | bail!("invalid lane environment key {key:?}"); |
| 361 | } |
| 362 | } |
| 363 | Ok(environment) |
| 364 | } |
| 365 | |
| 366 | fn write_lane_exit_receipt( |
| 367 | receipt_path: &Path, |
| 368 | receipt_tmp_path: &Path, |
| 369 | lane_id: &str, |
| 370 | exit_code: i32, |
| 371 | ) -> Result<()> { |
| 372 | let encoded = serde_json::to_vec(&LaneExitReceipt { |
| 373 | lane_id: lane_id.to_string(), |
| 374 | exit_code, |
| 375 | }) |
| 376 | .context("serialize lane exit receipt")?; |
| 377 | if encoded.len() as u64 > MAX_EXIT_RECEIPT_BYTES { |
| 378 | bail!("serialized lane exit receipt exceeds size bound"); |
| 379 | } |
| 380 | remove_file_if_present(receipt_tmp_path)?; |
| 381 | let mut options = OpenOptions::new(); |
| 382 | options.create_new(true).write(true); |
| 383 | #[cfg(unix)] |
| 384 | { |
| 385 | use std::os::unix::fs::OpenOptionsExt; |
| 386 | options.mode(0o600); |
| 387 | } |
| 388 | let result = (|| { |
| 389 | let mut file = options |
| 390 | .open(receipt_tmp_path) |
| 391 | .with_context(|| format!("create {}", receipt_tmp_path.display()))?; |
| 392 | file.write_all(&encoded) |
| 393 | .with_context(|| format!("write {}", receipt_tmp_path.display()))?; |
| 394 | file.sync_all() |
| 395 | .with_context(|| format!("sync {}", receipt_tmp_path.display()))?; |
| 396 | fs::rename(receipt_tmp_path, receipt_path) |
| 397 | .with_context(|| format!("publish {}", receipt_path.display()))?; |
| 398 | Ok(()) |
| 399 | })(); |
| 400 | if result.is_err() { |
| 401 | let _ = remove_file_if_present(receipt_tmp_path); |
| 402 | } |
| 403 | result |
| 404 | } |
| 405 | |
| 406 | fn exit_status_code(status: ExitStatus) -> i32 { |
| 407 | if let Some(code) = status.code() { |
| 408 | return code; |
| 409 | } |
| 410 | #[cfg(unix)] |
| 411 | { |
| 412 | use std::os::unix::process::ExitStatusExt; |
| 413 | if let Some(signal) = status.signal() { |
| 414 | return 128 + signal; |
| 415 | } |
| 416 | } |
| 417 | LANE_PROXY_FAILURE_EXIT_CODE |
| 418 | } |
| 419 | |
| 420 | fn append_proxy_failure(log_path: &Path, lane_id: &str, error: &anyhow::Error) -> Result<()> { |
| 421 | append_log_event( |
| 422 | log_path, |
| 423 | serde_json::json!({ |
| 424 | "type": "lane_proxy_error", |
| 425 | "lane_id": lane_id, |
| 426 | "error": format!("{error:#}"), |
| 427 | }), |
| 428 | ) |
| 429 | } |
| 430 | |
| 431 | /// Run a child command while framing all output as NDJSON and atomically |
| 432 | /// publishing a private exit receipt. Returns the process-style exit code the |
| 433 | /// hidden CLI should propagate to tmux. |
| 434 | pub fn run_lane_log_proxy(spec: LaneLogProxySpec) -> Result<i32> { |
| 435 | let LaneLogProxySpec { |
| 436 | command, |
| 437 | log_path, |
| 438 | receipt_path, |
| 439 | receipt_tmp_path, |
| 440 | environment_path, |
| 441 | lane_id, |
| 442 | } = spec; |
| 443 | |
| 444 | let fail = |error: anyhow::Error, exit_code: i32| -> Result<i32> { |
| 445 | append_proxy_failure(&log_path, &lane_id, &error)?; |
| 446 | write_lane_exit_receipt(&receipt_path, &receipt_tmp_path, &lane_id, exit_code)?; |
| 447 | Ok(exit_code) |
| 448 | }; |
| 449 | |
| 450 | if command.is_empty() { |
| 451 | return fail(anyhow::anyhow!("lane log proxy requires a command"), 127); |
| 452 | } |
| 453 | |
| 454 | let environment = if let Some(path) = environment_path.as_deref() { |
| 455 | let loaded = read_lane_environment(path); |
| 456 | let removed = remove_file_if_present(path); |
| 457 | match (loaded, removed) { |
| 458 | (Ok(environment), Ok(())) => environment, |
| 459 | (Err(error), Ok(())) => return fail(error, LANE_PROXY_FAILURE_EXIT_CODE), |
| 460 | (Ok(_), Err(error)) | (Err(_), Err(error)) => return Err(error), |
| 461 | } |
| 462 | } else { |
| 463 | Vec::new() |
| 464 | }; |
| 465 | |
| 466 | let mut child_command = Command::new(&command[0]); |
| 467 | child_command.args(&command[1..]); |
| 468 | child_command.envs(environment); |
| 469 | child_command.stdout(Stdio::piped()).stderr(Stdio::piped()); |
| 470 | let mut child = match child_command.spawn() { |
| 471 | Ok(child) => child, |
| 472 | Err(error) => { |
| 473 | return fail( |
| 474 | anyhow::Error::from(error).context(format!("spawn lane child command {command:?}")), |
| 475 | 127, |
| 476 | ); |
| 477 | } |
| 478 | }; |
| 479 | let stdout = match child.stdout.take() { |
| 480 | Some(stdout) => stdout, |
| 481 | None => return fail(anyhow::anyhow!("lane child stdout was not piped"), 125), |
| 482 | }; |
| 483 | let stderr = match child.stderr.take() { |
| 484 | Some(stderr) => stderr, |
| 485 | None => return fail(anyhow::anyhow!("lane child stderr was not piped"), 125), |
| 486 | }; |
| 487 | let stdout_log = log_path.clone(); |
| 488 | let stderr_log = log_path.clone(); |
| 489 | let stdout_thread = thread::spawn(move || stream_child_output(stdout, stdout_log, "stdout")); |
| 490 | let stderr_thread = thread::spawn(move || stream_child_output(stderr, stderr_log, "stderr")); |
| 491 | |
| 492 | let status = match child.wait() { |
| 493 | Ok(status) => status, |
| 494 | Err(error) => { |
| 495 | return fail( |
| 496 | anyhow::Error::from(error).context(format!("wait for lane child {command:?}")), |
| 497 | LANE_PROXY_FAILURE_EXIT_CODE, |
| 498 | ); |
| 499 | } |
| 500 | }; |
| 501 | let stdout_result = stdout_thread |
| 502 | .join() |
| 503 | .map_err(|_| anyhow::anyhow!("lane proxy stdout logger panicked"))?; |
| 504 | let stderr_result = stderr_thread |
| 505 | .join() |
| 506 | .map_err(|_| anyhow::anyhow!("lane proxy stderr logger panicked"))?; |
| 507 | let mut exit_code = exit_status_code(status); |
| 508 | if let Some(error) = stdout_result.err().or_else(|| stderr_result.err()) { |
| 509 | append_proxy_failure(&log_path, &lane_id, &error)?; |
| 510 | exit_code = LANE_PROXY_FAILURE_EXIT_CODE; |
| 511 | } |
| 512 | write_lane_exit_receipt(&receipt_path, &receipt_tmp_path, &lane_id, exit_code)?; |
| 513 | Ok(exit_code) |
| 514 | } |
| 515 | |
| 516 | fn read_lane_exit_receipt(log_path: &Path, lane_id: &str) -> Result<Option<LaneExitReceipt>> { |
| 517 | let path = lane_exit_receipt_path(log_path); |
| 518 | let metadata = match std::fs::metadata(&path) { |
| 519 | Ok(metadata) => metadata, |
| 520 | Err(err) if err.kind() == std::io::ErrorKind::NotFound => return Ok(None), |
| 521 | Err(err) => return Err(err).with_context(|| format!("stat {}", path.display())), |
| 522 | }; |
| 523 | if metadata.len() > MAX_EXIT_RECEIPT_BYTES { |
| 524 | bail!( |
| 525 | "lane exit receipt {} exceeds {} bytes", |
| 526 | path.display(), |
| 527 | MAX_EXIT_RECEIPT_BYTES |
| 528 | ); |
| 529 | } |
| 530 | let bytes = std::fs::read(&path).with_context(|| format!("read {}", path.display()))?; |
| 531 | let receipt: LaneExitReceipt = |
| 532 | serde_json::from_slice(&bytes).with_context(|| format!("parse {}", path.display()))?; |
| 533 | if receipt.lane_id != lane_id { |
| 534 | bail!( |
| 535 | "lane exit receipt {} belongs to {}, expected {}", |
| 536 | path.display(), |
| 537 | receipt.lane_id, |
| 538 | lane_id |
| 539 | ); |
| 540 | } |
| 541 | Ok(Some(receipt)) |
| 542 | } |
| 543 | |
| 544 | #[derive(Debug, Clone, Copy, PartialEq, Eq)] |
| 545 | enum TmuxSessionState { |
| 546 | Present, |
| 547 | Absent, |
| 548 | } |
| 549 | |
| 550 | fn tmux_command(socket: &Path) -> Command { |
| 551 | let mut command = Command::new("tmux"); |
| 552 | command.arg("-S").arg(socket); |
| 553 | command |
| 554 | } |
| 555 | |
| 556 | fn ensure_tmux_available() -> Result<()> { |
| 557 | let output = Command::new("tmux") |
| 558 | .arg("-V") |
| 559 | .output() |
| 560 | .context("tmux runtime requires the `tmux` executable")?; |
| 561 | if !output.status.success() { |
| 562 | bail!( |
| 563 | "tmux runtime is unavailable: `tmux -V` failed with {}: {}", |
| 564 | output.status, |
| 565 | String::from_utf8_lossy(&output.stderr).trim() |
| 566 | ); |
| 567 | } |
| 568 | Ok(()) |
| 569 | } |
| 570 | |
| 571 | fn tmux_session_state(socket: &Path, session: &str) -> Result<TmuxSessionState> { |
| 572 | let output = tmux_command(socket) |
| 573 | .args(["has-session", "-t", session]) |
| 574 | .stdout(Stdio::null()) |
| 575 | .stderr(Stdio::piped()) |
| 576 | .output() |
| 577 | .with_context(|| format!("query tmux session {session}"))?; |
| 578 | if output.status.success() { |
| 579 | return Ok(TmuxSessionState::Present); |
| 580 | } |
| 581 | let stderr = String::from_utf8_lossy(&output.stderr).to_ascii_lowercase(); |
| 582 | if stderr.contains("can't find session:") |
| 583 | || stderr.contains("no server running on") |
| 584 | || (stderr.contains("error connecting to") && stderr.contains("no such file or directory")) |
| 585 | { |
| 586 | return Ok(TmuxSessionState::Absent); |
| 587 | } |
| 588 | bail!( |
| 589 | "tmux has-session for {session} failed with {}: {}", |
| 590 | output.status, |
| 591 | stderr.trim() |
| 592 | ) |
| 593 | } |
| 594 | |
| 595 | fn stop_tmux_session(socket: &Path, session: &str) -> Result<()> { |
| 596 | let status = tmux_command(socket) |
| 597 | .args(["kill-session", "-t", session]) |
| 598 | .stdout(Stdio::null()) |
| 599 | .stderr(Stdio::null()) |
| 600 | .status() |
| 601 | .with_context(|| format!("kill tmux session {session}"))?; |
| 602 | match tmux_session_state(socket, session).with_context(|| { |
| 603 | format!( |
| 604 | "confirm tmux session {session} on {} stopped after kill-session ({status})", |
| 605 | socket.display() |
| 606 | ) |
| 607 | })? { |
| 608 | TmuxSessionState::Absent => {} |
| 609 | TmuxSessionState::Present => { |
| 610 | bail!("tmux session {session} remains active after kill-session ({status})") |
| 611 | } |
| 612 | } |
| 613 | // A nonzero kill can be benign when the process and session exited in the |
| 614 | // same instant. The explicit absence check above is the source of truth. |
| 615 | Ok(()) |
| 616 | } |
| 617 | |
| 618 | fn tmux_log_proxy_command( |
| 619 | proxy: &Path, |
| 620 | command: &[String], |
| 621 | log_path: &Path, |
| 622 | receipt_path: &Path, |
| 623 | receipt_tmp_path: &Path, |
| 624 | environment_path: Option<&Path>, |
| 625 | lane_id: &str, |
| 626 | ) -> String { |
| 627 | let mut argv = vec![ |
| 628 | proxy.display().to_string(), |
| 629 | "lane-log-proxy".to_string(), |
| 630 | "--log-path".to_string(), |
| 631 | log_path.display().to_string(), |
| 632 | "--receipt-path".to_string(), |
| 633 | receipt_path.display().to_string(), |
| 634 | "--receipt-tmp-path".to_string(), |
| 635 | receipt_tmp_path.display().to_string(), |
| 636 | "--lane-id".to_string(), |
| 637 | lane_id.to_string(), |
| 638 | ]; |
| 639 | if let Some(path) = environment_path { |
| 640 | argv.push("--environment-path".to_string()); |
| 641 | argv.push(path.display().to_string()); |
| 642 | } |
| 643 | argv.push("--".to_string()); |
| 644 | argv.extend(command.iter().cloned()); |
| 645 | format!("exec {}", shell_join(&argv)) |
| 646 | } |
| 647 | |
| 648 | fn apply_worktree(record: &mut LaneRecord, spec: &LaneStartSpec) -> Result<Option<PathBuf>> { |
| 649 | let Some(wt) = spec.worktree.as_ref() else { |
| 650 | return Ok(spec.cwd.clone()); |
| 651 | }; |
| 652 | let provisioned = provision_worktree(wt)?; |
| 653 | record.worktree_path = Some(provisioned.path.clone()); |
| 654 | record.branch = Some(provisioned.branch.clone()); |
| 655 | Ok(Some(provisioned.path)) |
| 656 | } |
| 657 | |
| 658 | /// Durable local tmux sessions + attach + stream-json log file. |
| 659 | #[derive(Debug, Default)] |
| 660 | pub struct TmuxRuntime; |
| 661 | |
| 662 | impl RuntimeBackend for TmuxRuntime { |
| 663 | fn kind(&self) -> RuntimeBackendKind { |
| 664 | RuntimeBackendKind::Tmux |
| 665 | } |
| 666 | |
| 667 | fn start( |
| 668 | &self, |
| 669 | registry: &LaneRegistry, |
| 670 | record: &mut LaneRecord, |
| 671 | spec: &LaneStartSpec, |
| 672 | ) -> Result<()> { |
| 673 | if spec.command.is_empty() { |
| 674 | bail!("tmux runtime requires a non-empty command"); |
| 675 | } |
| 676 | // Dry-run is an explicit test hook only. A missing/broken tmux binary |
| 677 | // must fail closed rather than persisting a fictional Running Lane. |
| 678 | let dry_run = std::env::var_os("CODEWHALE_LANE_TMUX_DRY_RUN").is_some(); |
| 679 | if !dry_run && let Err(error) = ensure_tmux_available() { |
| 680 | append_log_event( |
| 681 | &record.log_path, |
| 682 | serde_json::json!({ |
| 683 | "type": "lane_failed", |
| 684 | "lane_id": record.id, |
| 685 | "runtime": "tmux", |
| 686 | "error": error.to_string(), |
| 687 | }), |
| 688 | )?; |
| 689 | let _ = registry.mark_terminal_if_active(record, LaneStatus::Failed)?; |
| 690 | return Err(error); |
| 691 | } |
| 692 | |
| 693 | let cwd = apply_worktree(record, spec)?; |
| 694 | let session = format!("cw-{}", record.id); |
| 695 | let socket = registry.root().join("tmux.sock"); |
| 696 | record.tmux_session = Some(session.clone()); |
| 697 | record.tmux_socket = Some(socket.clone()); |
| 698 | record.attach_target = Some(format!( |
| 699 | "tmux -S {} attach -t {}", |
| 700 | shell_escape(&socket.display().to_string()), |
| 701 | shell_escape(&session) |
| 702 | )); |
| 703 | |
| 704 | append_log_event( |
| 705 | &record.log_path, |
| 706 | serde_json::json!({ |
| 707 | "type": "lane_started", |
| 708 | "lane_id": record.id, |
| 709 | "runtime": "tmux", |
| 710 | "session": session, |
| 711 | "workflow": record.workflow, |
| 712 | "fleet": record.fleet, |
| 713 | "issue": record.issue, |
| 714 | "dry_run": dry_run, |
| 715 | }), |
| 716 | )?; |
| 717 | |
| 718 | if dry_run { |
| 719 | append_log_event( |
| 720 | &record.log_path, |
| 721 | serde_json::json!({ |
| 722 | "type": "lane_log", |
| 723 | "message": "tmux dry-run: session recorded without spawning process", |
| 724 | "command": spec.command, |
| 725 | "cwd": cwd.as_ref().map(|p| p.display().to_string()), |
| 726 | }), |
| 727 | )?; |
| 728 | if !registry.mark_running_if_pending(record)? { |
| 729 | bail!( |
| 730 | "lane `{}` was stopped before tmux dry-run start completed", |
| 731 | record.id |
| 732 | ); |
| 733 | } |
| 734 | return Ok(()); |
| 735 | } |
| 736 | |
| 737 | let log_proxy = spec |
| 738 | .log_proxy |
| 739 | .as_deref() |
| 740 | .context("tmux runtime requires a lane log proxy executable")?; |
| 741 | |
| 742 | // Detached session: child output remains an operator journal only. |
| 743 | // Terminal control state goes through a separate bounded, atomically |
| 744 | // renamed receipt so child stdout cannot forge Lane completion. |
| 745 | let receipt_path = lane_exit_receipt_path(&record.log_path); |
| 746 | let receipt_tmp_path = lane_exit_receipt_tmp_path(&record.log_path); |
| 747 | let environment_path = lane_environment_path(&record.log_path); |
| 748 | remove_file_if_present(&receipt_path)?; |
| 749 | remove_file_if_present(&receipt_tmp_path)?; |
| 750 | remove_file_if_present(&environment_path)?; |
| 751 | let environment_path = if spec.environment.is_empty() { |
| 752 | None |
| 753 | } else { |
| 754 | write_lane_environment(&environment_path, &spec.environment)?; |
| 755 | Some(environment_path) |
| 756 | }; |
| 757 | let shell_cmd = tmux_log_proxy_command( |
| 758 | log_proxy, |
| 759 | &spec.command, |
| 760 | &record.log_path, |
| 761 | &receipt_path, |
| 762 | &receipt_tmp_path, |
| 763 | environment_path.as_deref(), |
| 764 | &record.id, |
| 765 | ); |
| 766 | |
| 767 | let mut cmd = tmux_command(&socket); |
| 768 | cmd.args(["new-session", "-d", "-s", &session]); |
| 769 | if let Some(cwd) = cwd.as_ref() { |
| 770 | cmd.arg("-c").arg(cwd); |
| 771 | } |
| 772 | cmd.arg(shell_cmd); |
| 773 | let proposed_record = record.clone(); |
| 774 | let spawned = std::cell::Cell::new(false); |
| 775 | let rolled_back = std::cell::Cell::new(false); |
| 776 | match registry.mark_running_if_pending_with( |
| 777 | record, |
| 778 | || { |
| 779 | let status = cmd |
| 780 | .status() |
| 781 | .with_context(|| format!("spawn tmux session {session}"))?; |
| 782 | if !status.success() { |
| 783 | bail!("tmux new-session failed with {status}"); |
| 784 | } |
| 785 | spawned.set(true); |
| 786 | Ok(()) |
| 787 | }, |
| 788 | || { |
| 789 | stop_tmux_session(&socket, &session)?; |
| 790 | rolled_back.set(true); |
| 791 | Ok(()) |
| 792 | }, |
| 793 | ) { |
| 794 | Ok(true) => {} |
| 795 | Ok(false) => { |
| 796 | if let Some(path) = environment_path.as_deref() { |
| 797 | remove_file_if_present(path)?; |
| 798 | } |
| 799 | let mut stopped_record = proposed_record; |
| 800 | stopped_record.stopped_at = record.stopped_at.clone(); |
| 801 | self.cleanup_worktree(&stopped_record)?; |
| 802 | bail!("lane `{}` was stopped before tmux launch", record.id); |
| 803 | } |
| 804 | Err(error) => { |
| 805 | if let Some(path) = environment_path.as_deref() { |
| 806 | let _ = remove_file_if_present(path); |
| 807 | } |
| 808 | if !spawned.get() || rolled_back.get() { |
| 809 | let _ = registry.mark_terminal_if_active(record, LaneStatus::Failed)?; |
| 810 | let mut failed_record = proposed_record; |
| 811 | failed_record.stopped_at = record.stopped_at.clone(); |
| 812 | self.cleanup_worktree(&failed_record)?; |
| 813 | } |
| 814 | return Err(error); |
| 815 | } |
| 816 | } |
| 817 | Ok(()) |
| 818 | } |
| 819 | |
| 820 | fn attach_command(&self, record: &LaneRecord) -> Option<String> { |
| 821 | if record.status != LaneStatus::Running { |
| 822 | return None; |
| 823 | } |
| 824 | let socket = record.tmux_socket.as_ref()?; |
| 825 | let session = record.tmux_session.as_deref()?; |
| 826 | Some(format!( |
| 827 | "tmux -S {} attach -t {}", |
| 828 | shell_escape(&socket.display().to_string()), |
| 829 | shell_escape(session) |
| 830 | )) |
| 831 | } |
| 832 | |
| 833 | fn stop( |
| 834 | &self, |
| 835 | registry: &LaneRegistry, |
| 836 | record: &mut LaneRecord, |
| 837 | fence: Option<u64>, |
| 838 | ) -> Result<TerminalTransition> { |
| 839 | let dry_run = std::env::var_os("CODEWHALE_LANE_TMUX_DRY_RUN").is_some(); |
| 840 | let transition = registry.mark_terminal_if_active_fenced( |
| 841 | record, |
| 842 | LaneStatus::Stopped, |
| 843 | fence, |
| 844 | |current| { |
| 845 | if !dry_run { |
| 846 | match ( |
| 847 | current.tmux_socket.as_deref(), |
| 848 | current.tmux_session.as_deref(), |
| 849 | ) { |
| 850 | (Some(socket), Some(session)) => stop_tmux_session(socket, session)?, |
| 851 | _ if current.status == LaneStatus::Running => { |
| 852 | bail!( |
| 853 | "running tmux lane `{}` has incomplete pinned session metadata; refusing unsafe stop", |
| 854 | current.id |
| 855 | ); |
| 856 | } |
| 857 | _ => {} |
| 858 | } |
| 859 | } |
| 860 | Ok(()) |
| 861 | }, |
| 862 | )?; |
| 863 | if transition.transitioned() { |
| 864 | append_log_event( |
| 865 | &record.log_path, |
| 866 | serde_json::json!({ |
| 867 | "type": "lane_stopped", |
| 868 | "lane_id": record.id, |
| 869 | "session": record.tmux_session, |
| 870 | }), |
| 871 | )?; |
| 872 | remove_file_if_present(&lane_environment_path(&record.log_path))?; |
| 873 | self.cleanup_worktree(record)?; |
| 874 | } |
| 875 | Ok(transition) |
| 876 | } |
| 877 | |
| 878 | fn reconcile(&self, registry: &LaneRegistry, record: &mut LaneRecord) -> Result<bool> { |
| 879 | // Pending is the pre-launch state. Never reconcile it: a concurrent |
| 880 | // status read must not race a fast `tmux new-session` and overwrite |
| 881 | // the start path's later Running transition. |
| 882 | if record.status != LaneStatus::Running { |
| 883 | return Ok(false); |
| 884 | } |
| 885 | |
| 886 | let receipt = read_lane_exit_receipt(&record.log_path, &record.id)?; |
| 887 | let (lane_status, exit_code, reason) = if let Some(receipt) = receipt { |
| 888 | ( |
| 889 | if receipt.exit_code == 0 { |
| 890 | LaneStatus::Completed |
| 891 | } else { |
| 892 | LaneStatus::Failed |
| 893 | }, |
| 894 | Some(receipt.exit_code), |
| 895 | "process_exit_receipt", |
| 896 | ) |
| 897 | } else { |
| 898 | if std::env::var_os("CODEWHALE_LANE_TMUX_DRY_RUN").is_some() { |
| 899 | return Ok(false); |
| 900 | } |
| 901 | let (Some(socket), Some(session)) = ( |
| 902 | record.tmux_socket.as_deref(), |
| 903 | record.tmux_session.as_deref(), |
| 904 | ) else { |
| 905 | bail!( |
| 906 | "running tmux lane `{}` lacks pinned socket/session metadata; refusing unsafe reconciliation", |
| 907 | record.id |
| 908 | ); |
| 909 | }; |
| 910 | match tmux_session_state(socket, session)? { |
| 911 | TmuxSessionState::Present => return Ok(false), |
| 912 | TmuxSessionState::Absent => {} |
| 913 | } |
| 914 | ( |
| 915 | LaneStatus::Failed, |
| 916 | None, |
| 917 | "tmux_session_missing_without_exit_receipt", |
| 918 | ) |
| 919 | }; |
| 920 | |
| 921 | if registry.mark_terminal_if_active(record, lane_status)? { |
| 922 | append_log_event( |
| 923 | &record.log_path, |
| 924 | serde_json::json!({ |
| 925 | "type": "lane_reconciled", |
| 926 | "lane_id": record.id, |
| 927 | "exit_code": exit_code, |
| 928 | "status": lane_status.as_str(), |
| 929 | "reason": reason, |
| 930 | }), |
| 931 | )?; |
| 932 | remove_file_if_present(&lane_environment_path(&record.log_path))?; |
| 933 | self.cleanup_worktree(record)?; |
| 934 | return Ok(true); |
| 935 | } |
| 936 | Ok(false) |
| 937 | } |
| 938 | } |
| 939 | |
| 940 | /// In-process / local command runtime (no tmux). Used for tests and headless. |
| 941 | #[derive(Debug, Default)] |
| 942 | pub struct InlineRuntime; |
| 943 | |
| 944 | impl RuntimeBackend for InlineRuntime { |
| 945 | fn kind(&self) -> RuntimeBackendKind { |
| 946 | RuntimeBackendKind::Inline |
| 947 | } |
| 948 | |
| 949 | fn start( |
| 950 | &self, |
| 951 | registry: &LaneRegistry, |
| 952 | record: &mut LaneRecord, |
| 953 | spec: &LaneStartSpec, |
| 954 | ) -> Result<()> { |
| 955 | if spec.command.is_empty() { |
| 956 | bail!("inline runtime requires a non-empty command"); |
| 957 | } |
| 958 | let cwd = apply_worktree(record, spec)?; |
| 959 | append_log_event( |
| 960 | &record.log_path, |
| 961 | serde_json::json!({ |
| 962 | "type": "lane_started", |
| 963 | "lane_id": record.id, |
| 964 | "runtime": "inline", |
| 965 | "command": spec.command, |
| 966 | }), |
| 967 | )?; |
| 968 | if !registry.mark_running_if_pending(record)? { |
| 969 | bail!( |
| 970 | "lane `{}` was stopped before inline start completed", |
| 971 | record.id |
| 972 | ); |
| 973 | } |
| 974 | |
| 975 | let mut cmd = Command::new(&spec.command[0]); |
| 976 | if spec.command.len() > 1 { |
| 977 | cmd.args(&spec.command[1..]); |
| 978 | } |
| 979 | if let Some(cwd) = cwd.as_ref() { |
| 980 | cmd.current_dir(cwd); |
| 981 | } |
| 982 | cmd.envs(spec.environment.iter().map(|(key, value)| (key, value))); |
| 983 | cmd.stdout(Stdio::piped()).stderr(Stdio::piped()); |
| 984 | let mut child = match cmd.spawn() { |
| 985 | Ok(child) => child, |
| 986 | Err(err) => { |
| 987 | append_log_event( |
| 988 | &record.log_path, |
| 989 | serde_json::json!({ |
| 990 | "type": "lane_failed", |
| 991 | "lane_id": record.id, |
| 992 | "error": err.to_string(), |
| 993 | }), |
| 994 | )?; |
| 995 | let _ = registry.mark_terminal_if_active(record, LaneStatus::Failed)?; |
| 996 | return Err(err).with_context(|| format!("run inline command {:?}", spec.command)); |
| 997 | } |
| 998 | }; |
| 999 | let stdout = child |
| 1000 | .stdout |
| 1001 | .take() |
| 1002 | .context("inline lane child stdout was not piped")?; |
| 1003 | let stderr = child |
| 1004 | .stderr |
| 1005 | .take() |
| 1006 | .context("inline lane child stderr was not piped")?; |
| 1007 | let stdout_log = record.log_path.clone(); |
| 1008 | let stderr_log = record.log_path.clone(); |
| 1009 | let stdout_thread = |
| 1010 | thread::spawn(move || stream_child_output(stdout, stdout_log, "stdout")); |
| 1011 | let stderr_thread = |
| 1012 | thread::spawn(move || stream_child_output(stderr, stderr_log, "stderr")); |
| 1013 | let status = child |
| 1014 | .wait() |
| 1015 | .with_context(|| format!("wait for inline command {:?}", spec.command))?; |
| 1016 | let stdout_result = stdout_thread |
| 1017 | .join() |
| 1018 | .map_err(|_| anyhow::anyhow!("inline stdout logger panicked"))?; |
| 1019 | let stderr_result = stderr_thread |
| 1020 | .join() |
| 1021 | .map_err(|_| anyhow::anyhow!("inline stderr logger panicked"))?; |
| 1022 | let logging_error = stdout_result.err().or_else(|| stderr_result.err()); |
| 1023 | |
| 1024 | if status.success() && logging_error.is_none() { |
| 1025 | append_log_event( |
| 1026 | &record.log_path, |
| 1027 | serde_json::json!({"type": "lane_completed", "lane_id": record.id}), |
| 1028 | )?; |
| 1029 | let _ = registry.mark_terminal_if_active(record, LaneStatus::Completed)?; |
| 1030 | } else { |
| 1031 | append_log_event( |
| 1032 | &record.log_path, |
| 1033 | serde_json::json!({ |
| 1034 | "type": "lane_failed", |
| 1035 | "lane_id": record.id, |
| 1036 | "status": format!("{status}"), |
| 1037 | "logging_error": logging_error.as_ref().map(ToString::to_string), |
| 1038 | }), |
| 1039 | )?; |
| 1040 | let _ = registry.mark_terminal_if_active(record, LaneStatus::Failed)?; |
| 1041 | } |
| 1042 | if let Some(err) = logging_error { |
| 1043 | return Err(err).context("stream inline lane output"); |
| 1044 | } |
| 1045 | Ok(()) |
| 1046 | } |
| 1047 | |
| 1048 | fn attach_command(&self, _record: &LaneRecord) -> Option<String> { |
| 1049 | None |
| 1050 | } |
| 1051 | |
| 1052 | fn stop( |
| 1053 | &self, |
| 1054 | registry: &LaneRegistry, |
| 1055 | record: &mut LaneRecord, |
| 1056 | fence: Option<u64>, |
| 1057 | ) -> Result<TerminalTransition> { |
| 1058 | let transition = registry.mark_terminal_if_active_fenced( |
| 1059 | record, |
| 1060 | LaneStatus::Stopped, |
| 1061 | fence, |
| 1062 | |current| { |
| 1063 | if current.status == LaneStatus::Running { |
| 1064 | bail!( |
| 1065 | "inline lane `{}` cannot be stopped safely from another process", |
| 1066 | current.id |
| 1067 | ); |
| 1068 | } |
| 1069 | Ok(()) |
| 1070 | }, |
| 1071 | )?; |
| 1072 | if transition.transitioned() { |
| 1073 | self.cleanup_worktree(record)?; |
| 1074 | } |
| 1075 | Ok(transition) |
| 1076 | } |
| 1077 | } |
| 1078 | |
| 1079 | /// Placeholder for remote VM / CI backends (surface only in Phase 1). |
| 1080 | #[derive(Debug)] |
| 1081 | struct StubRuntime { |
| 1082 | kind: RuntimeBackendKind, |
| 1083 | } |
| 1084 | |
| 1085 | impl RuntimeBackend for StubRuntime { |
| 1086 | fn kind(&self) -> RuntimeBackendKind { |
| 1087 | self.kind |
| 1088 | } |
| 1089 | |
| 1090 | fn start( |
| 1091 | &self, |
| 1092 | registry: &LaneRegistry, |
| 1093 | record: &mut LaneRecord, |
| 1094 | _spec: &LaneStartSpec, |
| 1095 | ) -> Result<()> { |
| 1096 | let error = format!( |
| 1097 | "{} runtime is not implemented; use tmux or inline", |
| 1098 | self.kind.as_str() |
| 1099 | ); |
| 1100 | append_log_event( |
| 1101 | &record.log_path, |
| 1102 | serde_json::json!({ |
| 1103 | "type": "lane_failed", |
| 1104 | "lane_id": record.id, |
| 1105 | "runtime": self.kind.as_str(), |
| 1106 | "error": &error, |
| 1107 | }), |
| 1108 | )?; |
| 1109 | let _ = registry.mark_terminal_if_active(record, LaneStatus::Failed)?; |
| 1110 | bail!("{error}") |
| 1111 | } |
| 1112 | |
| 1113 | fn attach_command(&self, _record: &LaneRecord) -> Option<String> { |
| 1114 | None |
| 1115 | } |
| 1116 | |
| 1117 | fn stop( |
| 1118 | &self, |
| 1119 | registry: &LaneRegistry, |
| 1120 | record: &mut LaneRecord, |
| 1121 | fence: Option<u64>, |
| 1122 | ) -> Result<TerminalTransition> { |
| 1123 | registry.mark_terminal_if_active_fenced(record, LaneStatus::Stopped, fence, |_| Ok(())) |
| 1124 | } |
| 1125 | } |
| 1126 | |
| 1127 | fn shell_escape(s: &str) -> String { |
| 1128 | format!("'{}'", s.replace('\'', "'\\''")) |
| 1129 | } |
| 1130 | |
| 1131 | fn shell_join(args: &[String]) -> String { |
| 1132 | args.iter() |
| 1133 | .map(|a| shell_escape(a)) |
| 1134 | .collect::<Vec<_>>() |
| 1135 | .join(" ") |
| 1136 | } |
| 1137 | |
| 1138 | #[cfg(test)] |
| 1139 | mod tests { |
| 1140 | use super::*; |
| 1141 | use std::ffi::OsString; |
| 1142 | use std::sync::{Mutex, MutexGuard, OnceLock}; |
| 1143 | use tempfile::tempdir; |
| 1144 | |
| 1145 | fn tmux_env_lock() -> MutexGuard<'static, ()> { |
| 1146 | static LOCK: OnceLock<Mutex<()>> = OnceLock::new(); |
| 1147 | LOCK.get_or_init(|| Mutex::new(())) |
| 1148 | .lock() |
| 1149 | .unwrap_or_else(|poisoned| poisoned.into_inner()) |
| 1150 | } |
| 1151 | |
| 1152 | struct ScopedEnvVar { |
| 1153 | name: &'static str, |
| 1154 | previous: Option<OsString>, |
| 1155 | } |
| 1156 | |
| 1157 | impl ScopedEnvVar { |
| 1158 | fn set(name: &'static str, value: &std::ffi::OsStr) -> Self { |
| 1159 | let previous = std::env::var_os(name); |
| 1160 | // SAFETY: tmux environment tests hold `tmux_env_lock` and restore |
| 1161 | // the process environment in Drop. |
| 1162 | unsafe { std::env::set_var(name, value) }; |
| 1163 | Self { name, previous } |
| 1164 | } |
| 1165 | |
| 1166 | #[cfg(unix)] |
| 1167 | fn remove(name: &'static str) -> Self { |
| 1168 | let previous = std::env::var_os(name); |
| 1169 | // SAFETY: tmux environment tests hold `tmux_env_lock` and restore |
| 1170 | // the process environment in Drop. |
| 1171 | unsafe { std::env::remove_var(name) }; |
| 1172 | Self { name, previous } |
| 1173 | } |
| 1174 | } |
| 1175 | |
| 1176 | impl Drop for ScopedEnvVar { |
| 1177 | fn drop(&mut self) { |
| 1178 | // SAFETY: paired with the serialized mutation above. |
| 1179 | unsafe { |
| 1180 | if let Some(previous) = self.previous.take() { |
| 1181 | std::env::set_var(self.name, previous); |
| 1182 | } else { |
| 1183 | std::env::remove_var(self.name); |
| 1184 | } |
| 1185 | } |
| 1186 | } |
| 1187 | } |
| 1188 | |
| 1189 | #[test] |
| 1190 | fn tmux_dry_run_start_attach_stop_roundtrip() { |
| 1191 | let _env_guard = tmux_env_lock(); |
| 1192 | let _dry_run = ScopedEnvVar::set("CODEWHALE_LANE_TMUX_DRY_RUN", std::ffi::OsStr::new("1")); |
| 1193 | let dir = tempdir().unwrap(); |
| 1194 | let reg = LaneRegistry::open(dir.path()).unwrap(); |
| 1195 | let mut record = reg |
| 1196 | .create_pending( |
| 1197 | Some("stopship".into()), |
| 1198 | Some("stopship".into()), |
| 1199 | Some("4375".into()), |
| 1200 | None, |
| 1201 | RuntimeBackendKind::Tmux, |
| 1202 | None, |
| 1203 | ) |
| 1204 | .unwrap(); |
| 1205 | let backend = TmuxRuntime; |
| 1206 | backend |
| 1207 | .start( |
| 1208 | ®, |
| 1209 | &mut record, |
| 1210 | &LaneStartSpec { |
| 1211 | command: vec!["echo".into(), "hello-lane".into()], |
| 1212 | cwd: None, |
| 1213 | environment: Vec::new(), |
| 1214 | log_proxy: None, |
| 1215 | worktree: None, |
| 1216 | }, |
| 1217 | ) |
| 1218 | .unwrap(); |
| 1219 | assert_eq!(record.status, LaneStatus::Running); |
| 1220 | assert!(record.tmux_session.is_some()); |
| 1221 | let expected_socket = reg.root().join("tmux.sock"); |
| 1222 | assert_eq!( |
| 1223 | record.tmux_socket.as_deref(), |
| 1224 | Some(expected_socket.as_path()) |
| 1225 | ); |
| 1226 | let attach = backend.attach_command(&record).expect("attach"); |
| 1227 | assert!(attach.contains("tmux -S")); |
| 1228 | assert!(attach.contains(&expected_socket.display().to_string())); |
| 1229 | assert!(attach.contains("attach -t")); |
| 1230 | let log = std::fs::read_to_string(&record.log_path).unwrap(); |
| 1231 | assert!(log.contains("lane_started")); |
| 1232 | |
| 1233 | backend.stop(®, &mut record, None).unwrap(); |
| 1234 | assert_eq!(record.status, LaneStatus::Stopped); |
| 1235 | let reloaded = reg.load(&record.id).unwrap(); |
| 1236 | assert_eq!(reloaded.status, LaneStatus::Stopped); |
| 1237 | } |
| 1238 | |
| 1239 | #[cfg(unix)] |
| 1240 | #[test] |
| 1241 | fn unavailable_tmux_fails_lane_instead_of_faking_running() { |
| 1242 | use std::os::unix::fs::symlink; |
| 1243 | |
| 1244 | let _env_guard = tmux_env_lock(); |
| 1245 | let dir = tempdir().unwrap(); |
| 1246 | let bin_dir = dir.path().join("bin"); |
| 1247 | fs::create_dir(&bin_dir).unwrap(); |
| 1248 | symlink("/usr/bin/false", bin_dir.join("tmux")).unwrap(); |
| 1249 | let prior_path = std::env::var_os("PATH").unwrap_or_default(); |
| 1250 | let combined_path = std::env::join_paths( |
| 1251 | std::iter::once(bin_dir).chain(std::env::split_paths(&prior_path)), |
| 1252 | ) |
| 1253 | .unwrap(); |
| 1254 | let _path = ScopedEnvVar::set("PATH", &combined_path); |
| 1255 | let _dry_run = ScopedEnvVar::remove("CODEWHALE_LANE_TMUX_DRY_RUN"); |
| 1256 | |
| 1257 | let reg = LaneRegistry::open(dir.path().join("registry")).unwrap(); |
| 1258 | let mut record = reg |
| 1259 | .create_pending(None, None, None, None, RuntimeBackendKind::Tmux, None) |
| 1260 | .unwrap(); |
| 1261 | let error = TmuxRuntime |
| 1262 | .start( |
| 1263 | ®, |
| 1264 | &mut record, |
| 1265 | &LaneStartSpec { |
| 1266 | command: vec!["/bin/true".to_string()], |
| 1267 | cwd: None, |
| 1268 | environment: Vec::new(), |
| 1269 | log_proxy: Some(PathBuf::from("/bin/true")), |
| 1270 | worktree: None, |
| 1271 | }, |
| 1272 | ) |
| 1273 | .unwrap_err(); |
| 1274 | assert!(error.to_string().contains("tmux runtime is unavailable")); |
| 1275 | assert_eq!(record.status, LaneStatus::Failed); |
| 1276 | assert_eq!(reg.load(&record.id).unwrap().status, LaneStatus::Failed); |
| 1277 | assert!( |
| 1278 | String::from_utf8_lossy(&fs::read(&record.log_path).unwrap()).contains("lane_failed") |
| 1279 | ); |
| 1280 | } |
| 1281 | |
| 1282 | #[test] |
| 1283 | fn unimplemented_remote_runtimes_fail_terminally() { |
| 1284 | for kind in [RuntimeBackendKind::Vm, RuntimeBackendKind::Ci] { |
| 1285 | let dir = tempdir().unwrap(); |
| 1286 | let reg = LaneRegistry::open(dir.path()).unwrap(); |
| 1287 | let mut record = reg |
| 1288 | .create_pending(None, None, None, None, kind, None) |
| 1289 | .unwrap(); |
| 1290 | let error = resolve_backend(kind) |
| 1291 | .start( |
| 1292 | ®, |
| 1293 | &mut record, |
| 1294 | &LaneStartSpec { |
| 1295 | command: vec!["/bin/true".to_string()], |
| 1296 | cwd: None, |
| 1297 | environment: Vec::new(), |
| 1298 | log_proxy: None, |
| 1299 | worktree: None, |
| 1300 | }, |
| 1301 | ) |
| 1302 | .unwrap_err(); |
| 1303 | assert!(error.to_string().contains("runtime is not implemented")); |
| 1304 | assert_eq!(record.status, LaneStatus::Failed); |
| 1305 | assert_eq!(reg.load(&record.id).unwrap().status, LaneStatus::Failed); |
| 1306 | } |
| 1307 | } |
| 1308 | |
| 1309 | #[test] |
| 1310 | fn inline_runtime_writes_stream_json_log() { |
| 1311 | let dir = tempdir().unwrap(); |
| 1312 | let reg = LaneRegistry::open(dir.path()).unwrap(); |
| 1313 | let mut record = reg |
| 1314 | .create_pending( |
| 1315 | Some("demo".into()), |
| 1316 | None, |
| 1317 | None, |
| 1318 | Some("echo".into()), |
| 1319 | RuntimeBackendKind::Inline, |
| 1320 | None, |
| 1321 | ) |
| 1322 | .unwrap(); |
| 1323 | InlineRuntime |
| 1324 | .start( |
| 1325 | ®, |
| 1326 | &mut record, |
| 1327 | &LaneStartSpec { |
| 1328 | command: vec!["echo".into(), "inline-ok".into()], |
| 1329 | cwd: None, |
| 1330 | environment: Vec::new(), |
| 1331 | log_proxy: None, |
| 1332 | worktree: None, |
| 1333 | }, |
| 1334 | ) |
| 1335 | .unwrap(); |
| 1336 | assert_eq!(record.status, LaneStatus::Completed); |
| 1337 | let log = std::fs::read_to_string(&record.log_path).unwrap(); |
| 1338 | assert!(log.contains("inline-ok"), "log={log}"); |
| 1339 | assert!(log.contains("lane_completed")); |
| 1340 | } |
| 1341 | |
| 1342 | #[test] |
| 1343 | fn lane_start_spec_debug_redacts_environment_values() { |
| 1344 | let spec = LaneStartSpec { |
| 1345 | command: vec!["codewhale-tui".into()], |
| 1346 | cwd: None, |
| 1347 | environment: vec![("DEEPSEEK_API_KEY".into(), "secret-value".into())], |
| 1348 | log_proxy: None, |
| 1349 | worktree: None, |
| 1350 | }; |
| 1351 | let rendered = format!("{spec:?}"); |
| 1352 | assert!(rendered.contains("DEEPSEEK_API_KEY")); |
| 1353 | assert!(!rendered.contains("secret-value")); |
| 1354 | } |
| 1355 | |
| 1356 | #[cfg(unix)] |
| 1357 | #[test] |
| 1358 | fn inline_runtime_persists_typed_receipts_before_process_exit() { |
| 1359 | let dir = tempdir().unwrap(); |
| 1360 | let reg = LaneRegistry::open(dir.path()).unwrap(); |
| 1361 | let record = reg |
| 1362 | .create_pending( |
| 1363 | Some("demo".into()), |
| 1364 | None, |
| 1365 | None, |
| 1366 | None, |
| 1367 | RuntimeBackendKind::Inline, |
| 1368 | None, |
| 1369 | ) |
| 1370 | .unwrap(); |
| 1371 | let log_path = record.log_path.clone(); |
| 1372 | let (done_tx, done_rx) = std::sync::mpsc::channel(); |
| 1373 | let handle = thread::spawn(move || { |
| 1374 | let mut record = record; |
| 1375 | let result = InlineRuntime.start( |
| 1376 | ®, |
| 1377 | &mut record, |
| 1378 | &LaneStartSpec { |
| 1379 | command: vec![ |
| 1380 | "sh".into(), |
| 1381 | "-c".into(), |
| 1382 | "printf '%s\\n' '{\"type\":\"workflow_event\",\"run_id\":\"workflow_live\",\"event\":{\"type\":\"run_started\"}}'; sleep 0.25; printf done" |
| 1383 | .into(), |
| 1384 | ], |
| 1385 | cwd: None, |
| 1386 | environment: Vec::new(), |
| 1387 | log_proxy: None, |
| 1388 | worktree: None, |
| 1389 | }, |
| 1390 | ); |
| 1391 | let _ = done_tx.send(result); |
| 1392 | }); |
| 1393 | |
| 1394 | let mut observed_live = false; |
| 1395 | for _ in 0..50 { |
| 1396 | let log = std::fs::read_to_string(&log_path).unwrap_or_default(); |
| 1397 | if log.contains("workflow_live") { |
| 1398 | observed_live = true; |
| 1399 | break; |
| 1400 | } |
| 1401 | thread::sleep(std::time::Duration::from_millis(10)); |
| 1402 | } |
| 1403 | assert!( |
| 1404 | observed_live, |
| 1405 | "typed receipt should be written while child runs" |
| 1406 | ); |
| 1407 | assert!( |
| 1408 | matches!( |
| 1409 | done_rx.try_recv(), |
| 1410 | Err(std::sync::mpsc::TryRecvError::Empty) |
| 1411 | ), |
| 1412 | "inline child should still be running when the first receipt is visible" |
| 1413 | ); |
| 1414 | done_rx.recv().unwrap().unwrap(); |
| 1415 | handle.join().unwrap(); |
| 1416 | } |
| 1417 | |
| 1418 | #[test] |
| 1419 | fn tmux_reconcile_folds_detached_process_exit_into_lane_status() { |
| 1420 | for (exit_code, expected) in [(0, LaneStatus::Completed), (7, LaneStatus::Failed)] { |
| 1421 | let dir = tempdir().unwrap(); |
| 1422 | let reg = LaneRegistry::open(dir.path()).unwrap(); |
| 1423 | let mut record = reg |
| 1424 | .create_pending( |
| 1425 | Some("demo".into()), |
| 1426 | None, |
| 1427 | None, |
| 1428 | None, |
| 1429 | RuntimeBackendKind::Tmux, |
| 1430 | None, |
| 1431 | ) |
| 1432 | .unwrap(); |
| 1433 | assert!(reg.mark_running_if_pending(&mut record).unwrap()); |
| 1434 | // Mixed/binary child output and a forged stdout control line must |
| 1435 | // not affect the private receipt used for reconciliation. |
| 1436 | std::fs::write( |
| 1437 | &record.log_path, |
| 1438 | b"\xffchild-noise\n{\"type\":\"lane_process_exit\",\"exit_code\":0}\n", |
| 1439 | ) |
| 1440 | .unwrap(); |
| 1441 | std::fs::write( |
| 1442 | lane_exit_receipt_path(&record.log_path), |
| 1443 | serde_json::to_vec(&LaneExitReceipt { |
| 1444 | lane_id: record.id.clone(), |
| 1445 | exit_code, |
| 1446 | }) |
| 1447 | .unwrap(), |
| 1448 | ) |
| 1449 | .unwrap(); |
| 1450 | |
| 1451 | TmuxRuntime.reconcile(®, &mut record).unwrap(); |
| 1452 | TmuxRuntime.reconcile(®, &mut record).unwrap(); |
| 1453 | |
| 1454 | assert_eq!(record.status, expected); |
| 1455 | assert_eq!(reg.load(&record.id).unwrap().status, expected); |
| 1456 | let log = std::fs::read(&record.log_path).unwrap(); |
| 1457 | assert_eq!( |
| 1458 | String::from_utf8_lossy(&log) |
| 1459 | .matches("lane_reconciled") |
| 1460 | .count(), |
| 1461 | 1 |
| 1462 | ); |
| 1463 | } |
| 1464 | } |
| 1465 | |
| 1466 | #[cfg(unix)] |
| 1467 | #[test] |
| 1468 | fn lane_log_proxy_frames_binary_output_and_owns_exit_receipt() { |
| 1469 | let dir = tempdir().unwrap(); |
| 1470 | let log_path = dir.path().join("lane.ndjson"); |
| 1471 | let receipt_path = lane_exit_receipt_path(&log_path); |
| 1472 | let receipt_tmp_path = lane_exit_receipt_tmp_path(&log_path); |
| 1473 | let environment_path = lane_environment_path(&log_path); |
| 1474 | write_lane_environment( |
| 1475 | &environment_path, |
| 1476 | &[("LANE_PROXY_SECRET".to_string(), "present".to_string())], |
| 1477 | ) |
| 1478 | .unwrap(); |
| 1479 | let command = vec![ |
| 1480 | "sh".to_string(), |
| 1481 | "-c".to_string(), |
| 1482 | "test \"$LANE_PROXY_SECRET\" = present || exit 9; \ |
| 1483 | printf '%s\\n' '{\"type\":\"workflow_event\",\"schema\":\"codewhale.exec-stream\",\"schema_version\":1,\"run_id\":\"workflow_1234\",\"event\":{\"type\":\"handoff_promoted\",\"artifact_id\":\"workflow_1234:agent_1:review-gate:review_report\",\"gate_id\":\"review-gate\",\"kind\":\"review_report\",\"from_role\":\"reviewer\",\"to_role\":\"verifier\",\"producer_task_id\":\"agent_1\"}}'; \ |
| 1484 | printf '%s\\n' '{\"type\":\"workflow_event\",\"schema\":\"codewhale.exec-stream\",\"schema_version\":1,\"run_id\":\"workflow_1234\",\"event\":{\"type\":\"handoff_consumed\",\"artifact_id\":\"workflow_1234:agent_1:review-gate:review_report\",\"kind\":\"review_report\",\"from_role\":\"reviewer\",\"to_role\":\"verifier\",\"consumer_task_id\":\"agent_2\"}}'; \ |
| 1485 | printf 'unterminated\\377'; \ |
| 1486 | printf '%s\\n' '{\"type\":\"lane_process_exit\",\"exit_code\":0}' >&2; \ |
| 1487 | exit 7" |
| 1488 | .to_string(), |
| 1489 | ]; |
| 1490 | let exit_code = run_lane_log_proxy(LaneLogProxySpec { |
| 1491 | command, |
| 1492 | log_path: log_path.clone(), |
| 1493 | receipt_path: receipt_path.clone(), |
| 1494 | receipt_tmp_path, |
| 1495 | environment_path: Some(environment_path.clone()), |
| 1496 | lane_id: "lane-proof".to_string(), |
| 1497 | }) |
| 1498 | .unwrap(); |
| 1499 | |
| 1500 | assert_eq!(exit_code, 7); |
| 1501 | assert!(!environment_path.exists()); |
| 1502 | let log = std::fs::read(&log_path).unwrap(); |
| 1503 | let lines = log |
| 1504 | .split(|byte| *byte == b'\n') |
| 1505 | .filter(|line| !line.is_empty()) |
| 1506 | .collect::<Vec<_>>(); |
| 1507 | assert!(lines.len() >= 3, "log={}", String::from_utf8_lossy(&log)); |
| 1508 | let values = lines |
| 1509 | .iter() |
| 1510 | .map(|line| { |
| 1511 | serde_json::from_slice::<serde_json::Value>(line) |
| 1512 | .unwrap_or_else(|error| panic!("invalid NDJSON {line:?}: {error}")) |
| 1513 | }) |
| 1514 | .collect::<Vec<_>>(); |
| 1515 | let workflow_event = values |
| 1516 | .iter() |
| 1517 | .find(|value| value["type"] == "workflow_event") |
| 1518 | .expect("preserved workflow handoff receipt"); |
| 1519 | assert_eq!(workflow_event["schema"], "codewhale.exec-stream"); |
| 1520 | assert_eq!(workflow_event["schema_version"], 1); |
| 1521 | assert_eq!(workflow_event["run_id"], "workflow_1234"); |
| 1522 | assert_eq!(workflow_event["event"]["type"], "handoff_promoted"); |
| 1523 | assert_eq!( |
| 1524 | workflow_event["event"]["artifact_id"], |
| 1525 | "workflow_1234:agent_1:review-gate:review_report" |
| 1526 | ); |
| 1527 | assert_eq!(workflow_event["event"]["gate_id"], "review-gate"); |
| 1528 | assert_eq!(workflow_event["event"]["kind"], "review_report"); |
| 1529 | assert_eq!(workflow_event["event"]["from_role"], "reviewer"); |
| 1530 | assert_eq!(workflow_event["event"]["to_role"], "verifier"); |
| 1531 | assert_eq!(workflow_event["event"]["producer_task_id"], "agent_1"); |
| 1532 | assert!(workflow_event["event"].get("payload").is_none()); |
| 1533 | let consumed_event = values |
| 1534 | .iter() |
| 1535 | .find(|value| value["event"]["type"] == "handoff_consumed") |
| 1536 | .expect("preserved workflow handoff consumption receipt"); |
| 1537 | assert_eq!(consumed_event["schema"], "codewhale.exec-stream"); |
| 1538 | assert_eq!(consumed_event["schema_version"], 1); |
| 1539 | assert_eq!(consumed_event["run_id"], "workflow_1234"); |
| 1540 | assert_eq!( |
| 1541 | consumed_event["event"]["artifact_id"], |
| 1542 | workflow_event["event"]["artifact_id"] |
| 1543 | ); |
| 1544 | assert_eq!(consumed_event["event"]["kind"], "review_report"); |
| 1545 | assert_eq!(consumed_event["event"]["from_role"], "reviewer"); |
| 1546 | assert_eq!(consumed_event["event"]["to_role"], "verifier"); |
| 1547 | assert_eq!(consumed_event["event"]["consumer_task_id"], "agent_2"); |
| 1548 | assert!(consumed_event["event"].get("payload").is_none()); |
| 1549 | let rendered = String::from_utf8_lossy(&log); |
| 1550 | assert!(rendered.contains("workflow_event")); |
| 1551 | assert!(rendered.contains("lane_process_exit")); |
| 1552 | assert!(rendered.contains("lane_log")); |
| 1553 | let receipt = read_lane_exit_receipt(&log_path, "lane-proof") |
| 1554 | .unwrap() |
| 1555 | .expect("private receipt"); |
| 1556 | assert_eq!(receipt.exit_code, 7); |
| 1557 | } |
| 1558 | |
| 1559 | #[test] |
| 1560 | fn invalid_environment_key_never_leaves_a_partial_secret_file() { |
| 1561 | let dir = tempdir().unwrap(); |
| 1562 | let path = dir.path().join("lane.env.json"); |
| 1563 | let result = write_lane_environment( |
| 1564 | &path, |
| 1565 | &[ |
| 1566 | ("VALID_KEY".to_string(), "first-secret".to_string()), |
| 1567 | ("INVALID-KEY".to_string(), "second-secret".to_string()), |
| 1568 | ], |
| 1569 | ); |
| 1570 | assert!(result.is_err()); |
| 1571 | assert!(!path.exists()); |
| 1572 | assert!(!lane_environment_tmp_path(&path).exists()); |
| 1573 | } |
| 1574 | |
| 1575 | #[cfg(unix)] |
| 1576 | #[test] |
| 1577 | fn private_environment_is_mode_0600_and_malformed_input_is_removed() { |
| 1578 | use std::os::unix::fs::PermissionsExt; |
| 1579 | |
| 1580 | let dir = tempdir().unwrap(); |
| 1581 | let path = dir.path().join("lane.env.json"); |
| 1582 | write_lane_environment(&path, &[("SECRET".to_string(), "sensitive".to_string())]).unwrap(); |
| 1583 | assert_eq!( |
| 1584 | fs::metadata(&path).unwrap().permissions().mode() & 0o777, |
| 1585 | 0o600 |
| 1586 | ); |
| 1587 | |
| 1588 | fs::write(&path, b"{malformed").unwrap(); |
| 1589 | let log_path = dir.path().join("lane.ndjson"); |
| 1590 | let exit_code = run_lane_log_proxy(LaneLogProxySpec { |
| 1591 | command: vec!["/bin/true".to_string()], |
| 1592 | receipt_path: lane_exit_receipt_path(&log_path), |
| 1593 | receipt_tmp_path: lane_exit_receipt_tmp_path(&log_path), |
| 1594 | environment_path: Some(path.clone()), |
| 1595 | lane_id: "lane-malformed-env".to_string(), |
| 1596 | log_path: log_path.clone(), |
| 1597 | }) |
| 1598 | .unwrap(); |
| 1599 | assert_eq!(exit_code, LANE_PROXY_FAILURE_EXIT_CODE); |
| 1600 | assert!(!path.exists()); |
| 1601 | assert!(String::from_utf8_lossy(&fs::read(log_path).unwrap()).contains("lane_proxy_error")); |
| 1602 | } |
| 1603 | |
| 1604 | #[cfg(unix)] |
| 1605 | #[test] |
| 1606 | fn tmux_stop_failure_keeps_lane_running_and_preserves_cleanup_targets() { |
| 1607 | use std::os::unix::fs::symlink; |
| 1608 | |
| 1609 | let _env_guard = tmux_env_lock(); |
| 1610 | let dir = tempdir().unwrap(); |
| 1611 | let bin_dir = dir.path().join("bin"); |
| 1612 | fs::create_dir(&bin_dir).unwrap(); |
| 1613 | symlink("/usr/bin/false", bin_dir.join("tmux")).unwrap(); |
| 1614 | let prior_path = std::env::var_os("PATH").unwrap_or_default(); |
| 1615 | let combined_path = std::env::join_paths( |
| 1616 | std::iter::once(bin_dir.clone()).chain(std::env::split_paths(&prior_path)), |
| 1617 | ) |
| 1618 | .unwrap(); |
| 1619 | let _path = ScopedEnvVar::set("PATH", &combined_path); |
| 1620 | let _dry_run = ScopedEnvVar::remove("CODEWHALE_LANE_TMUX_DRY_RUN"); |
| 1621 | |
| 1622 | let reg = LaneRegistry::open(dir.path().join("registry")).unwrap(); |
| 1623 | let mut record = reg |
| 1624 | .create_pending( |
| 1625 | Some("demo".into()), |
| 1626 | None, |
| 1627 | None, |
| 1628 | None, |
| 1629 | RuntimeBackendKind::Tmux, |
| 1630 | Some(0), |
| 1631 | ) |
| 1632 | .unwrap(); |
| 1633 | record.tmux_session = Some(format!("cw-{}", record.id)); |
| 1634 | record.tmux_socket = Some(reg.root().join("tmux.sock")); |
| 1635 | let worktree = dir.path().join("worktree"); |
| 1636 | fs::create_dir(&worktree).unwrap(); |
| 1637 | record.worktree_path = Some(worktree.clone()); |
| 1638 | let environment_path = lane_environment_path(&record.log_path); |
| 1639 | write_lane_environment( |
| 1640 | &environment_path, |
| 1641 | &[("SECRET".to_string(), "still-private".to_string())], |
| 1642 | ) |
| 1643 | .unwrap(); |
| 1644 | assert!(reg.mark_running_if_pending(&mut record).unwrap()); |
| 1645 | |
| 1646 | let error = TmuxRuntime.stop(®, &mut record, None).unwrap_err(); |
| 1647 | assert!(format!("{error:#}").contains("tmux has-session")); |
| 1648 | assert_eq!(record.status, LaneStatus::Running); |
| 1649 | assert_eq!(reg.load(&record.id).unwrap().status, LaneStatus::Running); |
| 1650 | assert!(environment_path.exists()); |
| 1651 | assert!(worktree.exists()); |
| 1652 | } |
| 1653 | |
| 1654 | #[cfg(unix)] |
| 1655 | #[test] |
| 1656 | fn tmux_reconcile_marks_vanished_session_failed_without_receipt() { |
| 1657 | use std::os::unix::fs::PermissionsExt; |
| 1658 | |
| 1659 | let _env_guard = tmux_env_lock(); |
| 1660 | let dir = tempdir().unwrap(); |
| 1661 | let bin_dir = dir.path().join("bin"); |
| 1662 | fs::create_dir(&bin_dir).unwrap(); |
| 1663 | let tmux = bin_dir.join("tmux"); |
| 1664 | fs::write( |
| 1665 | &tmux, |
| 1666 | "#!/bin/sh\nprintf '%s\\n' 'no server running on /tmp/codewhale-test.sock' >&2\nexit 1\n", |
| 1667 | ) |
| 1668 | .unwrap(); |
| 1669 | let mut permissions = fs::metadata(&tmux).unwrap().permissions(); |
| 1670 | permissions.set_mode(0o755); |
| 1671 | fs::set_permissions(&tmux, permissions).unwrap(); |
| 1672 | let prior_path = std::env::var_os("PATH").unwrap_or_default(); |
| 1673 | let combined_path = std::env::join_paths( |
| 1674 | std::iter::once(bin_dir).chain(std::env::split_paths(&prior_path)), |
| 1675 | ) |
| 1676 | .unwrap(); |
| 1677 | let _path = ScopedEnvVar::set("PATH", &combined_path); |
| 1678 | let _dry_run = ScopedEnvVar::remove("CODEWHALE_LANE_TMUX_DRY_RUN"); |
| 1679 | |
| 1680 | let reg = LaneRegistry::open(dir.path()).unwrap(); |
| 1681 | let mut record = reg |
| 1682 | .create_pending( |
| 1683 | Some("demo".into()), |
| 1684 | None, |
| 1685 | None, |
| 1686 | None, |
| 1687 | RuntimeBackendKind::Tmux, |
| 1688 | None, |
| 1689 | ) |
| 1690 | .unwrap(); |
| 1691 | record.tmux_session = Some(format!("missing-{}", record.id)); |
| 1692 | record.tmux_socket = Some(reg.root().join("tmux.sock")); |
| 1693 | assert!(reg.mark_running_if_pending(&mut record).unwrap()); |
| 1694 | |
| 1695 | TmuxRuntime.reconcile(®, &mut record).unwrap(); |
| 1696 | |
| 1697 | assert_eq!(record.status, LaneStatus::Failed); |
| 1698 | let log = std::fs::read_to_string(&record.log_path).unwrap(); |
| 1699 | assert!(log.contains("tmux_session_missing_without_exit_receipt")); |
| 1700 | } |
| 1701 | } |
| 1702 |