| 1 | //! Durable lane registry under `$CODEWHALE_HOME/lanes/`. |
| 2 | |
| 3 | use std::fs::{self, OpenOptions}; |
| 4 | use std::path::{Path, PathBuf}; |
| 5 | |
| 6 | use anyhow::{Context, Result, bail}; |
| 7 | use chrono::{SecondsFormat, Utc}; |
| 8 | use serde::{Deserialize, Serialize}; |
| 9 | |
| 10 | use crate::runtime::RuntimeBackendKind; |
| 11 | |
| 12 | const LANES_SUBDIR: &str = "lanes"; |
| 13 | const LOGS_SUBDIR: &str = "logs"; |
| 14 | |
| 15 | /// Lifecycle status for a running workflow instance. |
| 16 | #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)] |
| 17 | #[serde(rename_all = "snake_case")] |
| 18 | pub enum LaneStatus { |
| 19 | Pending, |
| 20 | Running, |
| 21 | Stopped, |
| 22 | Failed, |
| 23 | Completed, |
| 24 | } |
| 25 | |
| 26 | impl LaneStatus { |
| 27 | pub fn as_str(self) -> &'static str { |
| 28 | match self { |
| 29 | Self::Pending => "pending", |
| 30 | Self::Running => "running", |
| 31 | Self::Stopped => "stopped", |
| 32 | Self::Failed => "failed", |
| 33 | Self::Completed => "completed", |
| 34 | } |
| 35 | } |
| 36 | |
| 37 | pub fn is_active(self) -> bool { |
| 38 | matches!(self, Self::Pending | Self::Running) |
| 39 | } |
| 40 | } |
| 41 | |
| 42 | /// Result of an attempted terminal transition. |
| 43 | /// |
| 44 | /// The caller needs all three cases distinguished to report a truthful |
| 45 | /// receipt: "I stopped it", "it was already terminal", and "it moved on since |
| 46 | /// you read it, so I refused". Collapsing them into a bool made a concurrent |
| 47 | /// stop by another process indistinguishable from our own transition. |
| 48 | #[derive(Debug, Clone, Copy, PartialEq, Eq)] |
| 49 | pub enum TerminalTransition { |
| 50 | /// This call performed the active -> terminal transition. |
| 51 | Transitioned, |
| 52 | /// The record was already terminal; nothing changed. |
| 53 | AlreadyTerminal, |
| 54 | /// The caller pinned a lifecycle generation and the record has moved on. |
| 55 | /// Nothing was changed and no backend teardown ran. |
| 56 | FenceMismatch { observed: u64 }, |
| 57 | } |
| 58 | |
| 59 | impl TerminalTransition { |
| 60 | #[must_use] |
| 61 | pub const fn transitioned(self) -> bool { |
| 62 | matches!(self, Self::Transitioned) |
| 63 | } |
| 64 | } |
| 65 | |
| 66 | /// One lane record: a running (or completed) workflow instance. |
| 67 | #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)] |
| 68 | pub struct LaneRecord { |
| 69 | pub id: String, |
| 70 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 71 | pub workflow: Option<String>, |
| 72 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 73 | pub fleet: Option<String>, |
| 74 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 75 | pub issue: Option<String>, |
| 76 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 77 | pub goal: Option<String>, |
| 78 | pub runtime: RuntimeBackendKind, |
| 79 | pub status: LaneStatus, |
| 80 | /// Monotonic durable lifecycle sequence used by Work Graph reconciliation. |
| 81 | #[serde(default)] |
| 82 | pub lifecycle_seq: u64, |
| 83 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 84 | pub worktree_path: Option<PathBuf>, |
| 85 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 86 | pub branch: Option<String>, |
| 87 | /// tmux session name when `runtime == tmux`. |
| 88 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 89 | pub tmux_session: Option<String>, |
| 90 | /// Explicit tmux server socket used for this Lane. Pinning the socket keeps |
| 91 | /// start/attach/stop/reconcile in the same server namespace even when |
| 92 | /// `TMUX_TMPDIR` or the caller environment changes between commands. |
| 93 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 94 | pub tmux_socket: Option<PathBuf>, |
| 95 | /// Absolute path to the stream-json / NDJSON journal for this lane. |
| 96 | pub log_path: PathBuf, |
| 97 | pub started_at: String, |
| 98 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 99 | pub stopped_at: Option<String>, |
| 100 | /// Optional human-readable attach target (e.g. `tmux attach -t …`). |
| 101 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 102 | pub attach_target: Option<String>, |
| 103 | /// Worktree cleanup TTL in seconds (None = no auto-cleanup). |
| 104 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 105 | pub worktree_ttl_secs: Option<u64>, |
| 106 | } |
| 107 | |
| 108 | impl LaneRecord { |
| 109 | pub fn new_id() -> String { |
| 110 | let short = uuid::Uuid::new_v4().to_string(); |
| 111 | format!("lane-{}", &short[..8]) |
| 112 | } |
| 113 | |
| 114 | pub fn now_rfc3339() -> String { |
| 115 | Utc::now().to_rfc3339_opts(SecondsFormat::Secs, true) |
| 116 | } |
| 117 | } |
| 118 | |
| 119 | /// Registry root: `$CODEWHALE_HOME/lanes`. |
| 120 | pub fn lanes_dir() -> Result<PathBuf> { |
| 121 | codewhale_config::ensure_state_dir(LANES_SUBDIR) |
| 122 | } |
| 123 | |
| 124 | /// Where the Lane registry *would* live, without creating it. |
| 125 | /// |
| 126 | /// [`lanes_dir`] creates the directory as a side effect, which makes it |
| 127 | /// useless for answering "is there a durable Lane registry?" — a read-only |
| 128 | /// status surface must not conjure the store it is reporting on. Availability |
| 129 | /// probing goes through this instead (see [`crate::control::ControlContext`]). |
| 130 | pub fn lane_registry_root() -> Result<PathBuf> { |
| 131 | Ok(codewhale_config::codewhale_home()?.join(LANES_SUBDIR)) |
| 132 | } |
| 133 | |
| 134 | /// Persist and load lane records. |
| 135 | #[derive(Debug, Clone)] |
| 136 | pub struct LaneRegistry { |
| 137 | root: PathBuf, |
| 138 | } |
| 139 | |
| 140 | impl LaneRegistry { |
| 141 | /// Open the default registry under `$CODEWHALE_HOME/lanes`. |
| 142 | pub fn open_default() -> Result<Self> { |
| 143 | Self::open(lanes_dir()?) |
| 144 | } |
| 145 | |
| 146 | /// Open a registry at an explicit root (tests / custom homes). |
| 147 | pub fn open(root: impl Into<PathBuf>) -> Result<Self> { |
| 148 | let root = root.into(); |
| 149 | fs::create_dir_all(&root) |
| 150 | .with_context(|| format!("create lane registry {}", root.display()))?; |
| 151 | fs::create_dir_all(root.join(LOGS_SUBDIR)) |
| 152 | .with_context(|| format!("create lane logs under {}", root.display()))?; |
| 153 | Ok(Self { root }) |
| 154 | } |
| 155 | |
| 156 | pub fn root(&self) -> &Path { |
| 157 | &self.root |
| 158 | } |
| 159 | |
| 160 | pub fn logs_dir(&self) -> PathBuf { |
| 161 | self.root.join(LOGS_SUBDIR) |
| 162 | } |
| 163 | |
| 164 | pub fn record_path(&self, id: &str) -> PathBuf { |
| 165 | self.root.join(format!("{id}.json")) |
| 166 | } |
| 167 | |
| 168 | pub fn log_path_for(&self, id: &str) -> PathBuf { |
| 169 | self.logs_dir().join(format!("{id}.ndjson")) |
| 170 | } |
| 171 | |
| 172 | pub fn save(&self, record: &LaneRecord) -> Result<()> { |
| 173 | let path = self.record_path(&record.id); |
| 174 | let json = serde_json::to_string_pretty(record).context("serialize lane record")?; |
| 175 | let tmp = path.with_extension("json.tmp"); |
| 176 | fs::write(&tmp, json).with_context(|| format!("write {}", tmp.display()))?; |
| 177 | fs::rename(&tmp, &path).with_context(|| format!("rename {}", path.display()))?; |
| 178 | Ok(()) |
| 179 | } |
| 180 | |
| 181 | pub fn load(&self, id: &str) -> Result<LaneRecord> { |
| 182 | let path = self.record_path(id); |
| 183 | let text = fs::read_to_string(&path) |
| 184 | .with_context(|| format!("read lane record {}", path.display()))?; |
| 185 | serde_json::from_str(&text).with_context(|| format!("parse lane record {}", path.display())) |
| 186 | } |
| 187 | |
| 188 | pub fn list(&self) -> Result<Vec<LaneRecord>> { |
| 189 | let mut records = Vec::new(); |
| 190 | for entry in fs::read_dir(&self.root) |
| 191 | .with_context(|| format!("read lane registry {}", self.root.display()))? |
| 192 | { |
| 193 | let entry = entry?; |
| 194 | let path = entry.path(); |
| 195 | if path.extension().and_then(|e| e.to_str()) != Some("json") { |
| 196 | continue; |
| 197 | } |
| 198 | let text = |
| 199 | fs::read_to_string(&path).with_context(|| format!("read {}", path.display()))?; |
| 200 | match serde_json::from_str::<LaneRecord>(&text) { |
| 201 | Ok(record) => records.push(record), |
| 202 | Err(err) => { |
| 203 | // Skip corrupt records rather than failing the whole list. |
| 204 | eprintln!( |
| 205 | "warning: skip corrupt lane record {}: {err}", |
| 206 | path.display() |
| 207 | ); |
| 208 | } |
| 209 | } |
| 210 | } |
| 211 | records.sort_by(|a, b| b.started_at.cmp(&a.started_at)); |
| 212 | Ok(records) |
| 213 | } |
| 214 | |
| 215 | /// Create a pending lane with log file reserved. |
| 216 | pub fn create_pending( |
| 217 | &self, |
| 218 | workflow: Option<String>, |
| 219 | fleet: Option<String>, |
| 220 | issue: Option<String>, |
| 221 | goal: Option<String>, |
| 222 | runtime: RuntimeBackendKind, |
| 223 | worktree_ttl_secs: Option<u64>, |
| 224 | ) -> Result<LaneRecord> { |
| 225 | let id = LaneRecord::new_id(); |
| 226 | let log_path = self.log_path_for(&id); |
| 227 | // Touch the log so `lane logs` works immediately. |
| 228 | fs::write(&log_path, "").with_context(|| format!("create log {}", log_path.display()))?; |
| 229 | let record = LaneRecord { |
| 230 | id, |
| 231 | workflow, |
| 232 | fleet, |
| 233 | issue, |
| 234 | goal, |
| 235 | runtime, |
| 236 | status: LaneStatus::Pending, |
| 237 | lifecycle_seq: 1, |
| 238 | worktree_path: None, |
| 239 | branch: None, |
| 240 | tmux_session: None, |
| 241 | tmux_socket: None, |
| 242 | log_path, |
| 243 | started_at: LaneRecord::now_rfc3339(), |
| 244 | stopped_at: None, |
| 245 | attach_target: None, |
| 246 | worktree_ttl_secs, |
| 247 | }; |
| 248 | self.save(&record)?; |
| 249 | Ok(record) |
| 250 | } |
| 251 | |
| 252 | /// Atomically promote a Pending Lane to Running. |
| 253 | /// |
| 254 | /// A concurrent `lane stop` is allowed to win while a backend is still |
| 255 | /// launching. In that case this returns `false`, reloads the terminal |
| 256 | /// record, and the backend must tear down any process it just created. |
| 257 | pub fn mark_running_if_pending(&self, record: &mut LaneRecord) -> Result<bool> { |
| 258 | self.mark_running_if_pending_with(record, || Ok(()), || Ok(())) |
| 259 | } |
| 260 | |
| 261 | /// Atomically launch backend state and promote a Pending Lane to Running. |
| 262 | /// |
| 263 | /// The per-Lane lock spans the durable Pending check, `before_transition`, |
| 264 | /// and the Running save. A concurrent stop therefore either wins before |
| 265 | /// launch (and this returns `false` without calling the closure) or waits |
| 266 | /// until the Running record is visible. Backend metadata is first saved in |
| 267 | /// the Pending record so a failed final save still leaves enough data for |
| 268 | /// a later stop. `rollback` is attempted if that final save fails. |
| 269 | pub fn mark_running_if_pending_with<Start, Rollback>( |
| 270 | &self, |
| 271 | record: &mut LaneRecord, |
| 272 | before_transition: Start, |
| 273 | rollback: Rollback, |
| 274 | ) -> Result<bool> |
| 275 | where |
| 276 | Start: FnOnce() -> Result<()>, |
| 277 | Rollback: FnOnce() -> Result<()>, |
| 278 | { |
| 279 | let lock_path = self.root.join(format!("{}.lock", record.id)); |
| 280 | let lock_file = OpenOptions::new() |
| 281 | .create(true) |
| 282 | .truncate(false) |
| 283 | .read(true) |
| 284 | .write(true) |
| 285 | .open(&lock_path) |
| 286 | .with_context(|| format!("open lane lock {}", lock_path.display()))?; |
| 287 | let mut lock = fd_lock::RwLock::new(lock_file); |
| 288 | let _guard = lock |
| 289 | .write() |
| 290 | .with_context(|| format!("lock lane record {}", record.id))?; |
| 291 | |
| 292 | let current = self.load(&record.id)?; |
| 293 | if current.status != LaneStatus::Pending { |
| 294 | *record = current; |
| 295 | return Ok(false); |
| 296 | } |
| 297 | record.lifecycle_seq = current.lifecycle_seq.max(1); |
| 298 | |
| 299 | // `record` carries backend metadata (tmux session, worktree, attach |
| 300 | // target) populated during launch. Persist it while still Pending so |
| 301 | // a failed launch/final save remains discoverable and stoppable. |
| 302 | record.status = LaneStatus::Pending; |
| 303 | record.stopped_at = None; |
| 304 | self.save(record)?; |
| 305 | |
| 306 | before_transition()?; |
| 307 | record.status = LaneStatus::Running; |
| 308 | record.lifecycle_seq = record.lifecycle_seq.saturating_add(1); |
| 309 | record.stopped_at = None; |
| 310 | if let Err(save_error) = self.save(record) { |
| 311 | record.status = LaneStatus::Pending; |
| 312 | record.lifecycle_seq = current.lifecycle_seq.max(1); |
| 313 | if let Err(rollback_error) = rollback() { |
| 314 | return Err(save_error).context(format!( |
| 315 | "persist running Lane; backend rollback also failed: {rollback_error:#}" |
| 316 | )); |
| 317 | } |
| 318 | return Err(save_error).context("persist running Lane after backend launch"); |
| 319 | } |
| 320 | Ok(true) |
| 321 | } |
| 322 | |
| 323 | /// Atomically transition an active Lane to a terminal state. |
| 324 | /// |
| 325 | /// Detached tmux reconciliation, an explicit stop, and a second status |
| 326 | /// reader can race in separate CLI processes. Serialize those terminal |
| 327 | /// decisions on a per-Lane advisory lock, reload the live record under the |
| 328 | /// lock, and only let the first active -> terminal transition win. |
| 329 | pub fn mark_terminal_if_active( |
| 330 | &self, |
| 331 | record: &mut LaneRecord, |
| 332 | status: LaneStatus, |
| 333 | ) -> Result<bool> { |
| 334 | self.mark_terminal_if_active_with(record, status, |_| Ok(())) |
| 335 | } |
| 336 | |
| 337 | /// Atomically perform backend teardown and transition an active Lane. |
| 338 | /// |
| 339 | /// `before_transition` runs while holding the per-Lane lifecycle lock. |
| 340 | /// If teardown fails, the record remains active. This keeps a failed tmux |
| 341 | /// kill from being persisted as Stopped and prevents cleanup racing a |
| 342 | /// concurrent reconciliation decision. |
| 343 | pub fn mark_terminal_if_active_with<F>( |
| 344 | &self, |
| 345 | record: &mut LaneRecord, |
| 346 | status: LaneStatus, |
| 347 | before_transition: F, |
| 348 | ) -> Result<bool> |
| 349 | where |
| 350 | F: FnOnce(&LaneRecord) -> Result<()>, |
| 351 | { |
| 352 | self.mark_terminal_if_active_fenced(record, status, None, before_transition) |
| 353 | .map(TerminalTransition::transitioned) |
| 354 | } |
| 355 | |
| 356 | /// [`mark_terminal_if_active_with`] with an optional lifecycle fence. |
| 357 | /// |
| 358 | /// `expected_lifecycle_seq` is evaluated **after** the live record is |
| 359 | /// reloaded under the per-Lane advisory lock, not by the caller before it. |
| 360 | /// A pre-lock check is a TOCTOU: another process can transition the record |
| 361 | /// between the caller's read and this write, and the caller would then act |
| 362 | /// on a generation it never observed. Checking here means a stale fence |
| 363 | /// refuses without running `before_transition`, so no backend teardown |
| 364 | /// happens for a run the caller did not actually target. |
| 365 | /// |
| 366 | /// [`mark_terminal_if_active_with`]: Self::mark_terminal_if_active_with |
| 367 | pub fn mark_terminal_if_active_fenced<F>( |
| 368 | &self, |
| 369 | record: &mut LaneRecord, |
| 370 | status: LaneStatus, |
| 371 | expected_lifecycle_seq: Option<u64>, |
| 372 | before_transition: F, |
| 373 | ) -> Result<TerminalTransition> |
| 374 | where |
| 375 | F: FnOnce(&LaneRecord) -> Result<()>, |
| 376 | { |
| 377 | if status.is_active() { |
| 378 | bail!("terminal lane transition requires a terminal status"); |
| 379 | } |
| 380 | |
| 381 | let lock_path = self.root.join(format!("{}.lock", record.id)); |
| 382 | let lock_file = OpenOptions::new() |
| 383 | .create(true) |
| 384 | .truncate(false) |
| 385 | .read(true) |
| 386 | .write(true) |
| 387 | .open(&lock_path) |
| 388 | .with_context(|| format!("open lane lock {}", lock_path.display()))?; |
| 389 | let mut lock = fd_lock::RwLock::new(lock_file); |
| 390 | let _guard = lock |
| 391 | .write() |
| 392 | .with_context(|| format!("lock lane record {}", record.id))?; |
| 393 | |
| 394 | let mut current = self.load(&record.id)?; |
| 395 | // Fence first: a mismatched generation must not run backend teardown. |
| 396 | if let Some(expected) = expected_lifecycle_seq |
| 397 | && expected != current.lifecycle_seq |
| 398 | { |
| 399 | let observed = current.lifecycle_seq; |
| 400 | *record = current; |
| 401 | return Ok(TerminalTransition::FenceMismatch { observed }); |
| 402 | } |
| 403 | if !current.status.is_active() { |
| 404 | *record = current; |
| 405 | return Ok(TerminalTransition::AlreadyTerminal); |
| 406 | } |
| 407 | before_transition(¤t)?; |
| 408 | current.lifecycle_seq = current.lifecycle_seq.max(1).saturating_add(1); |
| 409 | current.status = status; |
| 410 | current.stopped_at = Some(LaneRecord::now_rfc3339()); |
| 411 | current.attach_target = None; |
| 412 | self.save(¤t)?; |
| 413 | *record = current; |
| 414 | Ok(TerminalTransition::Transitioned) |
| 415 | } |
| 416 | } |
| 417 | |
| 418 | #[cfg(test)] |
| 419 | mod tests { |
| 420 | use super::*; |
| 421 | use std::sync::atomic::{AtomicUsize, Ordering}; |
| 422 | use std::sync::{Arc, mpsc}; |
| 423 | use tempfile::tempdir; |
| 424 | |
| 425 | #[test] |
| 426 | fn registry_persists_across_open() { |
| 427 | let dir = tempdir().unwrap(); |
| 428 | let reg = LaneRegistry::open(dir.path()).unwrap(); |
| 429 | let record = reg |
| 430 | .create_pending( |
| 431 | Some("stopship".into()), |
| 432 | Some("stopship".into()), |
| 433 | Some("4375".into()), |
| 434 | None, |
| 435 | RuntimeBackendKind::Tmux, |
| 436 | Some(3600), |
| 437 | ) |
| 438 | .unwrap(); |
| 439 | let id = record.id.clone(); |
| 440 | |
| 441 | let reg2 = LaneRegistry::open(dir.path()).unwrap(); |
| 442 | let loaded = reg2.load(&id).unwrap(); |
| 443 | assert_eq!(loaded.workflow.as_deref(), Some("stopship")); |
| 444 | assert_eq!(loaded.fleet.as_deref(), Some("stopship")); |
| 445 | assert_eq!(loaded.issue.as_deref(), Some("4375")); |
| 446 | assert_eq!(loaded.runtime, RuntimeBackendKind::Tmux); |
| 447 | assert_eq!(loaded.status, LaneStatus::Pending); |
| 448 | assert_eq!(loaded.lifecycle_seq, 1); |
| 449 | assert!(loaded.log_path.is_file() || loaded.log_path.exists()); |
| 450 | |
| 451 | let listed = reg2.list().unwrap(); |
| 452 | assert_eq!(listed.len(), 1); |
| 453 | assert_eq!(listed[0].id, id); |
| 454 | } |
| 455 | |
| 456 | #[test] |
| 457 | fn terminal_lane_cannot_launch_after_stop_wins() { |
| 458 | let dir = tempdir().unwrap(); |
| 459 | let reg = LaneRegistry::open(dir.path()).unwrap(); |
| 460 | let mut record = reg |
| 461 | .create_pending(None, None, None, None, RuntimeBackendKind::Tmux, None) |
| 462 | .unwrap(); |
| 463 | assert!( |
| 464 | reg.mark_terminal_if_active(&mut record, LaneStatus::Stopped) |
| 465 | .unwrap() |
| 466 | ); |
| 467 | let starts = AtomicUsize::new(0); |
| 468 | assert!( |
| 469 | !reg.mark_running_if_pending_with( |
| 470 | &mut record, |
| 471 | || { |
| 472 | starts.fetch_add(1, Ordering::SeqCst); |
| 473 | Ok(()) |
| 474 | }, |
| 475 | || Ok(()), |
| 476 | ) |
| 477 | .unwrap() |
| 478 | ); |
| 479 | assert_eq!(starts.load(Ordering::SeqCst), 0); |
| 480 | assert_eq!(record.status, LaneStatus::Stopped); |
| 481 | assert_eq!(record.lifecycle_seq, 2); |
| 482 | let loaded = reg.load(&record.id).unwrap(); |
| 483 | assert_eq!(loaded.status, LaneStatus::Stopped); |
| 484 | assert_eq!(loaded.lifecycle_seq, 2); |
| 485 | } |
| 486 | |
| 487 | #[test] |
| 488 | fn start_and_stop_are_serialized_across_backend_launch() { |
| 489 | let dir = tempdir().unwrap(); |
| 490 | let reg = LaneRegistry::open(dir.path()).unwrap(); |
| 491 | let record = reg |
| 492 | .create_pending(None, None, None, None, RuntimeBackendKind::Tmux, None) |
| 493 | .unwrap(); |
| 494 | let id = record.id.clone(); |
| 495 | let starts = Arc::new(AtomicUsize::new(0)); |
| 496 | let teardowns = Arc::new(AtomicUsize::new(0)); |
| 497 | let (entered_tx, entered_rx) = mpsc::channel(); |
| 498 | let (release_tx, release_rx) = mpsc::channel(); |
| 499 | |
| 500 | let start_reg = reg.clone(); |
| 501 | let start_count = Arc::clone(&starts); |
| 502 | let start_thread = std::thread::spawn(move || { |
| 503 | let mut record = record; |
| 504 | let started = start_reg |
| 505 | .mark_running_if_pending_with( |
| 506 | &mut record, |
| 507 | || { |
| 508 | start_count.fetch_add(1, Ordering::SeqCst); |
| 509 | entered_tx.send(()).unwrap(); |
| 510 | release_rx.recv().unwrap(); |
| 511 | Ok(()) |
| 512 | }, |
| 513 | || Ok(()), |
| 514 | ) |
| 515 | .unwrap(); |
| 516 | assert!(started); |
| 517 | }); |
| 518 | entered_rx.recv().unwrap(); |
| 519 | |
| 520 | let stop_reg = reg.clone(); |
| 521 | let stop_id = id.clone(); |
| 522 | let teardown_count = Arc::clone(&teardowns); |
| 523 | let (stopped_tx, stopped_rx) = mpsc::channel(); |
| 524 | let stop_thread = std::thread::spawn(move || { |
| 525 | let mut record = stop_reg.load(&stop_id).unwrap(); |
| 526 | let stopped = stop_reg |
| 527 | .mark_terminal_if_active_with(&mut record, LaneStatus::Stopped, |_| { |
| 528 | teardown_count.fetch_add(1, Ordering::SeqCst); |
| 529 | Ok(()) |
| 530 | }) |
| 531 | .unwrap(); |
| 532 | stopped_tx.send(stopped).unwrap(); |
| 533 | }); |
| 534 | assert!(matches!( |
| 535 | stopped_rx.try_recv(), |
| 536 | Err(mpsc::TryRecvError::Empty) |
| 537 | )); |
| 538 | release_tx.send(()).unwrap(); |
| 539 | start_thread.join().unwrap(); |
| 540 | assert!(stopped_rx.recv().unwrap()); |
| 541 | stop_thread.join().unwrap(); |
| 542 | |
| 543 | assert_eq!(starts.load(Ordering::SeqCst), 1); |
| 544 | assert_eq!(teardowns.load(Ordering::SeqCst), 1); |
| 545 | let loaded = reg.load(&id).unwrap(); |
| 546 | assert_eq!(loaded.status, LaneStatus::Stopped); |
| 547 | assert_eq!( |
| 548 | loaded.lifecycle_seq, 3, |
| 549 | "pending, running, and stopped are three durable owner states" |
| 550 | ); |
| 551 | } |
| 552 | |
| 553 | #[test] |
| 554 | fn teardown_failure_keeps_durable_lane_active() { |
| 555 | let dir = tempdir().unwrap(); |
| 556 | let reg = LaneRegistry::open(dir.path()).unwrap(); |
| 557 | let mut record = reg |
| 558 | .create_pending(None, None, None, None, RuntimeBackendKind::Tmux, None) |
| 559 | .unwrap(); |
| 560 | assert!(reg.mark_running_if_pending(&mut record).unwrap()); |
| 561 | let error = reg |
| 562 | .mark_terminal_if_active_with(&mut record, LaneStatus::Stopped, |_| { |
| 563 | bail!("backend still alive") |
| 564 | }) |
| 565 | .unwrap_err(); |
| 566 | assert!(error.to_string().contains("backend still alive")); |
| 567 | assert_eq!(record.status, LaneStatus::Running); |
| 568 | assert_eq!(record.lifecycle_seq, 2); |
| 569 | let loaded = reg.load(&record.id).unwrap(); |
| 570 | assert_eq!(loaded.status, LaneStatus::Running); |
| 571 | assert_eq!(loaded.lifecycle_seq, 2); |
| 572 | } |
| 573 | } |
| 574 |