| 1 | //! Async daemon internals. |
| 2 | //! |
| 3 | //! `run()` is called from the daemonized grandchild process created by |
| 4 | //! `daemon::fork::spawn_daemon`. It boots a multi-threaded Tokio runtime, |
| 5 | //! writes the PID file, binds the Unix socket, and then services IPC |
| 6 | //! connections while a background task monitors managed processes. |
| 7 | |
| 8 | use std::{fs, sync::Arc}; |
| 9 | |
| 10 | use anyhow::Result; |
| 11 | use chrono::Utc; |
| 12 | use tokio::{ |
| 13 | io::{AsyncBufReadExt, AsyncWriteExt, BufReader}, |
| 14 | net::{UnixListener, UnixStream}, |
| 15 | sync::Mutex, |
| 16 | time::{sleep, Duration}, |
| 17 | }; |
| 18 | |
| 19 | use crate::{ |
| 20 | config, |
| 21 | daemon::ipc::{Request, Response}, |
| 22 | process::{runner, store::Store, Process, ProcessStatus}, |
| 23 | }; |
| 24 | |
| 25 | // ─── Public entry point ─────────────────────────────────────────────────────── |
| 26 | |
| 27 | /// Entry point called from `daemon/mod.rs` after the double-fork. |
| 28 | /// |
| 29 | /// Creates a multi-threaded Tokio runtime and drives `async_main` to |
| 30 | /// completion. Never returns — the process terminates via |
| 31 | /// `std::process::exit`. |
| 32 | pub fn run() -> ! { |
| 33 | // Initialise config inside the daemon process (safe to call multiple |
| 34 | // times; subsequent calls are no-ops via OnceCell). |
| 35 | config::init(); |
| 36 | |
| 37 | let rt = tokio::runtime::Builder::new_multi_thread() |
| 38 | .enable_all() |
| 39 | .build() |
| 40 | .expect("Failed to build Tokio runtime"); |
| 41 | |
| 42 | rt.block_on(async { |
| 43 | if let Err(e) = async_main().await { |
| 44 | eprintln!("[proses-daemon] fatal: {:#}", e); |
| 45 | std::process::exit(1); |
| 46 | } |
| 47 | }); |
| 48 | |
| 49 | std::process::exit(0); |
| 50 | } |
| 51 | |
| 52 | // ─── Async entry point ──────────────────────────────────────────────────────── |
| 53 | |
| 54 | async fn async_main() -> Result<()> { |
| 55 | let cfg = config::get(); |
| 56 | |
| 57 | // Write PID file so the CLI can find us. |
| 58 | let pid = std::process::id(); |
| 59 | fs::write(&cfg.pid_path, pid.to_string())?; |
| 60 | |
| 61 | // Remove any stale socket left by a previous crash. |
| 62 | if cfg.sock_path.exists() { |
| 63 | let _ = fs::remove_file(&cfg.sock_path); |
| 64 | } |
| 65 | |
| 66 | // Load (or create) the persistent process store. |
| 67 | let store: Arc<Mutex<Store>> = Arc::new(Mutex::new(Store::load(&cfg.store_path))); |
| 68 | |
| 69 | // Background task: periodically check for dead processes and restart them. |
| 70 | let monitor_store = store.clone(); |
| 71 | tokio::spawn(async move { |
| 72 | monitor_loop(monitor_store).await; |
| 73 | }); |
| 74 | |
| 75 | // Bind the Unix domain socket. |
| 76 | let listener = UnixListener::bind(&cfg.sock_path)?; |
| 77 | |
| 78 | loop { |
| 79 | match listener.accept().await { |
| 80 | Ok((stream, _addr)) => { |
| 81 | let store = store.clone(); |
| 82 | tokio::spawn(async move { |
| 83 | handle_connection(stream, store).await; |
| 84 | }); |
| 85 | } |
| 86 | Err(e) => { |
| 87 | eprintln!("[proses-daemon] accept error: {}", e); |
| 88 | } |
| 89 | } |
| 90 | } |
| 91 | } |
| 92 | |
| 93 | // ─── Monitor loop ───────────────────────────────────────────────────────────── |
| 94 | |
| 95 | /// Wakes every 2 seconds and restarts any `Running` process whose OS PID has |
| 96 | /// died. Processes explicitly stopped by the user (`Stopped` status) are |
| 97 | /// never restarted here. |
| 98 | async fn monitor_loop(store: Arc<Mutex<Store>>) { |
| 99 | loop { |
| 100 | sleep(Duration::from_secs(2)).await; |
| 101 | |
| 102 | let cfg = config::get(); |
| 103 | |
| 104 | // Snapshot the IDs of currently Running processes (short lock hold). |
| 105 | let running_ids: Vec<u32> = { |
| 106 | let s = store.lock().await; |
| 107 | s.processes |
| 108 | .values() |
| 109 | .filter(|p| p.status == ProcessStatus::Running) |
| 110 | .map(|p| p.id) |
| 111 | .collect() |
| 112 | }; |
| 113 | |
| 114 | for id in running_ids { |
| 115 | // Re-check with a fresh lock so we pick up Stop commands that |
| 116 | // arrived since we collected the IDs. |
| 117 | let (pid, restarts, max_restarts) = { |
| 118 | let s = store.lock().await; |
| 119 | match s.processes.get(&id) { |
| 120 | Some(p) if p.status == ProcessStatus::Running => { |
| 121 | (p.pid, p.restarts, p.max_restarts) |
| 122 | } |
| 123 | _ => continue, |
| 124 | } |
| 125 | }; |
| 126 | |
| 127 | // Fast path: the process is still alive. |
| 128 | if runner::is_running(pid) { |
| 129 | continue; |
| 130 | } |
| 131 | |
| 132 | // Process is dead — decide whether to restart or mark Errored. |
| 133 | let mut s = store.lock().await; |
| 134 | let proc = match s.processes.get_mut(&id) { |
| 135 | Some(p) if p.status == ProcessStatus::Running => p, |
| 136 | _ => continue, |
| 137 | }; |
| 138 | |
| 139 | // max_restarts == 0 means unlimited. |
| 140 | if max_restarts == 0 || restarts < max_restarts { |
| 141 | let snapshot = proc.clone(); |
| 142 | match runner::spawn(&snapshot) { |
| 143 | Ok(new_pid) => { |
| 144 | proc.pid = new_pid; |
| 145 | proc.restarts += 1; |
| 146 | proc.started_at = Utc::now(); |
| 147 | proc.status = ProcessStatus::Running; |
| 148 | } |
| 149 | Err(e) => { |
| 150 | eprintln!("[proses-daemon] failed to restart '{}': {}", proc.name, e); |
| 151 | proc.status = ProcessStatus::Errored; |
| 152 | } |
| 153 | } |
| 154 | } else { |
| 155 | eprintln!( |
| 156 | "[proses-daemon] '{}' exceeded max_restarts ({}), marking errored", |
| 157 | proc.name, max_restarts |
| 158 | ); |
| 159 | proc.status = ProcessStatus::Errored; |
| 160 | } |
| 161 | |
| 162 | let _ = s.save(&cfg.store_path); |
| 163 | } |
| 164 | } |
| 165 | } |
| 166 | |
| 167 | // ─── Connection handler ─────────────────────────────────────────────────────── |
| 168 | |
| 169 | /// Read one JSON line from `stream`, dispatch to `handle_request`, and |
| 170 | /// write back a single JSON response line. |
| 171 | async fn handle_connection(stream: UnixStream, store: Arc<Mutex<Store>>) { |
| 172 | let (reader_half, mut writer_half) = stream.into_split(); |
| 173 | let mut reader = BufReader::new(reader_half); |
| 174 | let mut line = String::new(); |
| 175 | |
| 176 | match reader.read_line(&mut line).await { |
| 177 | Ok(0) => return, // EOF — client disconnected without sending anything. |
| 178 | Err(e) => { |
| 179 | eprintln!("[proses-daemon] read error: {}", e); |
| 180 | return; |
| 181 | } |
| 182 | Ok(_) => {} |
| 183 | } |
| 184 | |
| 185 | let response = match serde_json::from_str::<Request>(line.trim()) { |
| 186 | Ok(req) => handle_request(req, store).await, |
| 187 | Err(e) => Response::err(format!("Malformed request: {}", e)), |
| 188 | }; |
| 189 | |
| 190 | let mut json = match serde_json::to_string(&response) { |
| 191 | Ok(s) => s, |
| 192 | Err(_) => r#"{"success":false,"message":"response serialisation error"}"#.to_string(), |
| 193 | }; |
| 194 | json.push('\n'); |
| 195 | |
| 196 | let _ = writer_half.write_all(json.as_bytes()).await; |
| 197 | } |
| 198 | |
| 199 | // ─── Request dispatcher ─────────────────────────────────────────────────────── |
| 200 | |
| 201 | async fn handle_request(req: Request, store: Arc<Mutex<Store>>) -> Response { |
| 202 | match req { |
| 203 | // ── Start ────────────────────────────────────────────────────────── |
| 204 | Request::Start { |
| 205 | name, |
| 206 | command, |
| 207 | cwd, |
| 208 | env, |
| 209 | max_restarts, |
| 210 | } => { |
| 211 | let cfg = config::get(); |
| 212 | let mut s = store.lock().await; |
| 213 | |
| 214 | if s.find(&name).is_some() { |
| 215 | return Response::err(format!( |
| 216 | "A process named '{}' already exists. \ |
| 217 | Use `restart` or `delete` it first.", |
| 218 | name |
| 219 | )); |
| 220 | } |
| 221 | |
| 222 | let id = s.alloc_id(); |
| 223 | let log_out = cfg.log_dir.join(format!("{}-{}.out", name, id)); |
| 224 | let log_err = cfg.log_dir.join(format!("{}-{}.err", name, id)); |
| 225 | |
| 226 | let mut proc = Process { |
| 227 | id, |
| 228 | name: name.clone(), |
| 229 | pid: 0, |
| 230 | command, |
| 231 | cwd: std::path::PathBuf::from(cwd), |
| 232 | env, |
| 233 | status: ProcessStatus::Stopped, |
| 234 | restarts: 0, |
| 235 | max_restarts, |
| 236 | started_at: Utc::now(), |
| 237 | log_out, |
| 238 | log_err, |
| 239 | }; |
| 240 | |
| 241 | match runner::spawn(&proc) { |
| 242 | Ok(pid) => { |
| 243 | proc.pid = pid; |
| 244 | proc.status = ProcessStatus::Running; |
| 245 | proc.started_at = Utc::now(); |
| 246 | let resp = Response::ok(format!("Process '{}' started (pid {})", name, pid)) |
| 247 | .with_process(proc.clone()); |
| 248 | s.processes.insert(id, proc); |
| 249 | let _ = s.save(&cfg.store_path); |
| 250 | resp |
| 251 | } |
| 252 | Err(e) => { |
| 253 | proc.status = ProcessStatus::Errored; |
| 254 | s.processes.insert(id, proc); |
| 255 | let _ = s.save(&cfg.store_path); |
| 256 | Response::err(format!("Failed to start '{}': {}", name, e)) |
| 257 | } |
| 258 | } |
| 259 | } |
| 260 | |
| 261 | // ── Stop ─────────────────────────────────────────────────────────── |
| 262 | Request::Stop { name_or_id } => { |
| 263 | let cfg = config::get(); |
| 264 | let mut s = store.lock().await; |
| 265 | |
| 266 | let id = match s.find_id(&name_or_id) { |
| 267 | Some(id) => id, |
| 268 | None => return Response::err(format!("Process '{}' not found", name_or_id)), |
| 269 | }; |
| 270 | |
| 271 | let pid = s.processes[&id].pid; |
| 272 | let _ = runner::stop(pid); |
| 273 | |
| 274 | if let Some(proc) = s.processes.get_mut(&id) { |
| 275 | proc.status = ProcessStatus::Stopped; |
| 276 | proc.pid = 0; |
| 277 | } |
| 278 | let _ = s.save(&cfg.store_path); |
| 279 | Response::ok(format!("Process '{}' stopped", name_or_id)) |
| 280 | } |
| 281 | |
| 282 | // ── Restart ──────────────────────────────────────────────────────── |
| 283 | // |
| 284 | // Per spec: release the lock, stop the old process, sleep 300 ms, |
| 285 | // respawn, re-acquire the lock to update pid / started_at. |
| 286 | Request::Restart { name_or_id } => { |
| 287 | let cfg = config::get(); |
| 288 | |
| 289 | // Step 1 — snapshot what we need, then release the lock. |
| 290 | let (old_pid, snapshot) = { |
| 291 | let s = store.lock().await; |
| 292 | match s.find(&name_or_id) { |
| 293 | None => return Response::err(format!("Process '{}' not found", name_or_id)), |
| 294 | Some(p) => (p.pid, p.clone()), |
| 295 | } |
| 296 | }; |
| 297 | |
| 298 | // Step 2 — stop old process (lock free). |
| 299 | let _ = runner::stop(old_pid); |
| 300 | |
| 301 | // Step 3 — give the OS a moment to reap the child. |
| 302 | sleep(Duration::from_millis(300)).await; |
| 303 | |
| 304 | // Step 4 — spawn the new instance (lock free). |
| 305 | let new_pid = match runner::spawn(&snapshot) { |
| 306 | Ok(pid) => pid, |
| 307 | Err(e) => { |
| 308 | return Response::err(format!("Failed to restart '{}': {}", name_or_id, e)) |
| 309 | } |
| 310 | }; |
| 311 | |
| 312 | // Step 5 — re-acquire lock and update the store record. |
| 313 | let mut s = store.lock().await; |
| 314 | if let Some(id) = s.find_id(&name_or_id) { |
| 315 | if let Some(proc) = s.processes.get_mut(&id) { |
| 316 | proc.pid = new_pid; |
| 317 | proc.started_at = Utc::now(); |
| 318 | proc.status = ProcessStatus::Running; |
| 319 | } |
| 320 | let _ = s.save(&cfg.store_path); |
| 321 | Response::ok(format!( |
| 322 | "Process '{}' restarted (pid {})", |
| 323 | name_or_id, new_pid |
| 324 | )) |
| 325 | } else { |
| 326 | Response::err(format!("Process '{}' vanished during restart", name_or_id)) |
| 327 | } |
| 328 | } |
| 329 | |
| 330 | // ── Delete ───────────────────────────────────────────────────────── |
| 331 | Request::Delete { name_or_id } => { |
| 332 | let cfg = config::get(); |
| 333 | let mut s = store.lock().await; |
| 334 | |
| 335 | let id = match s.find_id(&name_or_id) { |
| 336 | Some(id) => id, |
| 337 | None => return Response::err(format!("Process '{}' not found", name_or_id)), |
| 338 | }; |
| 339 | |
| 340 | // Stop if running before removing. |
| 341 | let pid_to_stop = s |
| 342 | .processes |
| 343 | .get(&id) |
| 344 | .filter(|p| p.status == ProcessStatus::Running) |
| 345 | .map(|p| p.pid); |
| 346 | |
| 347 | if let Some(pid) = pid_to_stop { |
| 348 | let _ = runner::stop(pid); |
| 349 | } |
| 350 | |
| 351 | s.processes.remove(&id); |
| 352 | let _ = s.save(&cfg.store_path); |
| 353 | Response::ok(format!("Process '{}' deleted", name_or_id)) |
| 354 | } |
| 355 | |
| 356 | // ── List ─────────────────────────────────────────────────────────── |
| 357 | Request::List => { |
| 358 | let s = store.lock().await; |
| 359 | let procs: Vec<Process> = s.processes.values().cloned().collect(); |
| 360 | let count = procs.len(); |
| 361 | Response::ok(format!("{} process(es)", count)).with_processes(procs) |
| 362 | } |
| 363 | |
| 364 | // ── Logs ─────────────────────────────────────────────────────────── |
| 365 | Request::Logs { name_or_id, lines } => { |
| 366 | let log_path = { |
| 367 | let s = store.lock().await; |
| 368 | match s.find(&name_or_id) { |
| 369 | None => return Response::err(format!("Process '{}' not found", name_or_id)), |
| 370 | Some(p) => p.log_out.clone(), |
| 371 | } |
| 372 | }; |
| 373 | |
| 374 | match fs::read_to_string(&log_path) { |
| 375 | Err(e) => Response::err(format!("Cannot read log {:?}: {}", log_path, e)), |
| 376 | Ok(content) => { |
| 377 | let all_lines: Vec<&str> = content.lines().collect(); |
| 378 | let start = all_lines.len().saturating_sub(lines); |
| 379 | let tail = all_lines[start..].join("\n"); |
| 380 | Response::ok(format!("Last {} line(s) of stdout log", lines)).with_logs(tail) |
| 381 | } |
| 382 | } |
| 383 | } |
| 384 | |
| 385 | // ── Show ─────────────────────────────────────────────────────────── |
| 386 | Request::Show { name_or_id } => { |
| 387 | let s = store.lock().await; |
| 388 | match s.find(&name_or_id) { |
| 389 | None => Response::err(format!("Process '{}' not found", name_or_id)), |
| 390 | Some(p) => { |
| 391 | Response::ok(format!("Process '{}'", name_or_id)).with_process(p.clone()) |
| 392 | } |
| 393 | } |
| 394 | } |
| 395 | |
| 396 | // ── Save ─────────────────────────────────────────────────────────── |
| 397 | Request::Save => { |
| 398 | let cfg = config::get(); |
| 399 | let s = store.lock().await; |
| 400 | match s.save(&cfg.store_path) { |
| 401 | Ok(_) => Response::ok("Store saved to disk"), |
| 402 | Err(e) => Response::err(format!("Save failed: {}", e)), |
| 403 | } |
| 404 | } |
| 405 | |
| 406 | // ── Resurrect ────────────────────────────────────────────────────── |
| 407 | // |
| 408 | // Re-spawn all processes whose persisted status is Running but whose |
| 409 | // PID is no longer alive. Useful after a daemon crash/restart. |
| 410 | Request::Resurrect => { |
| 411 | let cfg = config::get(); |
| 412 | let mut s = store.lock().await; |
| 413 | |
| 414 | let dead_running_ids: Vec<u32> = s |
| 415 | .processes |
| 416 | .values() |
| 417 | .filter(|p| p.status == ProcessStatus::Running && !runner::is_running(p.pid)) |
| 418 | .map(|p| p.id) |
| 419 | .collect(); |
| 420 | |
| 421 | let mut spawned = 0u32; |
| 422 | let mut failed = 0u32; |
| 423 | |
| 424 | for id in dead_running_ids { |
| 425 | if let Some(proc) = s.processes.get_mut(&id) { |
| 426 | let snapshot = proc.clone(); |
| 427 | match runner::spawn(&snapshot) { |
| 428 | Ok(new_pid) => { |
| 429 | proc.pid = new_pid; |
| 430 | proc.started_at = Utc::now(); |
| 431 | proc.status = ProcessStatus::Running; |
| 432 | spawned += 1; |
| 433 | } |
| 434 | Err(e) => { |
| 435 | eprintln!( |
| 436 | "[proses-daemon] resurrect failed for '{}': {}", |
| 437 | proc.name, e |
| 438 | ); |
| 439 | proc.status = ProcessStatus::Errored; |
| 440 | failed += 1; |
| 441 | } |
| 442 | } |
| 443 | } |
| 444 | } |
| 445 | |
| 446 | let _ = s.save(&cfg.store_path); |
| 447 | |
| 448 | if failed == 0 { |
| 449 | Response::ok(format!("Resurrected {} process(es)", spawned)) |
| 450 | } else { |
| 451 | Response::err(format!( |
| 452 | "Resurrected {}, failed to spawn {}", |
| 453 | spawned, failed |
| 454 | )) |
| 455 | } |
| 456 | } |
| 457 | |
| 458 | // ── Health ───────────────────────────────────────────────────────── |
| 459 | Request::Health => Response::ok("daemon is healthy"), |
| 460 | |
| 461 | // ── Shutdown ─────────────────────────────────────────────────────── |
| 462 | Request::Shutdown => { |
| 463 | let cfg = config::get(); |
| 464 | |
| 465 | // Stop every running process and persist the final state. |
| 466 | { |
| 467 | let mut s = store.lock().await; |
| 468 | let ids: Vec<u32> = s.processes.keys().cloned().collect(); |
| 469 | for id in ids { |
| 470 | if let Some(proc) = s.processes.get_mut(&id) { |
| 471 | if proc.status == ProcessStatus::Running { |
| 472 | let _ = runner::stop(proc.pid); |
| 473 | proc.status = ProcessStatus::Stopped; |
| 474 | proc.pid = 0; |
| 475 | } |
| 476 | } |
| 477 | } |
| 478 | let _ = s.save(&cfg.store_path); |
| 479 | } |
| 480 | |
| 481 | // Schedule the actual exit slightly after this response is flushed |
| 482 | // so the client always receives the response before the socket |
| 483 | // disappears. |
| 484 | tokio::spawn(async { |
| 485 | sleep(Duration::from_millis(150)).await; |
| 486 | let cfg = config::get(); |
| 487 | let _ = fs::remove_file(&cfg.pid_path); |
| 488 | let _ = fs::remove_file(&cfg.sock_path); |
| 489 | std::process::exit(0); |
| 490 | }); |
| 491 | |
| 492 | Response::ok("Daemon is shutting down") |
| 493 | } |
| 494 | } |
| 495 | } |