proxmox-rest-server/src/worker_task.rs

   1 use std::collections::{HashMap, VecDeque};
   2 use std::fs::File;
   3 use std::path::PathBuf;
   4 use std::io::{Read, Write, BufRead, BufReader};
   5 use std::panic::UnwindSafe;
   6 use std::sync::atomic::{AtomicBool, Ordering};
   7 use std::sync::{Arc, Mutex};
   8
   9 use anyhow::{bail, format_err, Error};
  10 use futures::*;
  11 use lazy_static::lazy_static;
  12 use serde_json::{json, Value};
  13 use serde::{Serialize, Deserialize};
  14 use tokio::sync::oneshot;
  15 use nix::fcntl::OFlag;
  16 use once_cell::sync::OnceCell;
  17
  18 use proxmox::sys::linux::procfs;
  19 use proxmox::try_block;
  20 use proxmox::tools::fs::{create_path, replace_file, atomic_open_or_create_file, CreateOptions};
  21 use proxmox::api::upid::UPID;
  22
  23 use pbs_tools::task::WorkerTaskContext;
  24 use pbs_tools::logrotate::{LogRotate, LogRotateFiles};
  25
  26 use crate::{CommandoSocket, FileLogger, FileLogOptions};
  27
  28 struct TaskListLockGuard(File);
  29
  30 struct WorkerTaskSetup {
  31     file_opts: CreateOptions,
  32     taskdir: PathBuf,
  33     task_lock_fn: PathBuf,
  34     active_tasks_fn: PathBuf,
  35     task_index_fn: PathBuf,
  36     task_archive_fn: PathBuf,
  37 }
  38
  39 static WORKER_TASK_SETUP: OnceCell<WorkerTaskSetup> = OnceCell::new();
  40
  41 fn worker_task_setup() -> Result<&'static WorkerTaskSetup, Error> {
  42     WORKER_TASK_SETUP.get()
  43         .ok_or_else(|| format_err!("WorkerTask library is not initialized"))
  44 }
  45
  46 impl WorkerTaskSetup {
  47
  48     fn new(basedir: PathBuf, file_opts: CreateOptions) -> Self {
  49
  50         let mut taskdir = basedir.clone();
  51         taskdir.push("tasks");
  52
  53         let mut task_lock_fn = taskdir.clone();
  54         task_lock_fn.push(".active.lock");
  55
  56         let mut active_tasks_fn = taskdir.clone();
  57         active_tasks_fn.push("active");
  58
  59         let mut task_index_fn = taskdir.clone();
  60         task_index_fn.push("index");
  61
  62         let mut task_archive_fn = taskdir.clone();
  63         task_archive_fn.push("archive");
  64
  65         Self {
  66             file_opts,
  67             taskdir,
  68             task_lock_fn,
  69             active_tasks_fn,
  70             task_index_fn,
  71             task_archive_fn,
  72         }
  73     }
  74
  75     fn lock_task_list_files(&self, exclusive: bool) -> Result<TaskListLockGuard, Error> {
  76         let options =  self.file_opts.clone()
  77             .perm(nix::sys::stat::Mode::from_bits_truncate(0o660));
  78
  79         let timeout = std::time::Duration::new(10, 0);
  80
  81         let file = proxmox::tools::fs::open_file_locked(
  82             &self.task_lock_fn,
  83             timeout,
  84             exclusive,
  85             options,
  86         )?;
  87
  88         Ok(TaskListLockGuard(file))
  89     }
  90
  91     fn log_path(&self, upid: &UPID) -> std::path::PathBuf {
  92         let mut path = self.taskdir.clone();
  93         path.push(format!("{:02X}", upid.pstart % 256));
  94         path.push(upid.to_string());
  95         path
  96     }
  97
  98     // atomically read/update the task list, update status of finished tasks
  99     // new_upid is added to the list when specified.
 100     fn update_active_workers(&self, new_upid: Option<&UPID>) -> Result<(), Error> {
 101
 102         let lock = self.lock_task_list_files(true)?;
 103
 104         // TODO remove with 1.x
 105         let mut finish_list: Vec<TaskListInfo> = read_task_file_from_path(&self.task_index_fn)?;
 106         let had_index_file = !finish_list.is_empty();
 107
 108         // We use filter_map because one negative case wants to *move* the data into `finish_list`,
 109         // clippy doesn't quite catch this!
 110         #[allow(clippy::unnecessary_filter_map)]
 111         let mut active_list: Vec<TaskListInfo> = read_task_file_from_path(&self.active_tasks_fn)?
 112             .into_iter()
 113             .filter_map(|info| {
 114                 if info.state.is_some() {
 115                     // this can happen when the active file still includes finished tasks
 116                     finish_list.push(info);
 117                     return None;
 118                 }
 119
 120                 if !worker_is_active_local(&info.upid) {
 121                     // println!("Detected stopped task '{}'", &info.upid_str);
 122                     let now = proxmox::tools::time::epoch_i64();
 123                     let status = upid_read_status(&info.upid).unwrap_or(TaskState::Unknown { endtime: now });
 124                     finish_list.push(TaskListInfo {
 125                         upid: info.upid,
 126                         upid_str: info.upid_str,
 127                         state: Some(status)
 128                     });
 129                     return None;
 130                 }
 131
 132                 Some(info)
 133             }).collect();
 134
 135         if let Some(upid) = new_upid {
 136             active_list.push(TaskListInfo { upid: upid.clone(), upid_str: upid.to_string(), state: None });
 137         }
 138
 139         let active_raw = render_task_list(&active_list);
 140
 141         let options =  self.file_opts.clone()
 142             .perm(nix::sys::stat::Mode::from_bits_truncate(0o660));
 143
 144         replace_file(
 145             &self.active_tasks_fn,
 146             active_raw.as_bytes(),
 147             options,
 148         )?;
 149
 150         finish_list.sort_unstable_by(|a, b| {
 151             match (&a.state, &b.state) {
 152                 (Some(s1), Some(s2)) => s1.cmp(&s2),
 153                 (Some(_), None) => std::cmp::Ordering::Less,
 154                 (None, Some(_)) => std::cmp::Ordering::Greater,
 155                 _ => a.upid.starttime.cmp(&b.upid.starttime),
 156             }
 157         });
 158
 159         if !finish_list.is_empty() {
 160             let options =  self.file_opts.clone()
 161                 .perm(nix::sys::stat::Mode::from_bits_truncate(0o660));
 162
 163             let mut writer = atomic_open_or_create_file(
 164                 &self.task_archive_fn,
 165                 OFlag::O_APPEND | OFlag::O_RDWR,
 166                 &[],
 167                 options,
 168             )?;
 169             for info in &finish_list {
 170                 writer.write_all(render_task_line(&info).as_bytes())?;
 171             }
 172         }
 173
 174         // TODO Remove with 1.x
 175         // for compatibility, if we had an INDEX file, we do not need it anymore
 176         if had_index_file {
 177             let _ = nix::unistd::unlink(&self.task_index_fn);
 178         }
 179
 180         drop(lock);
 181
 182         Ok(())
 183     }
 184
 185     // Create task log directory with correct permissions
 186     fn create_task_log_dirs(&self) -> Result<(), Error> {
 187
 188         try_block!({
 189             let dir_opts = self.file_opts.clone()
 190                 .perm(nix::sys::stat::Mode::from_bits_truncate(0o755));
 191
 192             create_path(&self.taskdir, Some(dir_opts.clone()), Some(dir_opts.clone()))?;
 193             // fixme:??? create_path(pbs_buildcfg::PROXMOX_BACKUP_RUN_DIR, None, Some(opts))?;
 194             Ok(())
 195         }).map_err(|err: Error| format_err!("unable to create task log dir - {}", err))
 196     }
 197 }
 198
 199 /// Initialize the WorkerTask library
 200 pub fn init_worker_tasks(basedir: PathBuf, file_opts: CreateOptions) -> Result<(), Error> {
 201     let setup = WorkerTaskSetup::new(basedir, file_opts);
 202     setup.create_task_log_dirs()?;
 203     WORKER_TASK_SETUP.set(setup)
 204         .map_err(|_| format_err!("init_worker_tasks failed - already initialized"))
 205 }
 206
 207 /// checks if the Task Archive is bigger that 'size_threshold' bytes, and
 208 /// rotates it if it is
 209 pub fn rotate_task_log_archive(size_threshold: u64, compress: bool, max_files: Option<usize>) -> Result<bool, Error> {
 210
 211     let setup = worker_task_setup()?;
 212
 213     let _lock = setup.lock_task_list_files(true)?;
 214
 215     let mut logrotate = LogRotate::new(&setup.task_archive_fn, compress)
 216             .ok_or_else(|| format_err!("could not get archive file names"))?;
 217
 218     logrotate.rotate(size_threshold, None, max_files)
 219 }
 220
 221
 222 /// Path to the worker log file
 223 pub fn upid_log_path(upid: &UPID) -> Result<std::path::PathBuf, Error> {
 224     let setup = worker_task_setup()?;
 225     Ok(setup.log_path(upid))
 226 }
 227
 228 /// Read endtime (time of last log line) and exitstatus from task log file
 229 /// If there is not a single line with at valid datetime, we assume the
 230 /// starttime to be the endtime
 231 pub fn upid_read_status(upid: &UPID) -> Result<TaskState, Error> {
 232
 233     let setup = worker_task_setup()?;
 234
 235     let mut status = TaskState::Unknown { endtime: upid.starttime };
 236
 237     let path = setup.log_path(upid);
 238
 239     let mut file = File::open(path)?;
 240
 241     /// speedup - only read tail
 242     use std::io::Seek;
 243     use std::io::SeekFrom;
 244     let _ = file.seek(SeekFrom::End(-8192)); // ignore errors
 245
 246     let mut data = Vec::with_capacity(8192);
 247     file.read_to_end(&mut data)?;
 248
 249     // strip newlines at the end of the task logs
 250     while data.last() == Some(&b'\n') {
 251         data.pop();
 252     }
 253
 254     let last_line = match data.iter().rposition(|c| *c == b'\n') {
 255         Some(start) if data.len() > (start+1) => &data[start+1..],
 256         Some(_) => &data, // should not happen, since we removed all trailing newlines
 257         None => &data,
 258     };
 259
 260     let last_line = std::str::from_utf8(last_line)
 261         .map_err(|err| format_err!("upid_read_status: utf8 parse failed: {}", err))?;
 262
 263     let mut iter = last_line.splitn(2, ": ");
 264     if let Some(time_str) = iter.next() {
 265         if let Ok(endtime) = proxmox::tools::time::parse_rfc3339(time_str) {
 266             // set the endtime even if we cannot parse the state
 267             status = TaskState::Unknown { endtime };
 268             if let Some(rest) = iter.next().and_then(|rest| rest.strip_prefix("TASK ")) {
 269                 if let Ok(state) = TaskState::from_endtime_and_message(endtime, rest) {
 270                     status = state;
 271                 }
 272             }
 273         }
 274     }
 275
 276     Ok(status)
 277 }
 278
 279 lazy_static! {
 280     static ref WORKER_TASK_LIST: Mutex<HashMap<usize, Arc<WorkerTask>>> = Mutex::new(HashMap::new());
 281 }
 282
 283 /// checks if the task UPID refers to a worker from this process
 284 fn is_local_worker(upid: &UPID) -> bool {
 285     upid.pid == crate::pid() && upid.pstart == crate::pstart()
 286 }
 287
 288 /// Test if the task is still running
 289 pub async fn worker_is_active(upid: &UPID) -> Result<bool, Error> {
 290     if is_local_worker(upid) {
 291         return Ok(WORKER_TASK_LIST.lock().unwrap().contains_key(&upid.task_id));
 292     }
 293
 294     if procfs::check_process_running_pstart(upid.pid, upid.pstart).is_none() {
 295         return Ok(false);
 296     }
 297
 298     let sock = crate::ctrl_sock_from_pid(upid.pid);
 299     let cmd = json!({
 300         "command": "worker-task-status",
 301         "args": {
 302             "upid": upid.to_string(),
 303         },
 304     });
 305     let status = crate::send_command(sock, &cmd).await?;
 306
 307     if let Some(active) = status.as_bool() {
 308         Ok(active)
 309     } else {
 310         bail!("got unexpected result {:?} (expected bool)", status);
 311     }
 312 }
 313
 314 /// Test if the task is still running (fast but inaccurate implementation)
 315 ///
 316 /// If the task is spawned from a different process, we simply return if
 317 /// that process is still running. This information is good enough to detect
 318 /// stale tasks...
 319 pub fn worker_is_active_local(upid: &UPID) -> bool {
 320     if is_local_worker(upid) {
 321         WORKER_TASK_LIST.lock().unwrap().contains_key(&upid.task_id)
 322     } else {
 323         procfs::check_process_running_pstart(upid.pid, upid.pstart).is_some()
 324     }
 325 }
 326
 327 pub fn register_task_control_commands(
 328     commando_sock: &mut CommandoSocket,
 329 ) -> Result<(), Error> {
 330     fn get_upid(args: Option<&Value>) -> Result<UPID, Error> {
 331         let args = if let Some(args) = args { args } else { bail!("missing args") };
 332         let upid = match args.get("upid") {
 333             Some(Value::String(upid)) => upid.parse::<UPID>()?,
 334             None => bail!("no upid in args"),
 335             _ => bail!("unable to parse upid"),
 336         };
 337         if !is_local_worker(&upid) {
 338             bail!("upid does not belong to this process");
 339         }
 340         Ok(upid)
 341     }
 342
 343     commando_sock.register_command("worker-task-abort".into(), move |args| {
 344         let upid = get_upid(args)?;
 345
 346         abort_local_worker(upid);
 347
 348         Ok(Value::Null)
 349     })?;
 350     commando_sock.register_command("worker-task-status".into(), move |args| {
 351         let upid = get_upid(args)?;
 352
 353         let active = WORKER_TASK_LIST.lock().unwrap().contains_key(&upid.task_id);
 354
 355         Ok(active.into())
 356     })?;
 357
 358     Ok(())
 359 }
 360
 361 pub fn abort_worker_async(upid: UPID) {
 362     tokio::spawn(async move {
 363         if let Err(err) = abort_worker(upid).await {
 364             eprintln!("abort worker failed - {}", err);
 365         }
 366     });
 367 }
 368
 369 pub async fn abort_worker(upid: UPID) -> Result<(), Error> {
 370
 371     let sock = crate::ctrl_sock_from_pid(upid.pid);
 372     let cmd = json!({
 373         "command": "worker-task-abort",
 374         "args": {
 375             "upid": upid.to_string(),
 376         },
 377     });
 378     crate::send_command(sock, &cmd).map_ok(|_| ()).await
 379 }
 380
 381 fn parse_worker_status_line(line: &str) -> Result<(String, UPID, Option<TaskState>), Error> {
 382
 383     let data = line.splitn(3, ' ').collect::<Vec<&str>>();
 384
 385     let len = data.len();
 386
 387     match len {
 388         1 => Ok((data[0].to_owned(), data[0].parse::<UPID>()?, None)),
 389         3 => {
 390             let endtime = i64::from_str_radix(data[1], 16)?;
 391             let state = TaskState::from_endtime_and_message(endtime, data[2])?;
 392             Ok((data[0].to_owned(), data[0].parse::<UPID>()?, Some(state)))
 393         }
 394         _ => bail!("wrong number of components"),
 395     }
 396 }
 397
 398 /// Task State
 399 #[derive(Debug, PartialEq, Eq, Serialize, Deserialize)]
 400 pub enum TaskState {
 401     /// The Task ended with an undefined state
 402     Unknown { endtime: i64 },
 403     /// The Task ended and there were no errors or warnings
 404     OK { endtime: i64 },
 405     /// The Task had 'count' amount of warnings and no errors
 406     Warning { count: u64, endtime: i64 },
 407     /// The Task ended with the error described in 'message'
 408     Error { message: String, endtime: i64 },
 409 }
 410
 411 impl TaskState {
 412     pub fn endtime(&self) -> i64 {
 413         match *self {
 414             TaskState::Unknown { endtime } => endtime,
 415             TaskState::OK { endtime } => endtime,
 416             TaskState::Warning { endtime, .. } => endtime,
 417             TaskState::Error { endtime, .. } => endtime,
 418         }
 419     }
 420
 421     fn result_text(&self) -> String {
 422         match self {
 423             TaskState::Error { message, .. } => format!("TASK ERROR: {}", message),
 424             other => format!("TASK {}", other),
 425         }
 426     }
 427
 428     fn from_endtime_and_message(endtime: i64, s: &str) -> Result<Self, Error> {
 429         if s == "unknown" {
 430             Ok(TaskState::Unknown { endtime })
 431         } else if s == "OK" {
 432             Ok(TaskState::OK { endtime })
 433         } else if let Some(warnings) = s.strip_prefix("WARNINGS: ") {
 434             let count: u64 = warnings.parse()?;
 435             Ok(TaskState::Warning{ count, endtime })
 436         } else if !s.is_empty() {
 437             let message = if let Some(err) = s.strip_prefix("ERROR: ") { err } else { s }.to_string();
 438             Ok(TaskState::Error{ message, endtime })
 439         } else {
 440             bail!("unable to parse Task Status '{}'", s);
 441         }
 442     }
 443 }
 444
 445 impl std::cmp::PartialOrd for TaskState {
 446     fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
 447         Some(self.endtime().cmp(&other.endtime()))
 448     }
 449 }
 450
 451 impl std::cmp::Ord for TaskState {
 452     fn cmp(&self, other: &Self) -> std::cmp::Ordering {
 453         self.endtime().cmp(&other.endtime())
 454     }
 455 }
 456
 457 impl std::fmt::Display for TaskState {
 458     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
 459         match self {
 460             TaskState::Unknown { .. } => write!(f, "unknown"),
 461             TaskState::OK { .. }=> write!(f, "OK"),
 462             TaskState::Warning { count, .. } => write!(f, "WARNINGS: {}", count),
 463             TaskState::Error { message, .. } => write!(f, "{}", message),
 464         }
 465     }
 466 }
 467
 468 /// Task details including parsed UPID
 469 ///
 470 /// If there is no `state`, the task is still running.
 471 #[derive(Debug)]
 472 pub struct TaskListInfo {
 473     /// The parsed UPID
 474     pub upid: UPID,
 475     /// UPID string representation
 476     pub upid_str: String,
 477     /// Task `(endtime, status)` if already finished
 478     pub state: Option<TaskState>, // endtime, status
 479 }
 480
 481 fn render_task_line(info: &TaskListInfo) -> String {
 482     let mut raw = String::new();
 483     if let Some(status) = &info.state {
 484         raw.push_str(&format!("{} {:08X} {}\n", info.upid_str, status.endtime(), status));
 485     } else {
 486         raw.push_str(&info.upid_str);
 487         raw.push('\n');
 488     }
 489
 490     raw
 491 }
 492
 493 fn render_task_list(list: &[TaskListInfo]) -> String {
 494     let mut raw = String::new();
 495     for info in list {
 496         raw.push_str(&render_task_line(&info));
 497     }
 498     raw
 499 }
 500
 501 // note this is not locked, caller has to make sure it is
 502 // this will skip (and log) lines that are not valid status lines
 503 fn read_task_file<R: Read>(reader: R) -> Result<Vec<TaskListInfo>, Error>
 504 {
 505     let reader = BufReader::new(reader);
 506     let mut list = Vec::new();
 507     for line in reader.lines() {
 508         let line = line?;
 509         match parse_worker_status_line(&line) {
 510             Ok((upid_str, upid, state)) => list.push(TaskListInfo {
 511                 upid_str,
 512                 upid,
 513                 state
 514             }),
 515             Err(err) => {
 516                 eprintln!("unable to parse worker status '{}' - {}", line, err);
 517                 continue;
 518             }
 519         };
 520     }
 521
 522     Ok(list)
 523 }
 524
 525 // note this is not locked, caller has to make sure it is
 526 fn read_task_file_from_path<P>(path: P) -> Result<Vec<TaskListInfo>, Error>
 527 where
 528     P: AsRef<std::path::Path> + std::fmt::Debug,
 529 {
 530     let file = match File::open(&path) {
 531         Ok(f) => f,
 532         Err(err) if err.kind() == std::io::ErrorKind::NotFound => return Ok(Vec::new()),
 533         Err(err) => bail!("unable to open task list {:?} - {}", path, err),
 534     };
 535
 536     read_task_file(file)
 537 }
 538
 539 pub struct TaskListInfoIterator {
 540     list: VecDeque<TaskListInfo>,
 541     end: bool,
 542     archive: Option<LogRotateFiles>,
 543     lock: Option<TaskListLockGuard>,
 544 }
 545
 546 impl TaskListInfoIterator {
 547     pub fn new(active_only: bool) -> Result<Self, Error> {
 548
 549         let setup = worker_task_setup()?;
 550
 551         let (read_lock, active_list) = {
 552             let lock = setup.lock_task_list_files(false)?;
 553             let active_list = read_task_file_from_path(&setup.active_tasks_fn)?;
 554
 555             let needs_update = active_list
 556                 .iter()
 557                 .any(|info| info.state.is_some() || !worker_is_active_local(&info.upid));
 558
 559             // TODO remove with 1.x
 560             let index_exists = setup.task_index_fn.is_file();
 561
 562             if needs_update || index_exists {
 563                 drop(lock);
 564                 setup.update_active_workers(None)?;
 565                 let lock = setup.lock_task_list_files(false)?;
 566                 let active_list = read_task_file_from_path(&setup.active_tasks_fn)?;
 567                 (lock, active_list)
 568             } else {
 569                 (lock, active_list)
 570             }
 571         };
 572
 573         let archive = if active_only {
 574             None
 575         } else {
 576             let logrotate = LogRotate::new(&setup.task_archive_fn, true)
 577                 .ok_or_else(|| format_err!("could not get archive file names"))?;
 578             Some(logrotate.files())
 579         };
 580
 581         let lock = if active_only { None } else { Some(read_lock) };
 582
 583         Ok(Self {
 584             list: active_list.into(),
 585             end: active_only,
 586             archive,
 587             lock,
 588         })
 589     }
 590 }
 591
 592 impl Iterator for TaskListInfoIterator {
 593     type Item = Result<TaskListInfo, Error>;
 594
 595     fn next(&mut self) -> Option<Self::Item> {
 596         loop {
 597             if let Some(element) = self.list.pop_back() {
 598                 return Some(Ok(element));
 599             } else if self.end {
 600                     return None;
 601             } else {
 602                 if let Some(mut archive) = self.archive.take() {
 603                     if let Some(file) = archive.next() {
 604                         let list = match read_task_file(file) {
 605                             Ok(list) => list,
 606                             Err(err) => return Some(Err(err)),
 607                         };
 608                         self.list.append(&mut list.into());
 609                         self.archive = Some(archive);
 610                         continue;
 611                     }
 612                 }
 613
 614                 self.end = true;
 615                 self.lock.take();
 616             }
 617         }
 618     }
 619 }
 620
 621 /// Launch long running worker tasks.
 622 ///
 623 /// A worker task can either be a whole thread, or a simply tokio
 624 /// task/future. Each task can `log()` messages, which are stored
 625 /// persistently to files. Task should poll the `abort_requested`
 626 /// flag, and stop execution when requested.
 627 pub struct WorkerTask {
 628     setup: &'static WorkerTaskSetup,
 629     upid: UPID,
 630     data: Mutex<WorkerTaskData>,
 631     abort_requested: AtomicBool,
 632 }
 633
 634 impl std::fmt::Display for WorkerTask {
 635
 636     fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
 637         self.upid.fmt(f)
 638     }
 639 }
 640
 641 struct WorkerTaskData {
 642     logger: FileLogger,
 643     progress: f64, // 0..1
 644     warn_count: u64,
 645     pub abort_listeners: Vec<oneshot::Sender<()>>,
 646 }
 647
 648 impl WorkerTask {
 649
 650     pub fn new(
 651         worker_type: &str,
 652         worker_id: Option<String>,
 653         auth_id: String,
 654         to_stdout: bool,
 655     ) -> Result<Arc<Self>, Error> {
 656
 657         let setup = worker_task_setup()?;
 658
 659         let upid = UPID::new(worker_type, worker_id, auth_id)?;
 660         let task_id = upid.task_id;
 661
 662         let mut path = setup.taskdir.clone();
 663
 664         path.push(format!("{:02X}", upid.pstart & 255));
 665
 666         let dir_opts = setup.file_opts.clone()
 667             .perm(nix::sys::stat::Mode::from_bits_truncate(0o755));
 668
 669         create_path(&path, None, Some(dir_opts))?;
 670
 671         path.push(upid.to_string());
 672
 673         let logger_options = FileLogOptions {
 674             to_stdout,
 675             exclusive: true,
 676             prefix_time: true,
 677             read: true,
 678             file_opts: setup.file_opts.clone(),
 679             ..Default::default()
 680         };
 681         let logger = FileLogger::new(&path, logger_options)?;
 682
 683         let worker = Arc::new(Self {
 684             setup,
 685             upid: upid.clone(),
 686             abort_requested: AtomicBool::new(false),
 687             data: Mutex::new(WorkerTaskData {
 688                 logger,
 689                 progress: 0.0,
 690                 warn_count: 0,
 691                 abort_listeners: vec![],
 692             }),
 693         });
 694
 695         // scope to drop the lock again after inserting
 696         {
 697             let mut hash = WORKER_TASK_LIST.lock().unwrap();
 698             hash.insert(task_id, worker.clone());
 699             crate::set_worker_count(hash.len());
 700         }
 701
 702         setup.update_active_workers(Some(&upid))?;
 703
 704         Ok(worker)
 705     }
 706
 707     /// Spawn a new tokio task/future.
 708     pub fn spawn<F, T>(
 709         worker_type: &str,
 710         worker_id: Option<String>,
 711         auth_id: String,
 712         to_stdout: bool,
 713         f: F,
 714     ) -> Result<String, Error>
 715         where F: Send + 'static + FnOnce(Arc<WorkerTask>) -> T,
 716               T: Send + 'static + Future<Output = Result<(), Error>>,
 717     {
 718         let worker = WorkerTask::new(worker_type, worker_id, auth_id, to_stdout)?;
 719         let upid_str = worker.upid.to_string();
 720         let f = f(worker.clone());
 721         tokio::spawn(async move {
 722             let result = f.await;
 723             worker.log_result(&result);
 724         });
 725
 726         Ok(upid_str)
 727     }
 728
 729     /// Create a new worker thread.
 730     pub fn new_thread<F>(
 731         worker_type: &str,
 732         worker_id: Option<String>,
 733         auth_id: String,
 734         to_stdout: bool,
 735         f: F,
 736     ) -> Result<String, Error>
 737         where F: Send + UnwindSafe + 'static + FnOnce(Arc<WorkerTask>) -> Result<(), Error>
 738     {
 739         let worker = WorkerTask::new(worker_type, worker_id, auth_id, to_stdout)?;
 740         let upid_str = worker.upid.to_string();
 741
 742         let _child = std::thread::Builder::new().name(upid_str.clone()).spawn(move || {
 743             let worker1 = worker.clone();
 744             let result = match std::panic::catch_unwind(move || f(worker1)) {
 745                 Ok(r) => r,
 746                 Err(panic) => {
 747                     match panic.downcast::<&str>() {
 748                         Ok(panic_msg) => {
 749                             Err(format_err!("worker panicked: {}", panic_msg))
 750                         }
 751                         Err(_) => {
 752                             Err(format_err!("worker panicked: unknown type."))
 753                         }
 754                     }
 755                 }
 756             };
 757
 758             worker.log_result(&result);
 759         });
 760
 761         Ok(upid_str)
 762     }
 763
 764     /// create state from self and a result
 765     pub fn create_state(&self, result: &Result<(), Error>) -> TaskState {
 766         let warn_count = self.data.lock().unwrap().warn_count;
 767
 768         let endtime = proxmox::tools::time::epoch_i64();
 769
 770         if let Err(err) = result {
 771             TaskState::Error { message: err.to_string(), endtime }
 772         } else if warn_count > 0 {
 773             TaskState::Warning { count: warn_count, endtime }
 774         } else {
 775             TaskState::OK { endtime }
 776         }
 777     }
 778
 779     /// Log task result, remove task from running list
 780     pub fn log_result(&self, result: &Result<(), Error>) {
 781         let state = self.create_state(result);
 782         self.log_message(state.result_text());
 783
 784         WORKER_TASK_LIST.lock().unwrap().remove(&self.upid.task_id);
 785         let _ = self.setup.update_active_workers(None);
 786         crate::set_worker_count(WORKER_TASK_LIST.lock().unwrap().len());
 787     }
 788
 789     /// Log a message.
 790     pub fn log_message<S: AsRef<str>>(&self, msg: S) {
 791         let mut data = self.data.lock().unwrap();
 792         data.logger.log(msg);
 793     }
 794
 795     /// Log a message as warning.
 796     pub fn log_warning<S: AsRef<str>>(&self, msg: S) {
 797         let mut data = self.data.lock().unwrap();
 798         data.logger.log(format!("WARN: {}", msg.as_ref()));
 799         data.warn_count += 1;
 800     }
 801
 802     /// Set progress indicator
 803     pub fn progress(&self, progress: f64) {
 804         if progress >= 0.0 && progress <= 1.0 {
 805             let mut data = self.data.lock().unwrap();
 806             data.progress = progress;
 807         } else {
 808            // fixme:  log!("task '{}': ignoring strange value for progress '{}'", self.upid, progress);
 809         }
 810     }
 811
 812     /// Request abort
 813     pub fn request_abort(&self) {
 814         eprintln!("set abort flag for worker {}", self.upid);
 815
 816         let prev_abort = self.abort_requested.swap(true, Ordering::SeqCst);
 817         if !prev_abort { // log abort one time
 818             self.log_message(format!("received abort request ..."));
 819         }
 820         // noitify listeners
 821         let mut data = self.data.lock().unwrap();
 822         loop {
 823             match data.abort_listeners.pop() {
 824                 None => { break; },
 825                 Some(ch) => {
 826                     let _ = ch.send(()); // ignore errors here
 827                 },
 828             }
 829         }
 830     }
 831
 832     /// Get a future which resolves on task abort
 833     pub fn abort_future(&self) ->  oneshot::Receiver<()> {
 834         let (tx, rx) = oneshot::channel::<()>();
 835
 836         let mut data = self.data.lock().unwrap();
 837         if self.abort_requested() {
 838             let _ = tx.send(());
 839         } else {
 840             data.abort_listeners.push(tx);
 841         }
 842         rx
 843     }
 844
 845     pub fn upid(&self) -> &UPID {
 846         &self.upid
 847     }
 848 }
 849
 850 impl WorkerTaskContext for WorkerTask {
 851
 852     fn abort_requested(&self) -> bool {
 853         self.abort_requested.load(Ordering::SeqCst)
 854     }
 855
 856     fn log(&self, level: log::Level, message: &std::fmt::Arguments) {
 857         match level {
 858             log::Level::Error => self.log_warning(&message.to_string()),
 859             log::Level::Warn => self.log_warning(&message.to_string()),
 860             log::Level::Info => self.log_message(&message.to_string()),
 861             log::Level::Debug => self.log_message(&format!("DEBUG: {}", message)),
 862             log::Level::Trace => self.log_message(&format!("TRACE: {}", message)),
 863         }
 864     }
 865 }
 866
 867 /// Wait for a locally spanned worker task
 868 ///
 869 /// Note: local workers should print logs to stdout, so there is no
 870 /// need to fetch/display logs. We just wait for the worker to finish.
 871 pub async fn wait_for_local_worker(upid_str: &str) -> Result<(), Error> {
 872
 873     let upid: UPID = upid_str.parse()?;
 874
 875     let sleep_duration = core::time::Duration::new(0, 100_000_000);
 876
 877     loop {
 878         if worker_is_active_local(&upid) {
 879             tokio::time::sleep(sleep_duration).await;
 880         } else {
 881             break;
 882         }
 883     }
 884     Ok(())
 885 }
 886
 887 /// Request abort of a local worker (if existing and running)
 888 pub fn abort_local_worker(upid: UPID) {
 889     if let Some(ref worker) = WORKER_TASK_LIST.lock().unwrap().get(&upid.task_id) {
 890         worker.request_abort();
 891     }
 892 }