]> git.proxmox.com Git - proxmox-backup.git/blob - src/server/worker_task.rs
proxmox-backup-proxy: add task archive rotation
[proxmox-backup.git] / src / server / worker_task.rs
1 use std::collections::{HashMap, VecDeque};
2 use std::fs::File;
3 use std::path::Path;
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 nix::unistd::Pid;
13 use serde_json::{json, Value};
14 use serde::{Serialize, Deserialize};
15 use tokio::sync::oneshot;
16
17 use proxmox::sys::linux::procfs;
18 use proxmox::try_block;
19 use proxmox::tools::fs::{create_path, open_file_locked, replace_file, CreateOptions};
20
21 use super::UPID;
22
23 use crate::tools::logrotate::{LogRotate, LogRotateFiles};
24 use crate::tools::FileLogger;
25 use crate::api2::types::Userid;
26
27 macro_rules! PROXMOX_BACKUP_VAR_RUN_DIR_M { () => ("/run/proxmox-backup") }
28 macro_rules! PROXMOX_BACKUP_LOG_DIR_M { () => ("/var/log/proxmox-backup") }
29 macro_rules! PROXMOX_BACKUP_TASK_DIR_M { () => (concat!( PROXMOX_BACKUP_LOG_DIR_M!(), "/tasks")) }
30
31 pub const PROXMOX_BACKUP_VAR_RUN_DIR: &str = PROXMOX_BACKUP_VAR_RUN_DIR_M!();
32 pub const PROXMOX_BACKUP_LOG_DIR: &str = PROXMOX_BACKUP_LOG_DIR_M!();
33 pub const PROXMOX_BACKUP_TASK_DIR: &str = PROXMOX_BACKUP_TASK_DIR_M!();
34 pub const PROXMOX_BACKUP_TASK_LOCK_FN: &str = concat!(PROXMOX_BACKUP_TASK_DIR_M!(), "/.active.lock");
35 pub const PROXMOX_BACKUP_ACTIVE_TASK_FN: &str = concat!(PROXMOX_BACKUP_TASK_DIR_M!(), "/active");
36 pub const PROXMOX_BACKUP_INDEX_TASK_FN: &str = concat!(PROXMOX_BACKUP_TASK_DIR_M!(), "/index");
37 pub const PROXMOX_BACKUP_ARCHIVE_TASK_FN: &str = concat!(PROXMOX_BACKUP_TASK_DIR_M!(), "/archive");
38
39 const MAX_INDEX_TASKS: usize = 1000;
40
41 lazy_static! {
42 static ref WORKER_TASK_LIST: Mutex<HashMap<usize, Arc<WorkerTask>>> = Mutex::new(HashMap::new());
43
44 static ref MY_PID: i32 = unsafe { libc::getpid() };
45 static ref MY_PID_PSTART: u64 = procfs::PidStat::read_from_pid(Pid::from_raw(*MY_PID))
46 .unwrap()
47 .starttime;
48 }
49
50 /// Test if the task is still running
51 pub async fn worker_is_active(upid: &UPID) -> Result<bool, Error> {
52 if (upid.pid == *MY_PID) && (upid.pstart == *MY_PID_PSTART) {
53 return Ok(WORKER_TASK_LIST.lock().unwrap().contains_key(&upid.task_id));
54 }
55
56 if !procfs::check_process_running_pstart(upid.pid, upid.pstart).is_some() {
57 return Ok(false);
58 }
59
60 let socketname = format!(
61 "\0{}/proxmox-task-control-{}.sock", PROXMOX_BACKUP_VAR_RUN_DIR, upid.pid);
62
63 let cmd = json!({
64 "command": "status",
65 "upid": upid.to_string(),
66 });
67
68 let status = super::send_command(socketname, cmd).await?;
69
70 if let Some(active) = status.as_bool() {
71 Ok(active)
72 } else {
73 bail!("got unexpected result {:?} (expected bool)", status);
74 }
75 }
76
77 /// Test if the task is still running (fast but inaccurate implementation)
78 ///
79 /// If the task is spanned from a different process, we simply return if
80 /// that process is still running. This information is good enough to detect
81 /// stale tasks...
82 pub fn worker_is_active_local(upid: &UPID) -> bool {
83 if (upid.pid == *MY_PID) && (upid.pstart == *MY_PID_PSTART) {
84 WORKER_TASK_LIST.lock().unwrap().contains_key(&upid.task_id)
85 } else {
86 procfs::check_process_running_pstart(upid.pid, upid.pstart).is_some()
87 }
88 }
89
90 pub fn create_task_control_socket() -> Result<(), Error> {
91
92 let socketname = format!(
93 "\0{}/proxmox-task-control-{}.sock", PROXMOX_BACKUP_VAR_RUN_DIR, *MY_PID);
94
95 let control_future = super::create_control_socket(socketname, |param| {
96 let param = param.as_object()
97 .ok_or_else(|| format_err!("unable to parse parameters (expected json object)"))?;
98 if param.keys().count() != 2 { bail!("wrong number of parameters"); }
99
100 let command = param["command"].as_str()
101 .ok_or_else(|| format_err!("unable to parse parameters (missing command)"))?;
102
103 // we have only two commands for now
104 if !(command == "abort-task" || command == "status") { bail!("got unknown command '{}'", command); }
105
106 let upid_str = param["upid"].as_str()
107 .ok_or_else(|| format_err!("unable to parse parameters (missing upid)"))?;
108
109 let upid = upid_str.parse::<UPID>()?;
110
111 if !((upid.pid == *MY_PID) && (upid.pstart == *MY_PID_PSTART)) {
112 bail!("upid does not belong to this process");
113 }
114
115 let hash = WORKER_TASK_LIST.lock().unwrap();
116
117 match command {
118 "abort-task" => {
119 if let Some(ref worker) = hash.get(&upid.task_id) {
120 worker.request_abort();
121 } else {
122 // assume task is already stopped
123 }
124 Ok(Value::Null)
125 }
126 "status" => {
127 let active = hash.contains_key(&upid.task_id);
128 Ok(active.into())
129 }
130 _ => {
131 bail!("got unknown command '{}'", command);
132 }
133 }
134 })?;
135
136 tokio::spawn(control_future);
137
138 Ok(())
139 }
140
141 pub fn abort_worker_async(upid: UPID) {
142 tokio::spawn(async move {
143 if let Err(err) = abort_worker(upid).await {
144 eprintln!("abort worker failed - {}", err);
145 }
146 });
147 }
148
149 pub async fn abort_worker(upid: UPID) -> Result<(), Error> {
150
151 let target_pid = upid.pid;
152
153 let socketname = format!(
154 "\0{}/proxmox-task-control-{}.sock", PROXMOX_BACKUP_VAR_RUN_DIR, target_pid);
155
156 let cmd = json!({
157 "command": "abort-task",
158 "upid": upid.to_string(),
159 });
160
161 super::send_command(socketname, cmd).map_ok(|_| ()).await
162 }
163
164 fn parse_worker_status_line(line: &str) -> Result<(String, UPID, Option<TaskState>), Error> {
165
166 let data = line.splitn(3, ' ').collect::<Vec<&str>>();
167
168 let len = data.len();
169
170 match len {
171 1 => Ok((data[0].to_owned(), data[0].parse::<UPID>()?, None)),
172 3 => {
173 let endtime = i64::from_str_radix(data[1], 16)?;
174 let state = TaskState::from_endtime_and_message(endtime, data[2])?;
175 Ok((data[0].to_owned(), data[0].parse::<UPID>()?, Some(state)))
176 }
177 _ => bail!("wrong number of components"),
178 }
179 }
180
181 /// Create task log directory with correct permissions
182 pub fn create_task_log_dirs() -> Result<(), Error> {
183
184 try_block!({
185 let backup_user = crate::backup::backup_user()?;
186 let opts = CreateOptions::new()
187 .owner(backup_user.uid)
188 .group(backup_user.gid);
189
190 create_path(PROXMOX_BACKUP_LOG_DIR, None, Some(opts.clone()))?;
191 create_path(PROXMOX_BACKUP_TASK_DIR, None, Some(opts.clone()))?;
192 create_path(PROXMOX_BACKUP_VAR_RUN_DIR, None, Some(opts))?;
193 Ok(())
194 }).map_err(|err: Error| format_err!("unable to create task log dir - {}", err))?;
195
196 Ok(())
197 }
198
199 /// Read endtime (time of last log line) and exitstatus from task log file
200 /// If there is not a single line with at valid datetime, we assume the
201 /// starttime to be the endtime
202 pub fn upid_read_status(upid: &UPID) -> Result<TaskState, Error> {
203
204 let mut status = TaskState::Unknown { endtime: upid.starttime };
205
206 let path = upid.log_path();
207
208 let mut file = File::open(path)?;
209
210 /// speedup - only read tail
211 use std::io::Seek;
212 use std::io::SeekFrom;
213 let _ = file.seek(SeekFrom::End(-8192)); // ignore errors
214
215 let mut data = Vec::with_capacity(8192);
216 file.read_to_end(&mut data)?;
217
218 // task logs should end with newline, we do not want it here
219 if data.len() > 0 && data[data.len()-1] == b'\n' {
220 data.pop();
221 }
222
223 let last_line = {
224 let mut start = 0;
225 for pos in (0..data.len()).rev() {
226 if data[pos] == b'\n' {
227 start = data.len().min(pos + 1);
228 break;
229 }
230 }
231 &data[start..]
232 };
233
234 let last_line = std::str::from_utf8(last_line)
235 .map_err(|err| format_err!("upid_read_status: utf8 parse failed: {}", err))?;
236
237 let mut iter = last_line.splitn(2, ": ");
238 if let Some(time_str) = iter.next() {
239 if let Ok(endtime) = proxmox::tools::time::parse_rfc3339(time_str) {
240 if let Some(rest) = iter.next().and_then(|rest| rest.strip_prefix("TASK ")) {
241 if let Ok(state) = TaskState::from_endtime_and_message(endtime, rest) {
242 status = state;
243 }
244 }
245 }
246 }
247
248 Ok(status)
249 }
250
251 /// Task State
252 #[derive(Debug, PartialEq, Eq, Serialize, Deserialize)]
253 pub enum TaskState {
254 /// The Task ended with an undefined state
255 Unknown { endtime: i64 },
256 /// The Task ended and there were no errors or warnings
257 OK { endtime: i64 },
258 /// The Task had 'count' amount of warnings and no errors
259 Warning { count: u64, endtime: i64 },
260 /// The Task ended with the error described in 'message'
261 Error { message: String, endtime: i64 },
262 }
263
264 impl TaskState {
265 pub fn endtime(&self) -> i64 {
266 match *self {
267 TaskState::Unknown { endtime } => endtime,
268 TaskState::OK { endtime } => endtime,
269 TaskState::Warning { endtime, .. } => endtime,
270 TaskState::Error { endtime, .. } => endtime,
271 }
272 }
273
274 fn result_text(&self) -> String {
275 match self {
276 TaskState::Error { message, .. } => format!("TASK ERROR: {}", message),
277 other => format!("TASK {}", other),
278 }
279 }
280
281 fn from_endtime_and_message(endtime: i64, s: &str) -> Result<Self, Error> {
282 if s == "unknown" {
283 Ok(TaskState::Unknown { endtime })
284 } else if s == "OK" {
285 Ok(TaskState::OK { endtime })
286 } else if s.starts_with("WARNINGS: ") {
287 let count: u64 = s[10..].parse()?;
288 Ok(TaskState::Warning{ count, endtime })
289 } else if s.len() > 0 {
290 let message = if s.starts_with("ERROR: ") { &s[7..] } else { s }.to_string();
291 Ok(TaskState::Error{ message, endtime })
292 } else {
293 bail!("unable to parse Task Status '{}'", s);
294 }
295 }
296 }
297
298 impl std::cmp::PartialOrd for TaskState {
299 fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
300 Some(self.endtime().cmp(&other.endtime()))
301 }
302 }
303
304 impl std::cmp::Ord for TaskState {
305 fn cmp(&self, other: &Self) -> std::cmp::Ordering {
306 self.endtime().cmp(&other.endtime())
307 }
308 }
309
310 impl std::fmt::Display for TaskState {
311 fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
312 match self {
313 TaskState::Unknown { .. } => write!(f, "unknown"),
314 TaskState::OK { .. }=> write!(f, "OK"),
315 TaskState::Warning { count, .. } => write!(f, "WARNINGS: {}", count),
316 TaskState::Error { message, .. } => write!(f, "{}", message),
317 }
318 }
319 }
320
321 /// Task details including parsed UPID
322 ///
323 /// If there is no `state`, the task is still running.
324 #[derive(Debug)]
325 pub struct TaskListInfo {
326 /// The parsed UPID
327 pub upid: UPID,
328 /// UPID string representation
329 pub upid_str: String,
330 /// Task `(endtime, status)` if already finished
331 pub state: Option<TaskState>, // endtime, status
332 }
333
334 fn lock_task_list_files(exclusive: bool) -> Result<std::fs::File, Error> {
335 let backup_user = crate::backup::backup_user()?;
336
337 let lock = open_file_locked(PROXMOX_BACKUP_TASK_LOCK_FN, std::time::Duration::new(10, 0), exclusive)?;
338 nix::unistd::chown(PROXMOX_BACKUP_TASK_LOCK_FN, Some(backup_user.uid), Some(backup_user.gid))?;
339
340 Ok(lock)
341 }
342
343 /// checks if the Task Archive is bigger that 'size_threshold' bytes, and
344 /// rotates it if it is
345 pub fn rotate_task_log_archive(size_threshold: u64, compress: bool, max_files: Option<usize>) -> Result<bool, Error> {
346 let _lock = lock_task_list_files(true)?;
347 let path = Path::new(PROXMOX_BACKUP_ARCHIVE_TASK_FN);
348 let metadata = path.metadata()?;
349 if metadata.len() > size_threshold {
350 let mut logrotate = LogRotate::new(PROXMOX_BACKUP_ARCHIVE_TASK_FN, compress).ok_or_else(|| format_err!("could not get archive file names"))?;
351 let backup_user = crate::backup::backup_user()?;
352 logrotate.rotate(
353 CreateOptions::new()
354 .owner(backup_user.uid)
355 .group(backup_user.gid),
356 max_files,
357 )?;
358 Ok(true)
359 } else {
360 Ok(false)
361 }
362 }
363
364 // atomically read/update the task list, update status of finished tasks
365 // new_upid is added to the list when specified.
366 fn update_active_workers(new_upid: Option<&UPID>) -> Result<(), Error> {
367
368 let backup_user = crate::backup::backup_user()?;
369
370 let lock = lock_task_list_files(true)?;
371
372 let mut finish_list: Vec<TaskListInfo> = read_task_file_from_path(PROXMOX_BACKUP_INDEX_TASK_FN)?;
373 let mut active_list: Vec<TaskListInfo> = read_task_file_from_path(PROXMOX_BACKUP_ACTIVE_TASK_FN)?
374 .into_iter()
375 .filter_map(|info| {
376 if info.state.is_some() {
377 // this can happen when the active file still includes finished tasks
378 finish_list.push(info);
379 return None;
380 }
381
382 if !worker_is_active_local(&info.upid) {
383 println!("Detected stopped UPID {}", &info.upid_str);
384 let now = proxmox::tools::time::epoch_i64();
385 let status = upid_read_status(&info.upid)
386 .unwrap_or_else(|_| TaskState::Unknown { endtime: now });
387 finish_list.push(TaskListInfo {
388 upid: info.upid,
389 upid_str: info.upid_str,
390 state: Some(status)
391 });
392 return None;
393 }
394
395 Some(info)
396 }).collect();
397
398 if let Some(upid) = new_upid {
399 active_list.push(TaskListInfo { upid: upid.clone(), upid_str: upid.to_string(), state: None });
400 }
401
402 let active_raw = render_task_list(&active_list);
403
404 replace_file(
405 PROXMOX_BACKUP_ACTIVE_TASK_FN,
406 active_raw.as_bytes(),
407 CreateOptions::new()
408 .owner(backup_user.uid)
409 .group(backup_user.gid),
410 )?;
411
412 finish_list.sort_unstable_by(|a, b| {
413 match (&a.state, &b.state) {
414 (Some(s1), Some(s2)) => s1.cmp(&s2),
415 (Some(_), None) => std::cmp::Ordering::Less,
416 (None, Some(_)) => std::cmp::Ordering::Greater,
417 _ => a.upid.starttime.cmp(&b.upid.starttime),
418 }
419 });
420
421 let start = (finish_list.len()-MAX_INDEX_TASKS).max(0);
422 let end = (start+MAX_INDEX_TASKS).min(finish_list.len());
423 let index_raw = render_task_list(&finish_list[start..end]);
424
425 replace_file(
426 PROXMOX_BACKUP_INDEX_TASK_FN,
427 index_raw.as_bytes(),
428 CreateOptions::new()
429 .owner(backup_user.uid)
430 .group(backup_user.gid),
431 )?;
432
433 if !finish_list.is_empty() && start > 0 {
434 match std::fs::OpenOptions::new().append(true).create(true).open(PROXMOX_BACKUP_ARCHIVE_TASK_FN) {
435 Ok(mut writer) => {
436 for info in &finish_list[0..start] {
437 writer.write_all(render_task_line(&info).as_bytes())?;
438 }
439 },
440 Err(err) => bail!("could not write task archive - {}", err),
441 }
442
443 nix::unistd::chown(PROXMOX_BACKUP_ARCHIVE_TASK_FN, Some(backup_user.uid), Some(backup_user.gid))?;
444 }
445
446 drop(lock);
447
448 Ok(())
449 }
450
451 fn render_task_line(info: &TaskListInfo) -> String {
452 let mut raw = String::new();
453 if let Some(status) = &info.state {
454 raw.push_str(&format!("{} {:08X} {}\n", info.upid_str, status.endtime(), status));
455 } else {
456 raw.push_str(&info.upid_str);
457 raw.push('\n');
458 }
459
460 raw
461 }
462
463 fn render_task_list(list: &[TaskListInfo]) -> String {
464 let mut raw = String::new();
465 for info in list {
466 raw.push_str(&render_task_line(&info));
467 }
468 raw
469 }
470
471 // note this is not locked, caller has to make sure it is
472 // this will skip (and log) lines that are not valid status lines
473 fn read_task_file<R: Read>(reader: R) -> Result<Vec<TaskListInfo>, Error>
474 {
475 let reader = BufReader::new(reader);
476 let mut list = Vec::new();
477 for line in reader.lines() {
478 let line = line?;
479 match parse_worker_status_line(&line) {
480 Ok((upid_str, upid, state)) => list.push(TaskListInfo {
481 upid_str,
482 upid,
483 state
484 }),
485 Err(err) => {
486 eprintln!("unable to parse worker status '{}' - {}", line, err);
487 continue;
488 }
489 };
490 }
491
492 Ok(list)
493 }
494
495 // note this is not locked, caller has to make sure it is
496 fn read_task_file_from_path<P>(path: P) -> Result<Vec<TaskListInfo>, Error>
497 where
498 P: AsRef<std::path::Path> + std::fmt::Debug,
499 {
500 let file = match File::open(&path) {
501 Ok(f) => f,
502 Err(err) if err.kind() == std::io::ErrorKind::NotFound => return Ok(Vec::new()),
503 Err(err) => bail!("unable to open task list {:?} - {}", path, err),
504 };
505
506 read_task_file(file)
507 }
508
509 enum TaskFile {
510 Active,
511 Index,
512 Archive,
513 End,
514 }
515
516 pub struct TaskListInfoIterator {
517 list: VecDeque<TaskListInfo>,
518 file: TaskFile,
519 archive: Option<LogRotateFiles>,
520 lock: Option<File>,
521 }
522
523 impl TaskListInfoIterator {
524 pub fn new(active_only: bool) -> Result<Self, Error> {
525 let (read_lock, active_list) = {
526 let lock = lock_task_list_files(false)?;
527 let active_list = read_task_file_from_path(PROXMOX_BACKUP_ACTIVE_TASK_FN)?;
528
529 let needs_update = active_list
530 .iter()
531 .any(|info| info.state.is_none() && !worker_is_active_local(&info.upid));
532
533 if needs_update {
534 drop(lock);
535 update_active_workers(None)?;
536 let lock = lock_task_list_files(false)?;
537 let active_list = read_task_file_from_path(PROXMOX_BACKUP_ACTIVE_TASK_FN)?;
538 (lock, active_list)
539 } else {
540 (lock, active_list)
541 }
542 };
543
544 let archive = if active_only {
545 None
546 } else {
547 let logrotate = LogRotate::new(PROXMOX_BACKUP_ARCHIVE_TASK_FN, true).ok_or_else(|| format_err!("could not get archive file names"))?;
548 Some(logrotate.files())
549 };
550
551 let file = if active_only { TaskFile::End } else { TaskFile::Active };
552 let lock = if active_only { None } else { Some(read_lock) };
553
554 Ok(Self {
555 list: active_list.into(),
556 file,
557 archive,
558 lock,
559 })
560 }
561 }
562
563 impl Iterator for TaskListInfoIterator {
564 type Item = Result<TaskListInfo, Error>;
565
566 fn next(&mut self) -> Option<Self::Item> {
567 loop {
568 if let Some(element) = self.list.pop_back() {
569 return Some(Ok(element));
570 } else {
571 match self.file {
572 TaskFile::Active => {
573 let index = match read_task_file_from_path(PROXMOX_BACKUP_INDEX_TASK_FN) {
574 Ok(index) => index,
575 Err(err) => return Some(Err(err)),
576 };
577 self.list.append(&mut index.into());
578 self.file = TaskFile::Index;
579 },
580 TaskFile::Index | TaskFile::Archive => {
581 if let Some(mut archive) = self.archive.take() {
582 if let Some(file) = archive.next() {
583 let list = match read_task_file(file) {
584 Ok(list) => list,
585 Err(err) => return Some(Err(err)),
586 };
587 self.list.append(&mut list.into());
588 self.archive = Some(archive);
589 self.file = TaskFile::Archive;
590 continue;
591 }
592 }
593 self.file = TaskFile::End;
594 self.lock.take();
595 return None;
596 }
597 TaskFile::End => return None,
598 }
599 }
600 }
601 }
602 }
603
604 /// Launch long running worker tasks.
605 ///
606 /// A worker task can either be a whole thread, or a simply tokio
607 /// task/future. Each task can `log()` messages, which are stored
608 /// persistently to files. Task should poll the `abort_requested`
609 /// flag, and stop execution when requested.
610 #[derive(Debug)]
611 pub struct WorkerTask {
612 upid: UPID,
613 data: Mutex<WorkerTaskData>,
614 abort_requested: AtomicBool,
615 }
616
617 impl std::fmt::Display for WorkerTask {
618
619 fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
620 self.upid.fmt(f)
621 }
622 }
623
624 #[derive(Debug)]
625 struct WorkerTaskData {
626 logger: FileLogger,
627 progress: f64, // 0..1
628 warn_count: u64,
629 pub abort_listeners: Vec<oneshot::Sender<()>>,
630 }
631
632 impl Drop for WorkerTask {
633
634 fn drop(&mut self) {
635 println!("unregister worker");
636 }
637 }
638
639 impl WorkerTask {
640
641 pub fn new(worker_type: &str, worker_id: Option<String>, userid: Userid, to_stdout: bool) -> Result<Arc<Self>, Error> {
642 println!("register worker");
643
644 let upid = UPID::new(worker_type, worker_id, userid)?;
645 let task_id = upid.task_id;
646
647 let mut path = std::path::PathBuf::from(PROXMOX_BACKUP_TASK_DIR);
648
649 path.push(format!("{:02X}", upid.pstart % 256));
650
651 let backup_user = crate::backup::backup_user()?;
652
653 create_path(&path, None, Some(CreateOptions::new().owner(backup_user.uid).group(backup_user.gid)))?;
654
655 path.push(upid.to_string());
656
657 println!("FILE: {:?}", path);
658
659 let logger = FileLogger::new(&path, to_stdout)?;
660 nix::unistd::chown(&path, Some(backup_user.uid), Some(backup_user.gid))?;
661
662 let worker = Arc::new(Self {
663 upid: upid.clone(),
664 abort_requested: AtomicBool::new(false),
665 data: Mutex::new(WorkerTaskData {
666 logger,
667 progress: 0.0,
668 warn_count: 0,
669 abort_listeners: vec![],
670 }),
671 });
672
673 // scope to drop the lock again after inserting
674 {
675 let mut hash = WORKER_TASK_LIST.lock().unwrap();
676 hash.insert(task_id, worker.clone());
677 super::set_worker_count(hash.len());
678 }
679
680 update_active_workers(Some(&upid))?;
681
682 Ok(worker)
683 }
684
685 /// Spawn a new tokio task/future.
686 pub fn spawn<F, T>(
687 worker_type: &str,
688 worker_id: Option<String>,
689 userid: Userid,
690 to_stdout: bool,
691 f: F,
692 ) -> Result<String, Error>
693 where F: Send + 'static + FnOnce(Arc<WorkerTask>) -> T,
694 T: Send + 'static + Future<Output = Result<(), Error>>,
695 {
696 let worker = WorkerTask::new(worker_type, worker_id, userid, to_stdout)?;
697 let upid_str = worker.upid.to_string();
698 let f = f(worker.clone());
699 tokio::spawn(async move {
700 let result = f.await;
701 worker.log_result(&result);
702 });
703
704 Ok(upid_str)
705 }
706
707 /// Create a new worker thread.
708 pub fn new_thread<F>(
709 worker_type: &str,
710 worker_id: Option<String>,
711 userid: Userid,
712 to_stdout: bool,
713 f: F,
714 ) -> Result<String, Error>
715 where F: Send + UnwindSafe + 'static + FnOnce(Arc<WorkerTask>) -> Result<(), Error>
716 {
717 println!("register worker thread");
718
719 let worker = WorkerTask::new(worker_type, worker_id, userid, to_stdout)?;
720 let upid_str = worker.upid.to_string();
721
722 let _child = std::thread::Builder::new().name(upid_str.clone()).spawn(move || {
723 let worker1 = worker.clone();
724 let result = match std::panic::catch_unwind(move || f(worker1)) {
725 Ok(r) => r,
726 Err(panic) => {
727 match panic.downcast::<&str>() {
728 Ok(panic_msg) => {
729 Err(format_err!("worker panicked: {}", panic_msg))
730 }
731 Err(_) => {
732 Err(format_err!("worker panicked: unknown type."))
733 }
734 }
735 }
736 };
737
738 worker.log_result(&result);
739 });
740
741 Ok(upid_str)
742 }
743
744 /// create state from self and a result
745 pub fn create_state(&self, result: &Result<(), Error>) -> TaskState {
746 let warn_count = self.data.lock().unwrap().warn_count;
747
748 let endtime = proxmox::tools::time::epoch_i64();
749
750 if let Err(err) = result {
751 TaskState::Error { message: err.to_string(), endtime }
752 } else if warn_count > 0 {
753 TaskState::Warning { count: warn_count, endtime }
754 } else {
755 TaskState::OK { endtime }
756 }
757 }
758
759 /// Log task result, remove task from running list
760 pub fn log_result(&self, result: &Result<(), Error>) {
761 let state = self.create_state(result);
762 self.log(state.result_text());
763
764 WORKER_TASK_LIST.lock().unwrap().remove(&self.upid.task_id);
765 let _ = update_active_workers(None);
766 super::set_worker_count(WORKER_TASK_LIST.lock().unwrap().len());
767 }
768
769 /// Log a message.
770 pub fn log<S: AsRef<str>>(&self, msg: S) {
771 let mut data = self.data.lock().unwrap();
772 data.logger.log(msg);
773 }
774
775 /// Log a message as warning.
776 pub fn warn<S: AsRef<str>>(&self, msg: S) {
777 let mut data = self.data.lock().unwrap();
778 data.logger.log(format!("WARN: {}", msg.as_ref()));
779 data.warn_count += 1;
780 }
781
782 /// Set progress indicator
783 pub fn progress(&self, progress: f64) {
784 if progress >= 0.0 && progress <= 1.0 {
785 let mut data = self.data.lock().unwrap();
786 data.progress = progress;
787 } else {
788 // fixme: log!("task '{}': ignoring strange value for progress '{}'", self.upid, progress);
789 }
790 }
791
792 /// Request abort
793 pub fn request_abort(&self) {
794 eprintln!("set abort flag for worker {}", self.upid);
795 self.abort_requested.store(true, Ordering::SeqCst);
796 // noitify listeners
797 let mut data = self.data.lock().unwrap();
798 loop {
799 match data.abort_listeners.pop() {
800 None => { break; },
801 Some(ch) => {
802 let _ = ch.send(()); // ignore erros here
803 },
804 }
805 }
806 }
807
808 /// Test if abort was requested.
809 pub fn abort_requested(&self) -> bool {
810 self.abort_requested.load(Ordering::SeqCst)
811 }
812
813 /// Fail if abort was requested.
814 pub fn fail_on_abort(&self) -> Result<(), Error> {
815 if self.abort_requested() {
816 bail!("abort requested - aborting task");
817 }
818 Ok(())
819 }
820
821 /// Get a future which resolves on task abort
822 pub fn abort_future(&self) -> oneshot::Receiver<()> {
823 let (tx, rx) = oneshot::channel::<()>();
824
825 let mut data = self.data.lock().unwrap();
826 if self.abort_requested() {
827 let _ = tx.send(());
828 } else {
829 data.abort_listeners.push(tx);
830 }
831 rx
832 }
833
834 pub fn upid(&self) -> &UPID {
835 &self.upid
836 }
837 }