1 //! Background worker that watches over the cache. 2 //! 3 //! It cleans up old cache, updates statistics and optimizes the cache. 4 //! We allow losing some messages (it doesn't hurt) and some races, 5 //! but we guarantee eventual consistency and fault tolerancy. 6 //! Background tasks can be CPU intensive, but the worker thread has low priority. 7 8 use super::{fs_write_atomic, CacheConfig}; 9 use log::{debug, info, trace, warn}; 10 use serde_derive::{Deserialize, Serialize}; 11 use std::cmp; 12 use std::collections::HashMap; 13 use std::ffi::OsStr; 14 use std::fmt; 15 use std::fs; 16 use std::path::{Path, PathBuf}; 17 use std::sync::mpsc::{sync_channel, Receiver, SyncSender}; 18 #[cfg(test)] 19 use std::sync::{Arc, Condvar, Mutex}; 20 use std::thread; 21 use std::time::Duration; 22 #[cfg(not(test))] 23 use std::time::SystemTime; 24 #[cfg(test)] 25 use tests::system_time_stub::SystemTimeStub as SystemTime; 26 27 #[derive(Clone)] 28 pub(super) struct Worker { 29 sender: SyncSender<CacheEvent>, 30 #[cfg(test)] 31 stats: Arc<(Mutex<WorkerStats>, Condvar)>, 32 } 33 34 struct WorkerThread { 35 receiver: Receiver<CacheEvent>, 36 cache_config: CacheConfig, 37 #[cfg(test)] 38 stats: Arc<(Mutex<WorkerStats>, Condvar)>, 39 } 40 41 #[cfg(test)] 42 #[derive(Default)] 43 struct WorkerStats { 44 dropped: u32, 45 sent: u32, 46 handled: u32, 47 } 48 49 #[derive(Debug, Clone)] 50 enum CacheEvent { 51 OnCacheGet(PathBuf), 52 OnCacheUpdate(PathBuf), 53 } 54 55 impl Worker { 56 pub(super) fn start_new(cache_config: &CacheConfig) -> Self { 57 let queue_size = match cache_config.worker_event_queue_size() { 58 num if num <= usize::max_value() as u64 => num as usize, 59 _ => usize::max_value(), 60 }; 61 let (tx, rx) = sync_channel(queue_size); 62 63 #[cfg(test)] 64 let stats = Arc::new((Mutex::new(WorkerStats::default()), Condvar::new())); 65 66 let worker_thread = WorkerThread { 67 receiver: rx, 68 cache_config: cache_config.clone(), 69 #[cfg(test)] 70 stats: stats.clone(), 71 }; 72 73 // when self is dropped, sender will be dropped, what will cause the channel 74 // to hang, and the worker thread to exit -- it happens in the tests 75 // non-tests binary has only a static worker, so Rust doesn't drop it 76 thread::spawn(move || worker_thread.run()); 77 78 Self { 79 sender: tx, 80 #[cfg(test)] 81 stats, 82 } 83 } 84 85 pub(super) fn on_cache_get_async(&self, path: impl AsRef<Path>) { 86 let event = CacheEvent::OnCacheGet(path.as_ref().to_path_buf()); 87 self.send_cache_event(event); 88 } 89 90 pub(super) fn on_cache_update_async(&self, path: impl AsRef<Path>) { 91 let event = CacheEvent::OnCacheUpdate(path.as_ref().to_path_buf()); 92 self.send_cache_event(event); 93 } 94 95 #[inline] 96 fn send_cache_event(&self, event: CacheEvent) { 97 let sent_event = self.sender.try_send(event.clone()); 98 99 if let Err(ref err) = sent_event { 100 info!( 101 "Failed to send asynchronously message to worker thread, \ 102 event: {:?}, error: {}", 103 event, err 104 ); 105 } 106 107 #[cfg(test)] 108 { 109 let mut stats = self 110 .stats 111 .0 112 .lock() 113 .expect("Failed to acquire worker stats lock"); 114 115 if sent_event.is_ok() { 116 stats.sent += 1; 117 } else { 118 stats.dropped += 1; 119 } 120 } 121 } 122 123 #[cfg(test)] 124 pub(super) fn events_dropped(&self) -> u32 { 125 let stats = self 126 .stats 127 .0 128 .lock() 129 .expect("Failed to acquire worker stats lock"); 130 stats.dropped 131 } 132 133 #[cfg(test)] 134 pub(super) fn wait_for_all_events_handled(&self) { 135 let (stats, condvar) = &*self.stats; 136 let mut stats = stats.lock().expect("Failed to acquire worker stats lock"); 137 while stats.handled != stats.sent { 138 stats = condvar 139 .wait(stats) 140 .expect("Failed to reacquire worker stats lock"); 141 } 142 } 143 } 144 145 impl fmt::Debug for Worker { 146 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { 147 f.debug_struct("Worker").finish() 148 } 149 } 150 151 #[derive(Serialize, Deserialize)] 152 struct ModuleCacheStatistics { 153 pub usages: u64, 154 #[serde(rename = "optimized-compression")] 155 pub compression_level: i32, 156 } 157 158 impl ModuleCacheStatistics { 159 fn default(cache_config: &CacheConfig) -> Self { 160 Self { 161 usages: 0, 162 compression_level: cache_config.baseline_compression_level(), 163 } 164 } 165 } 166 167 enum CacheEntry { 168 Recognized { 169 path: PathBuf, 170 mtime: SystemTime, 171 size: u64, 172 }, 173 Unrecognized { 174 path: PathBuf, 175 is_dir: bool, 176 }, 177 } 178 179 macro_rules! unwrap_or_warn { 180 ($result:expr, $cont:stmt, $err_msg:expr, $path:expr) => { 181 match $result { 182 Ok(val) => val, 183 Err(err) => { 184 warn!("{}, path: {}, msg: {}", $err_msg, $path.display(), err); 185 $cont 186 } 187 } 188 }; 189 } 190 191 impl WorkerThread { 192 fn run(self) { 193 debug!("Cache worker thread started."); 194 195 Self::lower_thread_priority(); 196 197 #[cfg(test)] 198 let (stats, condvar) = &*self.stats; 199 200 for event in self.receiver.iter() { 201 match event { 202 CacheEvent::OnCacheGet(path) => self.handle_on_cache_get(path), 203 CacheEvent::OnCacheUpdate(path) => self.handle_on_cache_update(path), 204 } 205 206 #[cfg(test)] 207 { 208 let mut stats = stats.lock().expect("Failed to acquire worker stats lock"); 209 stats.handled += 1; 210 condvar.notify_all(); 211 } 212 } 213 } 214 215 #[cfg(target_os = "fuchsia")] 216 fn lower_thread_priority() { 217 // TODO This needs to use Fuchsia thread profiles 218 // https://fuchsia.dev/fuchsia-src/reference/kernel_objects/profile 219 warn!( 220 "Lowering thread priority on Fuchsia is currently a noop. It might affect application performance." 221 ); 222 } 223 224 #[cfg(target_os = "windows")] 225 fn lower_thread_priority() { 226 use windows_sys::Win32::System::Threading::*; 227 228 // https://docs.microsoft.com/en-us/windows/win32/api/processthreadsapi/nf-processthreadsapi-setthreadpriority 229 // https://docs.microsoft.com/en-us/windows/win32/procthread/scheduling-priorities 230 231 if unsafe { 232 SetThreadPriority( 233 GetCurrentThread(), 234 THREAD_MODE_BACKGROUND_BEGIN.try_into().unwrap(), 235 ) 236 } == 0 237 { 238 warn!( 239 "Failed to lower worker thread priority. It might affect application performance." 240 ); 241 } 242 } 243 244 #[cfg(not(any(target_os = "windows", target_os = "fuchsia")))] 245 fn lower_thread_priority() { 246 // http://man7.org/linux/man-pages/man7/sched.7.html 247 248 const NICE_DELTA_FOR_BACKGROUND_TASKS: i32 = 3; 249 250 match rustix::process::nice(NICE_DELTA_FOR_BACKGROUND_TASKS) { 251 Ok(current_nice) => { 252 debug!("New nice value of worker thread: {}", current_nice); 253 } 254 Err(err) => { 255 warn!( 256 "Failed to lower worker thread priority ({:?}). It might affect application performance.", err); 257 } 258 }; 259 } 260 261 /// Increases the usage counter and recompresses the file 262 /// if the usage counter reached configurable threshold. 263 fn handle_on_cache_get(&self, path: PathBuf) { 264 trace!("handle_on_cache_get() for path: {}", path.display()); 265 266 // construct .stats file path 267 let filename = path.file_name().unwrap().to_str().unwrap(); 268 let stats_path = path.with_file_name(format!("{filename}.stats")); 269 270 // load .stats file (default if none or error) 271 let mut stats = read_stats_file(stats_path.as_ref()) 272 .unwrap_or_else(|| ModuleCacheStatistics::default(&self.cache_config)); 273 274 // step 1: update the usage counter & write to the disk 275 // it's racy, but it's fine (the counter will be just smaller, 276 // sometimes will retrigger recompression) 277 stats.usages += 1; 278 if !write_stats_file(stats_path.as_ref(), &stats) { 279 return; 280 } 281 282 // step 2: recompress if there's a need 283 let opt_compr_lvl = self.cache_config.optimized_compression_level(); 284 if stats.compression_level >= opt_compr_lvl 285 || stats.usages 286 < self 287 .cache_config 288 .optimized_compression_usage_counter_threshold() 289 { 290 return; 291 } 292 293 let lock_path = if let Some(p) = acquire_task_fs_lock( 294 path.as_ref(), 295 self.cache_config.optimizing_compression_task_timeout(), 296 self.cache_config 297 .allowed_clock_drift_for_files_from_future(), 298 ) { 299 p 300 } else { 301 return; 302 }; 303 304 trace!("Trying to recompress file: {}", path.display()); 305 306 // recompress, write to other file, rename (it's atomic file content exchange) 307 // and update the stats file 308 let compressed_cache_bytes = unwrap_or_warn!( 309 fs::read(&path), 310 return, 311 "Failed to read old cache file", 312 path 313 ); 314 315 let cache_bytes = unwrap_or_warn!( 316 zstd::decode_all(&compressed_cache_bytes[..]), 317 return, 318 "Failed to decompress cached code", 319 path 320 ); 321 322 let recompressed_cache_bytes = unwrap_or_warn!( 323 zstd::encode_all(&cache_bytes[..], opt_compr_lvl), 324 return, 325 "Failed to compress cached code", 326 path 327 ); 328 329 unwrap_or_warn!( 330 fs::write(&lock_path, &recompressed_cache_bytes), 331 return, 332 "Failed to write recompressed cache", 333 lock_path 334 ); 335 336 unwrap_or_warn!( 337 fs::rename(&lock_path, &path), 338 { 339 if let Err(error) = fs::remove_file(&lock_path) { 340 warn!( 341 "Failed to clean up (remove) recompressed cache, path {}, err: {}", 342 lock_path.display(), 343 error 344 ); 345 } 346 347 return; 348 }, 349 "Failed to rename recompressed cache", 350 lock_path 351 ); 352 353 // update stats file (reload it! recompression can take some time) 354 if let Some(mut new_stats) = read_stats_file(stats_path.as_ref()) { 355 if new_stats.compression_level >= opt_compr_lvl { 356 // Rare race: 357 // two instances with different opt_compr_lvl: we don't know in which order they updated 358 // the cache file and the stats file (they are not updated together atomically) 359 // Possible solution is to use directories per cache entry, but it complicates the system 360 // and is not worth it. 361 debug!( 362 "DETECTED task did more than once (or race with new file): \ 363 recompression of {}. Note: if optimized compression level setting \ 364 has changed in the meantine, the stats file might contain \ 365 inconsistent compression level due to race.", 366 path.display() 367 ); 368 } else { 369 new_stats.compression_level = opt_compr_lvl; 370 let _ = write_stats_file(stats_path.as_ref(), &new_stats); 371 } 372 373 if new_stats.usages < stats.usages { 374 debug!( 375 "DETECTED lower usage count (new file or race with counter \ 376 increasing): file {}", 377 path.display() 378 ); 379 } 380 } else { 381 debug!( 382 "Can't read stats file again to update compression level (it might got \ 383 cleaned up): file {}", 384 stats_path.display() 385 ); 386 } 387 388 trace!("Task finished: recompress file: {}", path.display()); 389 } 390 391 fn handle_on_cache_update(&self, path: PathBuf) { 392 trace!("handle_on_cache_update() for path: {}", path.display()); 393 394 // ---------------------- step 1: create .stats file 395 396 // construct .stats file path 397 let filename = path 398 .file_name() 399 .expect("Expected valid cache file name") 400 .to_str() 401 .expect("Expected valid cache file name"); 402 let stats_path = path.with_file_name(format!("{filename}.stats")); 403 404 // create and write stats file 405 let mut stats = ModuleCacheStatistics::default(&self.cache_config); 406 stats.usages += 1; 407 write_stats_file(&stats_path, &stats); 408 409 // ---------------------- step 2: perform cleanup task if needed 410 411 // acquire lock for cleanup task 412 // Lock is a proof of recent cleanup task, so we don't want to delete them. 413 // Expired locks will be deleted by the cleanup task. 414 let cleanup_file = self.cache_config.directory().join(".cleanup"); // some non existing marker file 415 if acquire_task_fs_lock( 416 &cleanup_file, 417 self.cache_config.cleanup_interval(), 418 self.cache_config 419 .allowed_clock_drift_for_files_from_future(), 420 ) 421 .is_none() 422 { 423 return; 424 } 425 426 trace!("Trying to clean up cache"); 427 428 let mut cache_index = self.list_cache_contents(); 429 let future_tolerance = SystemTime::now() 430 .checked_add( 431 self.cache_config 432 .allowed_clock_drift_for_files_from_future(), 433 ) 434 .expect("Brace your cache, the next Big Bang is coming (time overflow)"); 435 cache_index.sort_unstable_by(|lhs, rhs| { 436 // sort by age 437 use CacheEntry::*; 438 match (lhs, rhs) { 439 (Recognized { mtime: lhs_mt, .. }, Recognized { mtime: rhs_mt, .. }) => { 440 match (*lhs_mt > future_tolerance, *rhs_mt > future_tolerance) { 441 // later == younger 442 (false, false) => rhs_mt.cmp(lhs_mt), 443 // files from far future are treated as oldest recognized files 444 // we want to delete them, so the cache keeps track of recent files 445 // however, we don't delete them uncodintionally, 446 // because .stats file can be overwritten with a meaningful mtime 447 (true, false) => cmp::Ordering::Greater, 448 (false, true) => cmp::Ordering::Less, 449 (true, true) => cmp::Ordering::Equal, 450 } 451 } 452 // unrecognized is kind of infinity 453 (Recognized { .. }, Unrecognized { .. }) => cmp::Ordering::Less, 454 (Unrecognized { .. }, Recognized { .. }) => cmp::Ordering::Greater, 455 (Unrecognized { .. }, Unrecognized { .. }) => cmp::Ordering::Equal, 456 } 457 }); 458 459 // find "cut" boundary: 460 // - remove unrecognized files anyway, 461 // - remove some cache files if some quota has been exceeded 462 let mut total_size = 0u64; 463 let mut start_delete_idx = None; 464 let mut start_delete_idx_if_deleting_recognized_items: Option<usize> = None; 465 466 let total_size_limit = self.cache_config.files_total_size_soft_limit(); 467 let file_count_limit = self.cache_config.file_count_soft_limit(); 468 let tsl_if_deleting = total_size_limit 469 .checked_mul( 470 self.cache_config 471 .files_total_size_limit_percent_if_deleting() as u64, 472 ) 473 .unwrap() 474 / 100; 475 let fcl_if_deleting = file_count_limit 476 .checked_mul(self.cache_config.file_count_limit_percent_if_deleting() as u64) 477 .unwrap() 478 / 100; 479 480 for (idx, item) in cache_index.iter().enumerate() { 481 let size = if let CacheEntry::Recognized { size, .. } = item { 482 size 483 } else { 484 start_delete_idx = Some(idx); 485 break; 486 }; 487 488 total_size += size; 489 if start_delete_idx_if_deleting_recognized_items.is_none() 490 && (total_size > tsl_if_deleting || (idx + 1) as u64 > fcl_if_deleting) 491 { 492 start_delete_idx_if_deleting_recognized_items = Some(idx); 493 } 494 495 if total_size > total_size_limit || (idx + 1) as u64 > file_count_limit { 496 start_delete_idx = start_delete_idx_if_deleting_recognized_items; 497 break; 498 } 499 } 500 501 if let Some(idx) = start_delete_idx { 502 for item in &cache_index[idx..] { 503 let (result, path, entity) = match item { 504 CacheEntry::Recognized { path, .. } 505 | CacheEntry::Unrecognized { 506 path, 507 is_dir: false, 508 } => (fs::remove_file(path), path, "file"), 509 CacheEntry::Unrecognized { path, is_dir: true } => { 510 (fs::remove_dir_all(path), path, "directory") 511 } 512 }; 513 if let Err(err) = result { 514 warn!( 515 "Failed to remove {} during cleanup, path: {}, err: {}", 516 entity, 517 path.display(), 518 err 519 ); 520 } 521 } 522 } 523 524 trace!("Task finished: clean up cache"); 525 } 526 527 // Be fault tolerant: list as much as you can, and ignore the rest 528 fn list_cache_contents(&self) -> Vec<CacheEntry> { 529 fn enter_dir( 530 vec: &mut Vec<CacheEntry>, 531 dir_path: &Path, 532 level: u8, 533 cache_config: &CacheConfig, 534 ) { 535 macro_rules! add_unrecognized { 536 (file: $path:expr) => { 537 add_unrecognized!(false, $path) 538 }; 539 (dir: $path:expr) => { 540 add_unrecognized!(true, $path) 541 }; 542 ($is_dir:expr, $path:expr) => { 543 vec.push(CacheEntry::Unrecognized { 544 path: $path.to_path_buf(), 545 is_dir: $is_dir, 546 }) 547 }; 548 } 549 macro_rules! add_unrecognized_and { 550 ([ $( $ty:ident: $path:expr ),* ], $cont:stmt) => {{ 551 $( add_unrecognized!($ty: $path); )* 552 $cont 553 }}; 554 } 555 556 macro_rules! unwrap_or { 557 ($result:expr, $cont:stmt, $err_msg:expr) => { 558 unwrap_or!($result, $cont, $err_msg, dir_path) 559 }; 560 ($result:expr, $cont:stmt, $err_msg:expr, $path:expr) => { 561 unwrap_or_warn!( 562 $result, 563 $cont, 564 format!("{}, level: {}", $err_msg, level), 565 $path 566 ) 567 }; 568 } 569 570 // If we fail to list a directory, something bad is happening anyway 571 // (something touches our cache or we have disk failure) 572 // Try to delete it, so we can stay within soft limits of the cache size. 573 // This comment applies later in this function, too. 574 let it = unwrap_or!( 575 fs::read_dir(dir_path), 576 add_unrecognized_and!([dir: dir_path], return), 577 "Failed to list cache directory, deleting it" 578 ); 579 580 let mut cache_files = HashMap::new(); 581 for entry in it { 582 // read_dir() returns an iterator over results - in case some of them are errors 583 // we don't know their names, so we can't delete them. We don't want to delete 584 // the whole directory with good entries too, so we just ignore the erroneous entries. 585 let entry = unwrap_or!( 586 entry, 587 continue, 588 "Failed to read a cache dir entry (NOT deleting it, it still occupies space)" 589 ); 590 let path = entry.path(); 591 match (level, path.is_dir()) { 592 (0..=1, true) => enter_dir(vec, &path, level + 1, cache_config), 593 (0..=1, false) => { 594 if level == 0 595 && path.file_stem() == Some(OsStr::new(".cleanup")) 596 && path.extension().is_some() 597 // assume it's cleanup lock 598 && !is_fs_lock_expired( 599 Some(&entry), 600 &path, 601 cache_config.cleanup_interval(), 602 cache_config.allowed_clock_drift_for_files_from_future(), 603 ) 604 { 605 continue; // skip active lock 606 } 607 add_unrecognized!(file: path); 608 } 609 (2, false) => { 610 match path.extension().and_then(OsStr::to_str) { 611 // mod or stats file 612 None | Some("stats") => { 613 cache_files.insert(path, entry); 614 } 615 616 Some(ext) => { 617 // check if valid lock 618 let recognized = ext.starts_with("wip-") 619 && !is_fs_lock_expired( 620 Some(&entry), 621 &path, 622 cache_config.optimizing_compression_task_timeout(), 623 cache_config.allowed_clock_drift_for_files_from_future(), 624 ); 625 626 if !recognized { 627 add_unrecognized!(file: path); 628 } 629 } 630 } 631 } 632 (_, is_dir) => add_unrecognized!(is_dir, path), 633 } 634 } 635 636 // associate module with its stats & handle them 637 // assumption: just mods and stats 638 for (path, entry) in cache_files.iter() { 639 let path_buf: PathBuf; 640 let (mod_, stats_, is_mod) = match path.extension() { 641 Some(_) => { 642 path_buf = path.with_extension(""); 643 ( 644 cache_files.get(&path_buf).map(|v| (&path_buf, v)), 645 Some((path, entry)), 646 false, 647 ) 648 } 649 None => { 650 path_buf = path.with_extension("stats"); 651 ( 652 Some((path, entry)), 653 cache_files.get(&path_buf).map(|v| (&path_buf, v)), 654 true, 655 ) 656 } 657 }; 658 659 // construct a cache entry 660 match (mod_, stats_, is_mod) { 661 (Some((mod_path, mod_entry)), Some((stats_path, stats_entry)), true) => { 662 let mod_metadata = unwrap_or!( 663 mod_entry.metadata(), 664 add_unrecognized_and!([file: stats_path, file: mod_path], continue), 665 "Failed to get metadata, deleting BOTH module cache and stats files", 666 mod_path 667 ); 668 let stats_mtime = unwrap_or!( 669 stats_entry.metadata().and_then(|m| m.modified()), 670 add_unrecognized_and!( 671 [file: stats_path], 672 unwrap_or!( 673 mod_metadata.modified(), 674 add_unrecognized_and!( 675 [file: stats_path, file: mod_path], 676 continue 677 ), 678 "Failed to get mtime, deleting BOTH module cache and stats \ 679 files", 680 mod_path 681 ) 682 ), 683 "Failed to get metadata/mtime, deleting the file", 684 stats_path 685 ); 686 // .into() called for the SystemTimeStub if cfg(test) 687 vec.push(CacheEntry::Recognized { 688 path: mod_path.to_path_buf(), 689 mtime: stats_mtime.into(), 690 size: mod_metadata.len(), 691 }) 692 } 693 (Some(_), Some(_), false) => (), // was or will be handled by previous branch 694 (Some((mod_path, mod_entry)), None, _) => { 695 let (mod_metadata, mod_mtime) = unwrap_or!( 696 mod_entry 697 .metadata() 698 .and_then(|md| md.modified().map(|mt| (md, mt))), 699 add_unrecognized_and!([file: mod_path], continue), 700 "Failed to get metadata/mtime, deleting the file", 701 mod_path 702 ); 703 // .into() called for the SystemTimeStub if cfg(test) 704 vec.push(CacheEntry::Recognized { 705 path: mod_path.to_path_buf(), 706 mtime: mod_mtime.into(), 707 size: mod_metadata.len(), 708 }) 709 } 710 (None, Some((stats_path, _stats_entry)), _) => { 711 debug!("Found orphaned stats file: {}", stats_path.display()); 712 add_unrecognized!(file: stats_path); 713 } 714 _ => unreachable!(), 715 } 716 } 717 } 718 719 let mut vec = Vec::new(); 720 enter_dir( 721 &mut vec, 722 self.cache_config.directory(), 723 0, 724 &self.cache_config, 725 ); 726 vec 727 } 728 } 729 730 fn read_stats_file(path: &Path) -> Option<ModuleCacheStatistics> { 731 fs::read_to_string(path) 732 .map_err(|err| { 733 trace!( 734 "Failed to read stats file, path: {}, err: {}", 735 path.display(), 736 err 737 ) 738 }) 739 .and_then(|contents| { 740 toml::from_str::<ModuleCacheStatistics>(&contents).map_err(|err| { 741 trace!( 742 "Failed to parse stats file, path: {}, err: {}", 743 path.display(), 744 err, 745 ) 746 }) 747 }) 748 .ok() 749 } 750 751 fn write_stats_file(path: &Path, stats: &ModuleCacheStatistics) -> bool { 752 toml::to_string_pretty(&stats) 753 .map_err(|err| { 754 warn!( 755 "Failed to serialize stats file, path: {}, err: {}", 756 path.display(), 757 err 758 ) 759 }) 760 .and_then(|serialized| { 761 fs_write_atomic(path, "stats", serialized.as_bytes()).map_err(|_| ()) 762 }) 763 .is_ok() 764 } 765 766 /// Tries to acquire a lock for specific task. 767 /// 768 /// Returns Some(path) to the lock if succeeds. The task path must not 769 /// contain any extension and have file stem. 770 /// 771 /// To release a lock you need either manually rename or remove it, 772 /// or wait until it expires and cleanup task removes it. 773 /// 774 /// Note: this function is racy. Main idea is: be fault tolerant and 775 /// never block some task. The price is that we rarely do some task 776 /// more than once. 777 fn acquire_task_fs_lock( 778 task_path: &Path, 779 timeout: Duration, 780 allowed_future_drift: Duration, 781 ) -> Option<PathBuf> { 782 assert!(task_path.extension().is_none()); 783 assert!(task_path.file_stem().is_some()); 784 785 // list directory 786 let dir_path = task_path.parent()?; 787 let it = fs::read_dir(dir_path) 788 .map_err(|err| { 789 warn!( 790 "Failed to list cache directory, path: {}, err: {}", 791 dir_path.display(), 792 err 793 ) 794 }) 795 .ok()?; 796 797 // look for existing locks 798 for entry in it { 799 let entry = entry 800 .map_err(|err| { 801 warn!( 802 "Failed to list cache directory, path: {}, err: {}", 803 dir_path.display(), 804 err 805 ) 806 }) 807 .ok()?; 808 809 let path = entry.path(); 810 if path.is_dir() || path.file_stem() != task_path.file_stem() { 811 continue; 812 } 813 814 // check extension and mtime 815 match path.extension() { 816 None => continue, 817 Some(ext) => { 818 if let Some(ext_str) = ext.to_str() { 819 // if it's None, i.e. not valid UTF-8 string, then that's not our lock for sure 820 if ext_str.starts_with("wip-") 821 && !is_fs_lock_expired(Some(&entry), &path, timeout, allowed_future_drift) 822 { 823 return None; 824 } 825 } 826 } 827 } 828 } 829 830 // create the lock 831 let lock_path = task_path.with_extension(format!("wip-{}", std::process::id())); 832 let _file = fs::OpenOptions::new() 833 .create_new(true) 834 .write(true) 835 .open(&lock_path) 836 .map_err(|err| { 837 warn!( 838 "Failed to create lock file (note: it shouldn't exists): path: {}, err: {}", 839 lock_path.display(), 840 err 841 ) 842 }) 843 .ok()?; 844 845 Some(lock_path) 846 } 847 848 // we have either both, or just path; dir entry is desirable since on some platforms we can get 849 // metadata without extra syscalls 850 // furthermore: it's better to get a path if we have it instead of allocating a new one from the dir entry 851 fn is_fs_lock_expired( 852 entry: Option<&fs::DirEntry>, 853 path: &PathBuf, 854 threshold: Duration, 855 allowed_future_drift: Duration, 856 ) -> bool { 857 let mtime = match entry 858 .map_or_else(|| path.metadata(), |e| e.metadata()) 859 .and_then(|metadata| metadata.modified()) 860 { 861 Ok(mt) => mt, 862 Err(err) => { 863 warn!( 864 "Failed to get metadata/mtime, treating as an expired lock, path: {}, err: {}", 865 path.display(), 866 err 867 ); 868 return true; // can't read mtime, treat as expired, so this task will not be starved 869 } 870 }; 871 872 // DON'T use: mtime.elapsed() -- we must call SystemTime directly for the tests to be deterministic 873 match SystemTime::now().duration_since(mtime) { 874 Ok(elapsed) => elapsed >= threshold, 875 Err(err) => { 876 trace!( 877 "Found mtime in the future, treating as a not expired lock, path: {}, err: {}", 878 path.display(), 879 err 880 ); 881 // the lock is expired if the time is too far in the future 882 // it is fine to have network share and not synchronized clocks, 883 // but it's not good when user changes time in their system clock 884 err.duration() > allowed_future_drift 885 } 886 } 887 } 888 889 #[cfg(test)] 890 mod tests; 891