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::{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( 57 cache_config: &CacheConfig, 58 init_file_per_thread_logger: Option<&'static str>, 59 ) -> Self { 60 let queue_size = match cache_config.worker_event_queue_size() { 61 num if num <= usize::max_value() as u64 => num as usize, 62 _ => usize::max_value(), 63 }; 64 let (tx, rx) = sync_channel(queue_size); 65 66 #[cfg(test)] 67 let stats = Arc::new((Mutex::new(WorkerStats::default()), Condvar::new())); 68 69 let worker_thread = WorkerThread { 70 receiver: rx, 71 cache_config: cache_config.clone(), 72 #[cfg(test)] 73 stats: stats.clone(), 74 }; 75 76 // when self is dropped, sender will be dropped, what will cause the channel 77 // to hang, and the worker thread to exit -- it happens in the tests 78 // non-tests binary has only a static worker, so Rust doesn't drop it 79 thread::spawn(move || worker_thread.run(init_file_per_thread_logger)); 80 81 Self { 82 sender: tx, 83 #[cfg(test)] 84 stats, 85 } 86 } 87 88 pub(super) fn on_cache_get_async(&self, path: impl AsRef<Path>) { 89 let event = CacheEvent::OnCacheGet(path.as_ref().to_path_buf()); 90 self.send_cache_event(event); 91 } 92 93 pub(super) fn on_cache_update_async(&self, path: impl AsRef<Path>) { 94 let event = CacheEvent::OnCacheUpdate(path.as_ref().to_path_buf()); 95 self.send_cache_event(event); 96 } 97 98 #[inline] 99 fn send_cache_event(&self, event: CacheEvent) { 100 let sent_event = self.sender.try_send(event.clone()); 101 102 if let Err(ref err) = sent_event { 103 info!( 104 "Failed to send asynchronously message to worker thread, \ 105 event: {:?}, error: {}", 106 event, err 107 ); 108 } 109 110 #[cfg(test)] 111 { 112 let mut stats = self 113 .stats 114 .0 115 .lock() 116 .expect("Failed to acquire worker stats lock"); 117 118 if sent_event.is_ok() { 119 stats.sent += 1; 120 } else { 121 stats.dropped += 1; 122 } 123 } 124 } 125 126 #[cfg(test)] 127 pub(super) fn events_dropped(&self) -> u32 { 128 let stats = self 129 .stats 130 .0 131 .lock() 132 .expect("Failed to acquire worker stats lock"); 133 stats.dropped 134 } 135 136 #[cfg(test)] 137 pub(super) fn wait_for_all_events_handled(&self) { 138 let (stats, condvar) = &*self.stats; 139 let mut stats = stats.lock().expect("Failed to acquire worker stats lock"); 140 while stats.handled != stats.sent { 141 stats = condvar 142 .wait(stats) 143 .expect("Failed to reacquire worker stats lock"); 144 } 145 } 146 } 147 148 impl fmt::Debug for Worker { 149 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { 150 f.debug_struct("Worker").finish() 151 } 152 } 153 154 #[derive(Serialize, Deserialize)] 155 struct ModuleCacheStatistics { 156 pub usages: u64, 157 #[serde(rename = "optimized-compression")] 158 pub compression_level: i32, 159 } 160 161 impl ModuleCacheStatistics { 162 fn default(cache_config: &CacheConfig) -> Self { 163 Self { 164 usages: 0, 165 compression_level: cache_config.baseline_compression_level(), 166 } 167 } 168 } 169 170 enum CacheEntry { 171 Recognized { 172 path: PathBuf, 173 mtime: SystemTime, 174 size: u64, 175 }, 176 Unrecognized { 177 path: PathBuf, 178 is_dir: bool, 179 }, 180 } 181 182 macro_rules! unwrap_or_warn { 183 ($result:expr, $cont:stmt, $err_msg:expr, $path:expr) => { 184 match $result { 185 Ok(val) => val, 186 Err(err) => { 187 warn!("{}, path: {}, msg: {}", $err_msg, $path.display(), err); 188 $cont 189 } 190 } 191 }; 192 } 193 194 impl WorkerThread { 195 fn run(self, init_file_per_thread_logger: Option<&'static str>) { 196 if let Some(prefix) = init_file_per_thread_logger { 197 file_per_thread_logger::initialize(prefix); 198 } 199 200 debug!("Cache worker thread started."); 201 202 Self::lower_thread_priority(); 203 204 #[cfg(test)] 205 let (stats, condvar) = &*self.stats; 206 207 for event in self.receiver.iter() { 208 match event { 209 CacheEvent::OnCacheGet(path) => self.handle_on_cache_get(path), 210 CacheEvent::OnCacheUpdate(path) => self.handle_on_cache_update(path), 211 } 212 213 #[cfg(test)] 214 { 215 let mut stats = stats.lock().expect("Failed to acquire worker stats lock"); 216 stats.handled += 1; 217 condvar.notify_all(); 218 } 219 } 220 } 221 222 #[cfg(target_os = "fuchsia")] 223 fn lower_thread_priority() { 224 // TODO This needs to use Fuchsia thread profiles 225 // https://fuchsia.dev/fuchsia-src/reference/kernel_objects/profile 226 warn!( 227 "Lowering thread priority on Fuchsia is currently a noop. It might affect application performance." 228 ); 229 } 230 231 #[cfg(target_os = "windows")] 232 fn lower_thread_priority() { 233 use windows_sys::Win32::System::Threading::*; 234 235 // https://docs.microsoft.com/en-us/windows/win32/api/processthreadsapi/nf-processthreadsapi-setthreadpriority 236 // https://docs.microsoft.com/en-us/windows/win32/procthread/scheduling-priorities 237 238 if unsafe { 239 SetThreadPriority( 240 GetCurrentThread(), 241 THREAD_MODE_BACKGROUND_BEGIN.try_into().unwrap(), 242 ) 243 } == 0 244 { 245 warn!( 246 "Failed to lower worker thread priority. It might affect application performance." 247 ); 248 } 249 } 250 251 #[cfg(not(any(target_os = "windows", target_os = "fuchsia")))] 252 fn lower_thread_priority() { 253 // http://man7.org/linux/man-pages/man7/sched.7.html 254 255 const NICE_DELTA_FOR_BACKGROUND_TASKS: i32 = 3; 256 257 match rustix::process::nice(NICE_DELTA_FOR_BACKGROUND_TASKS) { 258 Ok(current_nice) => { 259 debug!("New nice value of worker thread: {}", current_nice); 260 } 261 Err(err) => { 262 warn!( 263 "Failed to lower worker thread priority ({:?}). It might affect application performance.", err); 264 } 265 }; 266 } 267 268 /// Increases the usage counter and recompresses the file 269 /// if the usage counter reached configurable treshold. 270 fn handle_on_cache_get(&self, path: PathBuf) { 271 trace!("handle_on_cache_get() for path: {}", path.display()); 272 273 // construct .stats file path 274 let filename = path.file_name().unwrap().to_str().unwrap(); 275 let stats_path = path.with_file_name(format!("{}.stats", filename)); 276 277 // load .stats file (default if none or error) 278 let mut stats = read_stats_file(stats_path.as_ref()) 279 .unwrap_or_else(|| ModuleCacheStatistics::default(&self.cache_config)); 280 281 // step 1: update the usage counter & write to the disk 282 // it's racy, but it's fine (the counter will be just smaller, 283 // sometimes will retrigger recompression) 284 stats.usages += 1; 285 if !write_stats_file(stats_path.as_ref(), &stats) { 286 return; 287 } 288 289 // step 2: recompress if there's a need 290 let opt_compr_lvl = self.cache_config.optimized_compression_level(); 291 if stats.compression_level >= opt_compr_lvl 292 || stats.usages 293 < self 294 .cache_config 295 .optimized_compression_usage_counter_threshold() 296 { 297 return; 298 } 299 300 let lock_path = if let Some(p) = acquire_task_fs_lock( 301 path.as_ref(), 302 self.cache_config.optimizing_compression_task_timeout(), 303 self.cache_config 304 .allowed_clock_drift_for_files_from_future(), 305 ) { 306 p 307 } else { 308 return; 309 }; 310 311 trace!("Trying to recompress file: {}", path.display()); 312 313 // recompress, write to other file, rename (it's atomic file content exchange) 314 // and update the stats file 315 let compressed_cache_bytes = unwrap_or_warn!( 316 fs::read(&path), 317 return, 318 "Failed to read old cache file", 319 path 320 ); 321 322 let cache_bytes = unwrap_or_warn!( 323 zstd::decode_all(&compressed_cache_bytes[..]), 324 return, 325 "Failed to decompress cached code", 326 path 327 ); 328 329 let recompressed_cache_bytes = unwrap_or_warn!( 330 zstd::encode_all(&cache_bytes[..], opt_compr_lvl), 331 return, 332 "Failed to compress cached code", 333 path 334 ); 335 336 unwrap_or_warn!( 337 fs::write(&lock_path, &recompressed_cache_bytes), 338 return, 339 "Failed to write recompressed cache", 340 lock_path 341 ); 342 343 unwrap_or_warn!( 344 fs::rename(&lock_path, &path), 345 { 346 if let Err(error) = fs::remove_file(&lock_path) { 347 warn!( 348 "Failed to clean up (remove) recompressed cache, path {}, err: {}", 349 lock_path.display(), 350 error 351 ); 352 } 353 354 return; 355 }, 356 "Failed to rename recompressed cache", 357 lock_path 358 ); 359 360 // update stats file (reload it! recompression can take some time) 361 if let Some(mut new_stats) = read_stats_file(stats_path.as_ref()) { 362 if new_stats.compression_level >= opt_compr_lvl { 363 // Rare race: 364 // two instances with different opt_compr_lvl: we don't know in which order they updated 365 // the cache file and the stats file (they are not updated together atomically) 366 // Possible solution is to use directories per cache entry, but it complicates the system 367 // and is not worth it. 368 debug!( 369 "DETECTED task did more than once (or race with new file): \ 370 recompression of {}. Note: if optimized compression level setting \ 371 has changed in the meantine, the stats file might contain \ 372 inconsistent compression level due to race.", 373 path.display() 374 ); 375 } else { 376 new_stats.compression_level = opt_compr_lvl; 377 let _ = write_stats_file(stats_path.as_ref(), &new_stats); 378 } 379 380 if new_stats.usages < stats.usages { 381 debug!( 382 "DETECTED lower usage count (new file or race with counter \ 383 increasing): file {}", 384 path.display() 385 ); 386 } 387 } else { 388 debug!( 389 "Can't read stats file again to update compression level (it might got \ 390 cleaned up): file {}", 391 stats_path.display() 392 ); 393 } 394 395 trace!("Task finished: recompress file: {}", path.display()); 396 } 397 398 fn handle_on_cache_update(&self, path: PathBuf) { 399 trace!("handle_on_cache_update() for path: {}", path.display()); 400 401 // ---------------------- step 1: create .stats file 402 403 // construct .stats file path 404 let filename = path 405 .file_name() 406 .expect("Expected valid cache file name") 407 .to_str() 408 .expect("Expected valid cache file name"); 409 let stats_path = path.with_file_name(format!("{}.stats", filename)); 410 411 // create and write stats file 412 let mut stats = ModuleCacheStatistics::default(&self.cache_config); 413 stats.usages += 1; 414 write_stats_file(&stats_path, &stats); 415 416 // ---------------------- step 2: perform cleanup task if needed 417 418 // acquire lock for cleanup task 419 // Lock is a proof of recent cleanup task, so we don't want to delete them. 420 // Expired locks will be deleted by the cleanup task. 421 let cleanup_file = self.cache_config.directory().join(".cleanup"); // some non existing marker file 422 if acquire_task_fs_lock( 423 &cleanup_file, 424 self.cache_config.cleanup_interval(), 425 self.cache_config 426 .allowed_clock_drift_for_files_from_future(), 427 ) 428 .is_none() 429 { 430 return; 431 } 432 433 trace!("Trying to clean up cache"); 434 435 let mut cache_index = self.list_cache_contents(); 436 let future_tolerance = SystemTime::now() 437 .checked_add( 438 self.cache_config 439 .allowed_clock_drift_for_files_from_future(), 440 ) 441 .expect("Brace your cache, the next Big Bang is coming (time overflow)"); 442 cache_index.sort_unstable_by(|lhs, rhs| { 443 // sort by age 444 use CacheEntry::*; 445 match (lhs, rhs) { 446 (Recognized { mtime: lhs_mt, .. }, Recognized { mtime: rhs_mt, .. }) => { 447 match (*lhs_mt > future_tolerance, *rhs_mt > future_tolerance) { 448 // later == younger 449 (false, false) => rhs_mt.cmp(lhs_mt), 450 // files from far future are treated as oldest recognized files 451 // we want to delete them, so the cache keeps track of recent files 452 // however, we don't delete them uncodintionally, 453 // because .stats file can be overwritten with a meaningful mtime 454 (true, false) => cmp::Ordering::Greater, 455 (false, true) => cmp::Ordering::Less, 456 (true, true) => cmp::Ordering::Equal, 457 } 458 } 459 // unrecognized is kind of infinity 460 (Recognized { .. }, Unrecognized { .. }) => cmp::Ordering::Less, 461 (Unrecognized { .. }, Recognized { .. }) => cmp::Ordering::Greater, 462 (Unrecognized { .. }, Unrecognized { .. }) => cmp::Ordering::Equal, 463 } 464 }); 465 466 // find "cut" boundary: 467 // - remove unrecognized files anyway, 468 // - remove some cache files if some quota has been exceeded 469 let mut total_size = 0u64; 470 let mut start_delete_idx = None; 471 let mut start_delete_idx_if_deleting_recognized_items: Option<usize> = None; 472 473 let total_size_limit = self.cache_config.files_total_size_soft_limit(); 474 let file_count_limit = self.cache_config.file_count_soft_limit(); 475 let tsl_if_deleting = total_size_limit 476 .checked_mul( 477 self.cache_config 478 .files_total_size_limit_percent_if_deleting() as u64, 479 ) 480 .unwrap() 481 / 100; 482 let fcl_if_deleting = file_count_limit 483 .checked_mul(self.cache_config.file_count_limit_percent_if_deleting() as u64) 484 .unwrap() 485 / 100; 486 487 for (idx, item) in cache_index.iter().enumerate() { 488 let size = if let CacheEntry::Recognized { size, .. } = item { 489 size 490 } else { 491 start_delete_idx = Some(idx); 492 break; 493 }; 494 495 total_size += size; 496 if start_delete_idx_if_deleting_recognized_items.is_none() 497 && (total_size > tsl_if_deleting || (idx + 1) as u64 > fcl_if_deleting) 498 { 499 start_delete_idx_if_deleting_recognized_items = Some(idx); 500 } 501 502 if total_size > total_size_limit || (idx + 1) as u64 > file_count_limit { 503 start_delete_idx = start_delete_idx_if_deleting_recognized_items; 504 break; 505 } 506 } 507 508 if let Some(idx) = start_delete_idx { 509 for item in &cache_index[idx..] { 510 let (result, path, entity) = match item { 511 CacheEntry::Recognized { path, .. } 512 | CacheEntry::Unrecognized { 513 path, 514 is_dir: false, 515 } => (fs::remove_file(path), path, "file"), 516 CacheEntry::Unrecognized { path, is_dir: true } => { 517 (fs::remove_dir_all(path), path, "directory") 518 } 519 }; 520 if let Err(err) = result { 521 warn!( 522 "Failed to remove {} during cleanup, path: {}, err: {}", 523 entity, 524 path.display(), 525 err 526 ); 527 } 528 } 529 } 530 531 trace!("Task finished: clean up cache"); 532 } 533 534 // Be fault tolerant: list as much as you can, and ignore the rest 535 fn list_cache_contents(&self) -> Vec<CacheEntry> { 536 fn enter_dir( 537 vec: &mut Vec<CacheEntry>, 538 dir_path: &Path, 539 level: u8, 540 cache_config: &CacheConfig, 541 ) { 542 macro_rules! add_unrecognized { 543 (file: $path:expr) => { 544 add_unrecognized!(false, $path) 545 }; 546 (dir: $path:expr) => { 547 add_unrecognized!(true, $path) 548 }; 549 ($is_dir:expr, $path:expr) => { 550 vec.push(CacheEntry::Unrecognized { 551 path: $path.to_path_buf(), 552 is_dir: $is_dir, 553 }) 554 }; 555 } 556 macro_rules! add_unrecognized_and { 557 ([ $( $ty:ident: $path:expr ),* ], $cont:stmt) => {{ 558 $( add_unrecognized!($ty: $path); )* 559 $cont 560 }}; 561 } 562 563 macro_rules! unwrap_or { 564 ($result:expr, $cont:stmt, $err_msg:expr) => { 565 unwrap_or!($result, $cont, $err_msg, dir_path) 566 }; 567 ($result:expr, $cont:stmt, $err_msg:expr, $path:expr) => { 568 unwrap_or_warn!( 569 $result, 570 $cont, 571 format!("{}, level: {}", $err_msg, level), 572 $path 573 ) 574 }; 575 } 576 577 // If we fail to list a directory, something bad is happening anyway 578 // (something touches our cache or we have disk failure) 579 // Try to delete it, so we can stay within soft limits of the cache size. 580 // This comment applies later in this function, too. 581 let it = unwrap_or!( 582 fs::read_dir(dir_path), 583 add_unrecognized_and!([dir: dir_path], return), 584 "Failed to list cache directory, deleting it" 585 ); 586 587 let mut cache_files = HashMap::new(); 588 for entry in it { 589 // read_dir() returns an iterator over results - in case some of them are errors 590 // we don't know their names, so we can't delete them. We don't want to delete 591 // the whole directory with good entries too, so we just ignore the erroneous entries. 592 let entry = unwrap_or!( 593 entry, 594 continue, 595 "Failed to read a cache dir entry (NOT deleting it, it still occupies space)" 596 ); 597 let path = entry.path(); 598 match (level, path.is_dir()) { 599 (0..=1, true) => enter_dir(vec, &path, level + 1, cache_config), 600 (0..=1, false) => { 601 if level == 0 602 && path.file_stem() == Some(OsStr::new(".cleanup")) 603 && path.extension().is_some() 604 // assume it's cleanup lock 605 && !is_fs_lock_expired( 606 Some(&entry), 607 &path, 608 cache_config.cleanup_interval(), 609 cache_config.allowed_clock_drift_for_files_from_future(), 610 ) 611 { 612 continue; // skip active lock 613 } 614 add_unrecognized!(file: path); 615 } 616 (2, false) => { 617 match path.extension().and_then(OsStr::to_str) { 618 // mod or stats file 619 None | Some("stats") => { 620 cache_files.insert(path, entry); 621 } 622 623 Some(ext) => { 624 // check if valid lock 625 let recognized = ext.starts_with("wip-") 626 && !is_fs_lock_expired( 627 Some(&entry), 628 &path, 629 cache_config.optimizing_compression_task_timeout(), 630 cache_config.allowed_clock_drift_for_files_from_future(), 631 ); 632 633 if !recognized { 634 add_unrecognized!(file: path); 635 } 636 } 637 } 638 } 639 (_, is_dir) => add_unrecognized!(is_dir, path), 640 } 641 } 642 643 // associate module with its stats & handle them 644 // assumption: just mods and stats 645 for (path, entry) in cache_files.iter() { 646 let path_buf: PathBuf; 647 let (mod_, stats_, is_mod) = match path.extension() { 648 Some(_) => { 649 path_buf = path.with_extension(""); 650 ( 651 cache_files.get(&path_buf).map(|v| (&path_buf, v)), 652 Some((path, entry)), 653 false, 654 ) 655 } 656 None => { 657 path_buf = path.with_extension("stats"); 658 ( 659 Some((path, entry)), 660 cache_files.get(&path_buf).map(|v| (&path_buf, v)), 661 true, 662 ) 663 } 664 }; 665 666 // construct a cache entry 667 match (mod_, stats_, is_mod) { 668 (Some((mod_path, mod_entry)), Some((stats_path, stats_entry)), true) => { 669 let mod_metadata = unwrap_or!( 670 mod_entry.metadata(), 671 add_unrecognized_and!([file: stats_path, file: mod_path], continue), 672 "Failed to get metadata, deleting BOTH module cache and stats files", 673 mod_path 674 ); 675 let stats_mtime = unwrap_or!( 676 stats_entry.metadata().and_then(|m| m.modified()), 677 add_unrecognized_and!( 678 [file: stats_path], 679 unwrap_or!( 680 mod_metadata.modified(), 681 add_unrecognized_and!( 682 [file: stats_path, file: mod_path], 683 continue 684 ), 685 "Failed to get mtime, deleting BOTH module cache and stats \ 686 files", 687 mod_path 688 ) 689 ), 690 "Failed to get metadata/mtime, deleting the file", 691 stats_path 692 ); 693 // .into() called for the SystemTimeStub if cfg(test) 694 #[allow(clippy::identity_conversion)] 695 vec.push(CacheEntry::Recognized { 696 path: mod_path.to_path_buf(), 697 mtime: stats_mtime.into(), 698 size: mod_metadata.len(), 699 }) 700 } 701 (Some(_), Some(_), false) => (), // was or will be handled by previous branch 702 (Some((mod_path, mod_entry)), None, _) => { 703 let (mod_metadata, mod_mtime) = unwrap_or!( 704 mod_entry 705 .metadata() 706 .and_then(|md| md.modified().map(|mt| (md, mt))), 707 add_unrecognized_and!([file: mod_path], continue), 708 "Failed to get metadata/mtime, deleting the file", 709 mod_path 710 ); 711 // .into() called for the SystemTimeStub if cfg(test) 712 #[allow(clippy::identity_conversion)] 713 vec.push(CacheEntry::Recognized { 714 path: mod_path.to_path_buf(), 715 mtime: mod_mtime.into(), 716 size: mod_metadata.len(), 717 }) 718 } 719 (None, Some((stats_path, _stats_entry)), _) => { 720 debug!("Found orphaned stats file: {}", stats_path.display()); 721 add_unrecognized!(file: stats_path); 722 } 723 _ => unreachable!(), 724 } 725 } 726 } 727 728 let mut vec = Vec::new(); 729 enter_dir( 730 &mut vec, 731 self.cache_config.directory(), 732 0, 733 &self.cache_config, 734 ); 735 vec 736 } 737 } 738 739 fn read_stats_file(path: &Path) -> Option<ModuleCacheStatistics> { 740 fs::read(path) 741 .map_err(|err| { 742 trace!( 743 "Failed to read stats file, path: {}, err: {}", 744 path.display(), 745 err 746 ) 747 }) 748 .and_then(|bytes| { 749 toml::from_slice::<ModuleCacheStatistics>(&bytes[..]).map_err(|err| { 750 trace!( 751 "Failed to parse stats file, path: {}, err: {}", 752 path.display(), 753 err, 754 ) 755 }) 756 }) 757 .ok() 758 } 759 760 fn write_stats_file(path: &Path, stats: &ModuleCacheStatistics) -> bool { 761 toml::to_string_pretty(&stats) 762 .map_err(|err| { 763 warn!( 764 "Failed to serialize stats file, path: {}, err: {}", 765 path.display(), 766 err 767 ) 768 }) 769 .and_then(|serialized| { 770 if fs_write_atomic(path, "stats", serialized.as_bytes()) { 771 Ok(()) 772 } else { 773 Err(()) 774 } 775 }) 776 .is_ok() 777 } 778 779 /// Tries to acquire a lock for specific task. 780 /// 781 /// Returns Some(path) to the lock if succeeds. The task path must not 782 /// contain any extension and have file stem. 783 /// 784 /// To release a lock you need either manually rename or remove it, 785 /// or wait until it expires and cleanup task removes it. 786 /// 787 /// Note: this function is racy. Main idea is: be fault tolerant and 788 /// never block some task. The price is that we rarely do some task 789 /// more than once. 790 fn acquire_task_fs_lock( 791 task_path: &Path, 792 timeout: Duration, 793 allowed_future_drift: Duration, 794 ) -> Option<PathBuf> { 795 assert!(task_path.extension().is_none()); 796 assert!(task_path.file_stem().is_some()); 797 798 // list directory 799 let dir_path = task_path.parent()?; 800 let it = fs::read_dir(dir_path) 801 .map_err(|err| { 802 warn!( 803 "Failed to list cache directory, path: {}, err: {}", 804 dir_path.display(), 805 err 806 ) 807 }) 808 .ok()?; 809 810 // look for existing locks 811 for entry in it { 812 let entry = entry 813 .map_err(|err| { 814 warn!( 815 "Failed to list cache directory, path: {}, err: {}", 816 dir_path.display(), 817 err 818 ) 819 }) 820 .ok()?; 821 822 let path = entry.path(); 823 if path.is_dir() || path.file_stem() != task_path.file_stem() { 824 continue; 825 } 826 827 // check extension and mtime 828 match path.extension() { 829 None => continue, 830 Some(ext) => { 831 if let Some(ext_str) = ext.to_str() { 832 // if it's None, i.e. not valid UTF-8 string, then that's not our lock for sure 833 if ext_str.starts_with("wip-") 834 && !is_fs_lock_expired(Some(&entry), &path, timeout, allowed_future_drift) 835 { 836 return None; 837 } 838 } 839 } 840 } 841 } 842 843 // create the lock 844 let lock_path = task_path.with_extension(format!("wip-{}", std::process::id())); 845 let _file = fs::OpenOptions::new() 846 .create_new(true) 847 .write(true) 848 .open(&lock_path) 849 .map_err(|err| { 850 warn!( 851 "Failed to create lock file (note: it shouldn't exists): path: {}, err: {}", 852 lock_path.display(), 853 err 854 ) 855 }) 856 .ok()?; 857 858 Some(lock_path) 859 } 860 861 // we have either both, or just path; dir entry is desirable since on some platforms we can get 862 // metadata without extra syscalls 863 // futhermore: it's better to get a path if we have it instead of allocating a new one from the dir entry 864 fn is_fs_lock_expired( 865 entry: Option<&fs::DirEntry>, 866 path: &PathBuf, 867 threshold: Duration, 868 allowed_future_drift: Duration, 869 ) -> bool { 870 let mtime = match entry 871 .map_or_else(|| path.metadata(), |e| e.metadata()) 872 .and_then(|metadata| metadata.modified()) 873 { 874 Ok(mt) => mt, 875 Err(err) => { 876 warn!( 877 "Failed to get metadata/mtime, treating as an expired lock, path: {}, err: {}", 878 path.display(), 879 err 880 ); 881 return true; // can't read mtime, treat as expired, so this task will not be starved 882 } 883 }; 884 885 // DON'T use: mtime.elapsed() -- we must call SystemTime directly for the tests to be deterministic 886 match SystemTime::now().duration_since(mtime) { 887 Ok(elapsed) => elapsed >= threshold, 888 Err(err) => { 889 trace!( 890 "Found mtime in the future, treating as a not expired lock, path: {}, err: {}", 891 path.display(), 892 err 893 ); 894 // the lock is expired if the time is too far in the future 895 // it is fine to have network share and not synchronized clocks, 896 // but it's not good when user changes time in their system clock 897 err.duration() > allowed_future_drift 898 } 899 } 900 } 901 902 #[cfg(test)] 903 mod tests; 904