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 winapi::um::processthreadsapi::{GetCurrentThread, SetThreadPriority}; 234 use winapi::um::winbase::THREAD_MODE_BACKGROUND_BEGIN; 235 236 // https://docs.microsoft.com/en-us/windows/win32/api/processthreadsapi/nf-processthreadsapi-setthreadpriority 237 // https://docs.microsoft.com/en-us/windows/win32/procthread/scheduling-priorities 238 239 if unsafe { 240 SetThreadPriority( 241 GetCurrentThread(), 242 THREAD_MODE_BACKGROUND_BEGIN.try_into().unwrap(), 243 ) 244 } == 0 245 { 246 warn!( 247 "Failed to lower worker thread priority. It might affect application performance." 248 ); 249 } 250 } 251 252 #[cfg(not(any(target_os = "windows", target_os = "fuchsia")))] 253 fn lower_thread_priority() { 254 // http://man7.org/linux/man-pages/man7/sched.7.html 255 256 const NICE_DELTA_FOR_BACKGROUND_TASKS: i32 = 3; 257 258 match rustix::process::nice(NICE_DELTA_FOR_BACKGROUND_TASKS) { 259 Ok(current_nice) => { 260 debug!("New nice value of worker thread: {}", current_nice); 261 } 262 Err(err) => { 263 warn!( 264 "Failed to lower worker thread priority ({:?}). It might affect application performance.", err); 265 } 266 }; 267 } 268 269 /// Increases the usage counter and recompresses the file 270 /// if the usage counter reached configurable treshold. 271 fn handle_on_cache_get(&self, path: PathBuf) { 272 trace!("handle_on_cache_get() for path: {}", path.display()); 273 274 // construct .stats file path 275 let filename = path.file_name().unwrap().to_str().unwrap(); 276 let stats_path = path.with_file_name(format!("{}.stats", filename)); 277 278 // load .stats file (default if none or error) 279 let mut stats = read_stats_file(stats_path.as_ref()) 280 .unwrap_or_else(|| ModuleCacheStatistics::default(&self.cache_config)); 281 282 // step 1: update the usage counter & write to the disk 283 // it's racy, but it's fine (the counter will be just smaller, 284 // sometimes will retrigger recompression) 285 stats.usages += 1; 286 if !write_stats_file(stats_path.as_ref(), &stats) { 287 return; 288 } 289 290 // step 2: recompress if there's a need 291 let opt_compr_lvl = self.cache_config.optimized_compression_level(); 292 if stats.compression_level >= opt_compr_lvl 293 || stats.usages 294 < self 295 .cache_config 296 .optimized_compression_usage_counter_threshold() 297 { 298 return; 299 } 300 301 let lock_path = if let Some(p) = acquire_task_fs_lock( 302 path.as_ref(), 303 self.cache_config.optimizing_compression_task_timeout(), 304 self.cache_config 305 .allowed_clock_drift_for_files_from_future(), 306 ) { 307 p 308 } else { 309 return; 310 }; 311 312 trace!("Trying to recompress file: {}", path.display()); 313 314 // recompress, write to other file, rename (it's atomic file content exchange) 315 // and update the stats file 316 let compressed_cache_bytes = unwrap_or_warn!( 317 fs::read(&path), 318 return, 319 "Failed to read old cache file", 320 path 321 ); 322 323 let cache_bytes = unwrap_or_warn!( 324 zstd::decode_all(&compressed_cache_bytes[..]), 325 return, 326 "Failed to decompress cached code", 327 path 328 ); 329 330 let recompressed_cache_bytes = unwrap_or_warn!( 331 zstd::encode_all(&cache_bytes[..], opt_compr_lvl), 332 return, 333 "Failed to compress cached code", 334 path 335 ); 336 337 unwrap_or_warn!( 338 fs::write(&lock_path, &recompressed_cache_bytes), 339 return, 340 "Failed to write recompressed cache", 341 lock_path 342 ); 343 344 unwrap_or_warn!( 345 fs::rename(&lock_path, &path), 346 { 347 if let Err(error) = fs::remove_file(&lock_path) { 348 warn!( 349 "Failed to clean up (remove) recompressed cache, path {}, err: {}", 350 lock_path.display(), 351 error 352 ); 353 } 354 355 return; 356 }, 357 "Failed to rename recompressed cache", 358 lock_path 359 ); 360 361 // update stats file (reload it! recompression can take some time) 362 if let Some(mut new_stats) = read_stats_file(stats_path.as_ref()) { 363 if new_stats.compression_level >= opt_compr_lvl { 364 // Rare race: 365 // two instances with different opt_compr_lvl: we don't know in which order they updated 366 // the cache file and the stats file (they are not updated together atomically) 367 // Possible solution is to use directories per cache entry, but it complicates the system 368 // and is not worth it. 369 debug!( 370 "DETECTED task did more than once (or race with new file): \ 371 recompression of {}. Note: if optimized compression level setting \ 372 has changed in the meantine, the stats file might contain \ 373 inconsistent compression level due to race.", 374 path.display() 375 ); 376 } else { 377 new_stats.compression_level = opt_compr_lvl; 378 let _ = write_stats_file(stats_path.as_ref(), &new_stats); 379 } 380 381 if new_stats.usages < stats.usages { 382 debug!( 383 "DETECTED lower usage count (new file or race with counter \ 384 increasing): file {}", 385 path.display() 386 ); 387 } 388 } else { 389 debug!( 390 "Can't read stats file again to update compression level (it might got \ 391 cleaned up): file {}", 392 stats_path.display() 393 ); 394 } 395 396 trace!("Task finished: recompress file: {}", path.display()); 397 } 398 399 fn handle_on_cache_update(&self, path: PathBuf) { 400 trace!("handle_on_cache_update() for path: {}", path.display()); 401 402 // ---------------------- step 1: create .stats file 403 404 // construct .stats file path 405 let filename = path 406 .file_name() 407 .expect("Expected valid cache file name") 408 .to_str() 409 .expect("Expected valid cache file name"); 410 let stats_path = path.with_file_name(format!("{}.stats", filename)); 411 412 // create and write stats file 413 let mut stats = ModuleCacheStatistics::default(&self.cache_config); 414 stats.usages += 1; 415 write_stats_file(&stats_path, &stats); 416 417 // ---------------------- step 2: perform cleanup task if needed 418 419 // acquire lock for cleanup task 420 // Lock is a proof of recent cleanup task, so we don't want to delete them. 421 // Expired locks will be deleted by the cleanup task. 422 let cleanup_file = self.cache_config.directory().join(".cleanup"); // some non existing marker file 423 if acquire_task_fs_lock( 424 &cleanup_file, 425 self.cache_config.cleanup_interval(), 426 self.cache_config 427 .allowed_clock_drift_for_files_from_future(), 428 ) 429 .is_none() 430 { 431 return; 432 } 433 434 trace!("Trying to clean up cache"); 435 436 let mut cache_index = self.list_cache_contents(); 437 let future_tolerance = SystemTime::now() 438 .checked_add( 439 self.cache_config 440 .allowed_clock_drift_for_files_from_future(), 441 ) 442 .expect("Brace your cache, the next Big Bang is coming (time overflow)"); 443 cache_index.sort_unstable_by(|lhs, rhs| { 444 // sort by age 445 use CacheEntry::*; 446 match (lhs, rhs) { 447 (Recognized { mtime: lhs_mt, .. }, Recognized { mtime: rhs_mt, .. }) => { 448 match (*lhs_mt > future_tolerance, *rhs_mt > future_tolerance) { 449 // later == younger 450 (false, false) => rhs_mt.cmp(lhs_mt), 451 // files from far future are treated as oldest recognized files 452 // we want to delete them, so the cache keeps track of recent files 453 // however, we don't delete them uncodintionally, 454 // because .stats file can be overwritten with a meaningful mtime 455 (true, false) => cmp::Ordering::Greater, 456 (false, true) => cmp::Ordering::Less, 457 (true, true) => cmp::Ordering::Equal, 458 } 459 } 460 // unrecognized is kind of infinity 461 (Recognized { .. }, Unrecognized { .. }) => cmp::Ordering::Less, 462 (Unrecognized { .. }, Recognized { .. }) => cmp::Ordering::Greater, 463 (Unrecognized { .. }, Unrecognized { .. }) => cmp::Ordering::Equal, 464 } 465 }); 466 467 // find "cut" boundary: 468 // - remove unrecognized files anyway, 469 // - remove some cache files if some quota has been exceeded 470 let mut total_size = 0u64; 471 let mut start_delete_idx = None; 472 let mut start_delete_idx_if_deleting_recognized_items: Option<usize> = None; 473 474 let total_size_limit = self.cache_config.files_total_size_soft_limit(); 475 let file_count_limit = self.cache_config.file_count_soft_limit(); 476 let tsl_if_deleting = total_size_limit 477 .checked_mul( 478 self.cache_config 479 .files_total_size_limit_percent_if_deleting() as u64, 480 ) 481 .unwrap() 482 / 100; 483 let fcl_if_deleting = file_count_limit 484 .checked_mul(self.cache_config.file_count_limit_percent_if_deleting() as u64) 485 .unwrap() 486 / 100; 487 488 for (idx, item) in cache_index.iter().enumerate() { 489 let size = if let CacheEntry::Recognized { size, .. } = item { 490 size 491 } else { 492 start_delete_idx = Some(idx); 493 break; 494 }; 495 496 total_size += size; 497 if start_delete_idx_if_deleting_recognized_items.is_none() 498 && (total_size > tsl_if_deleting || (idx + 1) as u64 > fcl_if_deleting) 499 { 500 start_delete_idx_if_deleting_recognized_items = Some(idx); 501 } 502 503 if total_size > total_size_limit || (idx + 1) as u64 > file_count_limit { 504 start_delete_idx = start_delete_idx_if_deleting_recognized_items; 505 break; 506 } 507 } 508 509 if let Some(idx) = start_delete_idx { 510 for item in &cache_index[idx..] { 511 let (result, path, entity) = match item { 512 CacheEntry::Recognized { path, .. } 513 | CacheEntry::Unrecognized { 514 path, 515 is_dir: false, 516 } => (fs::remove_file(path), path, "file"), 517 CacheEntry::Unrecognized { path, is_dir: true } => { 518 (fs::remove_dir_all(path), path, "directory") 519 } 520 }; 521 if let Err(err) = result { 522 warn!( 523 "Failed to remove {} during cleanup, path: {}, err: {}", 524 entity, 525 path.display(), 526 err 527 ); 528 } 529 } 530 } 531 532 trace!("Task finished: clean up cache"); 533 } 534 535 // Be fault tolerant: list as much as you can, and ignore the rest 536 fn list_cache_contents(&self) -> Vec<CacheEntry> { 537 fn enter_dir( 538 vec: &mut Vec<CacheEntry>, 539 dir_path: &Path, 540 level: u8, 541 cache_config: &CacheConfig, 542 ) { 543 macro_rules! add_unrecognized { 544 (file: $path:expr) => { 545 add_unrecognized!(false, $path) 546 }; 547 (dir: $path:expr) => { 548 add_unrecognized!(true, $path) 549 }; 550 ($is_dir:expr, $path:expr) => { 551 vec.push(CacheEntry::Unrecognized { 552 path: $path.to_path_buf(), 553 is_dir: $is_dir, 554 }) 555 }; 556 } 557 macro_rules! add_unrecognized_and { 558 ([ $( $ty:ident: $path:expr ),* ], $cont:stmt) => {{ 559 $( add_unrecognized!($ty: $path); )* 560 $cont 561 }}; 562 } 563 564 macro_rules! unwrap_or { 565 ($result:expr, $cont:stmt, $err_msg:expr) => { 566 unwrap_or!($result, $cont, $err_msg, dir_path) 567 }; 568 ($result:expr, $cont:stmt, $err_msg:expr, $path:expr) => { 569 unwrap_or_warn!( 570 $result, 571 $cont, 572 format!("{}, level: {}", $err_msg, level), 573 $path 574 ) 575 }; 576 } 577 578 // If we fail to list a directory, something bad is happening anyway 579 // (something touches our cache or we have disk failure) 580 // Try to delete it, so we can stay within soft limits of the cache size. 581 // This comment applies later in this function, too. 582 let it = unwrap_or!( 583 fs::read_dir(dir_path), 584 add_unrecognized_and!([dir: dir_path], return), 585 "Failed to list cache directory, deleting it" 586 ); 587 588 let mut cache_files = HashMap::new(); 589 for entry in it { 590 // read_dir() returns an iterator over results - in case some of them are errors 591 // we don't know their names, so we can't delete them. We don't want to delete 592 // the whole directory with good entries too, so we just ignore the erroneous entries. 593 let entry = unwrap_or!( 594 entry, 595 continue, 596 "Failed to read a cache dir entry (NOT deleting it, it still occupies space)" 597 ); 598 let path = entry.path(); 599 match (level, path.is_dir()) { 600 (0..=1, true) => enter_dir(vec, &path, level + 1, cache_config), 601 (0..=1, false) => { 602 if level == 0 603 && path.file_stem() == Some(OsStr::new(".cleanup")) 604 && path.extension().is_some() 605 // assume it's cleanup lock 606 && !is_fs_lock_expired( 607 Some(&entry), 608 &path, 609 cache_config.cleanup_interval(), 610 cache_config.allowed_clock_drift_for_files_from_future(), 611 ) 612 { 613 continue; // skip active lock 614 } 615 add_unrecognized!(file: path); 616 } 617 (2, false) => { 618 match path.extension().and_then(OsStr::to_str) { 619 // mod or stats file 620 None | Some("stats") => { 621 cache_files.insert(path, entry); 622 } 623 624 Some(ext) => { 625 // check if valid lock 626 let recognized = ext.starts_with("wip-") 627 && !is_fs_lock_expired( 628 Some(&entry), 629 &path, 630 cache_config.optimizing_compression_task_timeout(), 631 cache_config.allowed_clock_drift_for_files_from_future(), 632 ); 633 634 if !recognized { 635 add_unrecognized!(file: path); 636 } 637 } 638 } 639 } 640 (_, is_dir) => add_unrecognized!(is_dir, path), 641 } 642 } 643 644 // associate module with its stats & handle them 645 // assumption: just mods and stats 646 for (path, entry) in cache_files.iter() { 647 let path_buf: PathBuf; 648 let (mod_, stats_, is_mod) = match path.extension() { 649 Some(_) => { 650 path_buf = path.with_extension(""); 651 ( 652 cache_files.get(&path_buf).map(|v| (&path_buf, v)), 653 Some((path, entry)), 654 false, 655 ) 656 } 657 None => { 658 path_buf = path.with_extension("stats"); 659 ( 660 Some((path, entry)), 661 cache_files.get(&path_buf).map(|v| (&path_buf, v)), 662 true, 663 ) 664 } 665 }; 666 667 // construct a cache entry 668 match (mod_, stats_, is_mod) { 669 (Some((mod_path, mod_entry)), Some((stats_path, stats_entry)), true) => { 670 let mod_metadata = unwrap_or!( 671 mod_entry.metadata(), 672 add_unrecognized_and!([file: stats_path, file: mod_path], continue), 673 "Failed to get metadata, deleting BOTH module cache and stats files", 674 mod_path 675 ); 676 let stats_mtime = unwrap_or!( 677 stats_entry.metadata().and_then(|m| m.modified()), 678 add_unrecognized_and!( 679 [file: stats_path], 680 unwrap_or!( 681 mod_metadata.modified(), 682 add_unrecognized_and!( 683 [file: stats_path, file: mod_path], 684 continue 685 ), 686 "Failed to get mtime, deleting BOTH module cache and stats \ 687 files", 688 mod_path 689 ) 690 ), 691 "Failed to get metadata/mtime, deleting the file", 692 stats_path 693 ); 694 // .into() called for the SystemTimeStub if cfg(test) 695 #[allow(clippy::identity_conversion)] 696 vec.push(CacheEntry::Recognized { 697 path: mod_path.to_path_buf(), 698 mtime: stats_mtime.into(), 699 size: mod_metadata.len(), 700 }) 701 } 702 (Some(_), Some(_), false) => (), // was or will be handled by previous branch 703 (Some((mod_path, mod_entry)), None, _) => { 704 let (mod_metadata, mod_mtime) = unwrap_or!( 705 mod_entry 706 .metadata() 707 .and_then(|md| md.modified().map(|mt| (md, mt))), 708 add_unrecognized_and!([file: mod_path], continue), 709 "Failed to get metadata/mtime, deleting the file", 710 mod_path 711 ); 712 // .into() called for the SystemTimeStub if cfg(test) 713 #[allow(clippy::identity_conversion)] 714 vec.push(CacheEntry::Recognized { 715 path: mod_path.to_path_buf(), 716 mtime: mod_mtime.into(), 717 size: mod_metadata.len(), 718 }) 719 } 720 (None, Some((stats_path, _stats_entry)), _) => { 721 debug!("Found orphaned stats file: {}", stats_path.display()); 722 add_unrecognized!(file: stats_path); 723 } 724 _ => unreachable!(), 725 } 726 } 727 } 728 729 let mut vec = Vec::new(); 730 enter_dir( 731 &mut vec, 732 self.cache_config.directory(), 733 0, 734 &self.cache_config, 735 ); 736 vec 737 } 738 } 739 740 fn read_stats_file(path: &Path) -> Option<ModuleCacheStatistics> { 741 fs::read(path) 742 .map_err(|err| { 743 trace!( 744 "Failed to read stats file, path: {}, err: {}", 745 path.display(), 746 err 747 ) 748 }) 749 .and_then(|bytes| { 750 toml::from_slice::<ModuleCacheStatistics>(&bytes[..]).map_err(|err| { 751 trace!( 752 "Failed to parse stats file, path: {}, err: {}", 753 path.display(), 754 err, 755 ) 756 }) 757 }) 758 .ok() 759 } 760 761 fn write_stats_file(path: &Path, stats: &ModuleCacheStatistics) -> bool { 762 toml::to_string_pretty(&stats) 763 .map_err(|err| { 764 warn!( 765 "Failed to serialize stats file, path: {}, err: {}", 766 path.display(), 767 err 768 ) 769 }) 770 .and_then(|serialized| { 771 if fs_write_atomic(path, "stats", serialized.as_bytes()) { 772 Ok(()) 773 } else { 774 Err(()) 775 } 776 }) 777 .is_ok() 778 } 779 780 /// Tries to acquire a lock for specific task. 781 /// 782 /// Returns Some(path) to the lock if succeeds. The task path must not 783 /// contain any extension and have file stem. 784 /// 785 /// To release a lock you need either manually rename or remove it, 786 /// or wait until it expires and cleanup task removes it. 787 /// 788 /// Note: this function is racy. Main idea is: be fault tolerant and 789 /// never block some task. The price is that we rarely do some task 790 /// more than once. 791 fn acquire_task_fs_lock( 792 task_path: &Path, 793 timeout: Duration, 794 allowed_future_drift: Duration, 795 ) -> Option<PathBuf> { 796 assert!(task_path.extension().is_none()); 797 assert!(task_path.file_stem().is_some()); 798 799 // list directory 800 let dir_path = task_path.parent()?; 801 let it = fs::read_dir(dir_path) 802 .map_err(|err| { 803 warn!( 804 "Failed to list cache directory, path: {}, err: {}", 805 dir_path.display(), 806 err 807 ) 808 }) 809 .ok()?; 810 811 // look for existing locks 812 for entry in it { 813 let entry = entry 814 .map_err(|err| { 815 warn!( 816 "Failed to list cache directory, path: {}, err: {}", 817 dir_path.display(), 818 err 819 ) 820 }) 821 .ok()?; 822 823 let path = entry.path(); 824 if path.is_dir() || path.file_stem() != task_path.file_stem() { 825 continue; 826 } 827 828 // check extension and mtime 829 match path.extension() { 830 None => continue, 831 Some(ext) => { 832 if let Some(ext_str) = ext.to_str() { 833 // if it's None, i.e. not valid UTF-8 string, then that's not our lock for sure 834 if ext_str.starts_with("wip-") 835 && !is_fs_lock_expired(Some(&entry), &path, timeout, allowed_future_drift) 836 { 837 return None; 838 } 839 } 840 } 841 } 842 } 843 844 // create the lock 845 let lock_path = task_path.with_extension(format!("wip-{}", std::process::id())); 846 let _file = fs::OpenOptions::new() 847 .create_new(true) 848 .write(true) 849 .open(&lock_path) 850 .map_err(|err| { 851 warn!( 852 "Failed to create lock file (note: it shouldn't exists): path: {}, err: {}", 853 lock_path.display(), 854 err 855 ) 856 }) 857 .ok()?; 858 859 Some(lock_path) 860 } 861 862 // we have either both, or just path; dir entry is desirable since on some platforms we can get 863 // metadata without extra syscalls 864 // futhermore: it's better to get a path if we have it instead of allocating a new one from the dir entry 865 fn is_fs_lock_expired( 866 entry: Option<&fs::DirEntry>, 867 path: &PathBuf, 868 threshold: Duration, 869 allowed_future_drift: Duration, 870 ) -> bool { 871 let mtime = match entry 872 .map_or_else(|| path.metadata(), |e| e.metadata()) 873 .and_then(|metadata| metadata.modified()) 874 { 875 Ok(mt) => mt, 876 Err(err) => { 877 warn!( 878 "Failed to get metadata/mtime, treating as an expired lock, path: {}, err: {}", 879 path.display(), 880 err 881 ); 882 return true; // can't read mtime, treat as expired, so this task will not be starved 883 } 884 }; 885 886 // DON'T use: mtime.elapsed() -- we must call SystemTime directly for the tests to be deterministic 887 match SystemTime::now().duration_since(mtime) { 888 Ok(elapsed) => elapsed >= threshold, 889 Err(err) => { 890 trace!( 891 "Found mtime in the future, treating as a not expired lock, path: {}, err: {}", 892 path.display(), 893 err 894 ); 895 // the lock is expired if the time is too far in the future 896 // it is fine to have network share and not synchronized clocks, 897 // but it's not good when user changes time in their system clock 898 err.duration() > allowed_future_drift 899 } 900 } 901 } 902 903 #[cfg(test)] 904 mod tests; 905