1 //! This module manages how the incremental compilation cache is represented in
4 //! Incremental compilation caches are managed according to a copy-on-write
5 //! strategy: Once a complete, consistent cache version is finalized, it is
6 //! never modified. Instead, when a subsequent compilation session is started,
7 //! the compiler will allocate a new version of the cache that starts out as
8 //! a copy of the previous version. Then only this new copy is modified and it
9 //! will not be visible to other processes until it is finalized. This ensures
10 //! that multiple compiler processes can be executed concurrently for the same
11 //! crate without interfering with each other or blocking each other.
13 //! More concretely this is implemented via the following protocol:
15 //! 1. For a newly started compilation session, the compiler allocates a
16 //! new `session` directory within the incremental compilation directory.
17 //! This session directory will have a unique name that ends with the suffix
18 //! "-working" and that contains a creation timestamp.
19 //! 2. Next, the compiler looks for the newest finalized session directory,
20 //! that is, a session directory from a previous compilation session that
21 //! has been marked as valid and consistent. A session directory is
22 //! considered finalized if the "-working" suffix in the directory name has
23 //! been replaced by the SVH of the crate.
24 //! 3. Once the compiler has found a valid, finalized session directory, it will
25 //! hard-link/copy its contents into the new "-working" directory. If all
26 //! goes well, it will have its own, private copy of the source directory and
27 //! subsequently not have to worry about synchronizing with other compiler
29 //! 4. Now the compiler can do its normal compilation process, which involves
30 //! reading and updating its private session directory.
31 //! 5. When compilation finishes without errors, the private session directory
32 //! will be in a state where it can be used as input for other compilation
33 //! sessions. That is, it will contain a dependency graph and cache artifacts
34 //! that are consistent with the state of the source code it was compiled
35 //! from, with no need to change them ever again. At this point, the compiler
36 //! finalizes and "publishes" its private session directory by renaming it
37 //! from "s-{timestamp}-{random}-working" to "s-{timestamp}-{SVH}".
38 //! 6. At this point the "old" session directory that we copied our data from
39 //! at the beginning of the session has become obsolete because we have just
40 //! published a more current version. Thus the compiler will delete it.
42 //! ## Garbage Collection
44 //! Naively following the above protocol might lead to old session directories
45 //! piling up if a compiler instance crashes for some reason before its able to
46 //! remove its private session directory. In order to avoid wasting disk space,
47 //! the compiler also does some garbage collection each time it is started in
48 //! incremental compilation mode. Specifically, it will scan the incremental
49 //! compilation directory for private session directories that are not in use
50 //! any more and will delete those. It will also delete any finalized session
51 //! directories for a given crate except for the most recent one.
53 //! ## Synchronization
55 //! There is some synchronization needed in order for the compiler to be able to
56 //! determine whether a given private session directory is not in used any more.
57 //! This is done by creating a lock file for each session directory and
58 //! locking it while the directory is still being used. Since file locks have
59 //! operating system support, we can rely on the lock being released if the
60 //! compiler process dies for some unexpected reason. Thus, when garbage
61 //! collecting private session directories, the collecting process can determine
62 //! whether the directory is still in use by trying to acquire a lock on the
63 //! file. If locking the file fails, the original process must still be alive.
64 //! If locking the file succeeds, we know that the owning process is not alive
65 //! any more and we can safely delete the directory.
66 //! There is still a small time window between the original process creating the
67 //! lock file and actually locking it. In order to minimize the chance that
68 //! another process tries to acquire the lock in just that instance, only
69 //! session directories that are older than a few seconds are considered for
70 //! garbage collection.
72 //! Another case that has to be considered is what happens if one process
73 //! deletes a finalized session directory that another process is currently
74 //! trying to copy from. This case is also handled via the lock file. Before
75 //! a process starts copying a finalized session directory, it will acquire a
76 //! shared lock on the directory's lock file. Any garbage collecting process,
77 //! on the other hand, will acquire an exclusive lock on the lock file.
78 //! Thus, if a directory is being collected, any reader process will fail
79 //! acquiring the shared lock and will leave the directory alone. Conversely,
80 //! if a collecting process can't acquire the exclusive lock because the
81 //! directory is currently being read from, it will leave collecting that
82 //! directory to another process at a later point in time.
83 //! The exact same scheme is also used when reading the metadata hashes file
84 //! from an extern crate. When a crate is compiled, the hash values of its
85 //! metadata are stored in a file in its session directory. When the
86 //! compilation session of another crate imports the first crate's metadata,
87 //! it also has to read in the accompanying metadata hashes. It thus will access
88 //! the finalized session directory of all crates it links to and while doing
89 //! so, it will also place a read lock on that the respective session directory
90 //! so that it won't be deleted while the metadata hashes are loaded.
94 //! This system relies on two features being available in the file system in
95 //! order to work really well: file locking and hard linking.
96 //! If hard linking is not available (like on FAT) the data in the cache
97 //! actually has to be copied at the beginning of each session.
98 //! If file locking does not work reliably (like on NFS), some of the
99 //! synchronization will go haywire.
100 //! In both cases we recommend to locate the incremental compilation directory
101 //! on a file system that supports these things.
102 //! It might be a good idea though to try and detect whether we are on an
103 //! unsupported file system and emit a warning in that case. This is not yet
106 use rustc::session::{Session, CrateDisambiguator};
107 use rustc_fs_util::{link_or_copy, LinkOrCopy};
108 use rustc_data_structures::{flock, base_n};
109 use rustc_data_structures::fx::{FxHashSet, FxHashMap};
110 use rustc_data_structures::svh::Svh;
112 use std::fs as std_fs;
115 use std::path::{Path, PathBuf};
116 use std::time::{UNIX_EPOCH, SystemTime, Duration};
118 use rand::{RngCore, thread_rng};
120 const LOCK_FILE_EXT: &str = ".lock";
121 const DEP_GRAPH_FILENAME: &str = "dep-graph.bin";
122 const WORK_PRODUCTS_FILENAME: &str = "work-products.bin";
123 const QUERY_CACHE_FILENAME: &str = "query-cache.bin";
125 // We encode integers using the following base, so they are shorter than decimal
126 // or hexadecimal numbers (we want short file and directory names). Since these
127 // numbers will be used in file names, we choose an encoding that is not
128 // case-sensitive (as opposed to base64, for example).
129 const INT_ENCODE_BASE: usize = base_n::CASE_INSENSITIVE;
131 pub fn dep_graph_path(sess: &Session) -> PathBuf {
132 in_incr_comp_dir_sess(sess, DEP_GRAPH_FILENAME)
134 pub fn dep_graph_path_from(incr_comp_session_dir: &Path) -> PathBuf {
135 in_incr_comp_dir(incr_comp_session_dir, DEP_GRAPH_FILENAME)
138 pub fn work_products_path(sess: &Session) -> PathBuf {
139 in_incr_comp_dir_sess(sess, WORK_PRODUCTS_FILENAME)
142 pub fn query_cache_path(sess: &Session) -> PathBuf {
143 in_incr_comp_dir_sess(sess, QUERY_CACHE_FILENAME)
146 pub fn lock_file_path(session_dir: &Path) -> PathBuf {
147 let crate_dir = session_dir.parent().unwrap();
149 let directory_name = session_dir.file_name().unwrap().to_string_lossy();
150 assert_no_characters_lost(&directory_name);
152 let dash_indices: Vec<_> = directory_name.match_indices("-")
155 if dash_indices.len() != 3 {
156 bug!("Encountered incremental compilation session directory with \
158 session_dir.display())
161 crate_dir.join(&directory_name[0 .. dash_indices[2]])
162 .with_extension(&LOCK_FILE_EXT[1..])
165 pub fn in_incr_comp_dir_sess(sess: &Session, file_name: &str) -> PathBuf {
166 in_incr_comp_dir(&sess.incr_comp_session_dir(), file_name)
169 pub fn in_incr_comp_dir(incr_comp_session_dir: &Path, file_name: &str) -> PathBuf {
170 incr_comp_session_dir.join(file_name)
173 /// Allocates the private session directory. The boolean in the Ok() result
174 /// indicates whether we should try loading a dep graph from the successfully
175 /// initialized directory, or not.
176 /// The post-condition of this fn is that we have a valid incremental
177 /// compilation session directory, if the result is `Ok`. A valid session
178 /// directory is one that contains a locked lock file. It may or may not contain
179 /// a dep-graph and work products from a previous session.
180 /// If the call fails, the fn may leave behind an invalid session directory.
181 /// The garbage collection will take care of it.
182 pub fn prepare_session_directory(sess: &Session,
184 crate_disambiguator: CrateDisambiguator) {
185 if sess.opts.incremental.is_none() {
189 debug!("prepare_session_directory");
191 // {incr-comp-dir}/{crate-name-and-disambiguator}
192 let crate_dir = crate_path(sess, crate_name, crate_disambiguator);
193 debug!("crate-dir: {}", crate_dir.display());
194 if create_dir(sess, &crate_dir, "crate").is_err() {
198 // Hack: canonicalize the path *after creating the directory*
199 // because, on windows, long paths can cause problems;
200 // canonicalization inserts this weird prefix that makes windows
201 // tolerate long paths.
202 let crate_dir = match crate_dir.canonicalize() {
205 sess.err(&format!("incremental compilation: error canonicalizing path `{}`: {}",
206 crate_dir.display(), err));
211 let mut source_directories_already_tried = FxHashSet::default();
214 // Generate a session directory of the form:
216 // {incr-comp-dir}/{crate-name-and-disambiguator}/s-{timestamp}-{random}-working
217 let session_dir = generate_session_dir_path(&crate_dir);
218 debug!("session-dir: {}", session_dir.display());
220 // Lock the new session directory. If this fails, return an
221 // error without retrying
222 let (directory_lock, lock_file_path) = match lock_directory(sess, &session_dir) {
227 // Now that we have the lock, we can actually create the session
229 if create_dir(sess, &session_dir, "session").is_err() {
233 // Find a suitable source directory to copy from. Ignore those that we
234 // have already tried before.
235 let source_directory = find_source_directory(&crate_dir,
236 &source_directories_already_tried);
238 let source_directory = if let Some(dir) = source_directory {
241 // There's nowhere to copy from, we're done
242 debug!("no source directory found. Continuing with empty session \
245 sess.init_incr_comp_session(session_dir, directory_lock, false);
249 debug!("attempting to copy data from source: {}",
250 source_directory.display());
254 // Try copying over all files from the source directory
255 if let Ok(allows_links) = copy_files(sess,
258 debug!("successfully copied data from: {}",
259 source_directory.display());
262 sess.warn(&format!("Hard linking files in the incremental \
263 compilation cache failed. Copying files \
264 instead. Consider moving the cache \
265 directory to a file system which supports \
266 hard linking in session dir `{}`",
267 session_dir.display())
271 sess.init_incr_comp_session(session_dir, directory_lock, true);
274 debug!("copying failed - trying next directory");
276 // Something went wrong while trying to copy/link files from the
277 // source directory. Try again with a different one.
278 source_directories_already_tried.insert(source_directory);
280 // Try to remove the session directory we just allocated. We don't
281 // know if there's any garbage in it from the failed copy action.
282 if let Err(err) = safe_remove_dir_all(&session_dir) {
283 sess.warn(&format!("Failed to delete partly initialized \
284 session dir `{}`: {}",
285 session_dir.display(),
289 delete_session_dir_lock_file(sess, &lock_file_path);
290 mem::drop(directory_lock);
296 /// This function finalizes and thus 'publishes' the session directory by
297 /// renaming it to `s-{timestamp}-{svh}` and releasing the file lock.
298 /// If there have been compilation errors, however, this function will just
299 /// delete the presumably invalid session directory.
300 pub fn finalize_session_directory(sess: &Session, svh: Svh) {
301 if sess.opts.incremental.is_none() {
305 let incr_comp_session_dir: PathBuf = sess.incr_comp_session_dir().clone();
307 if sess.has_errors() {
308 // If there have been any errors during compilation, we don't want to
309 // publish this session directory. Rather, we'll just delete it.
311 debug!("finalize_session_directory() - invalidating session directory: {}",
312 incr_comp_session_dir.display());
314 if let Err(err) = safe_remove_dir_all(&*incr_comp_session_dir) {
315 sess.warn(&format!("Error deleting incremental compilation \
316 session directory `{}`: {}",
317 incr_comp_session_dir.display(),
321 let lock_file_path = lock_file_path(&*incr_comp_session_dir);
322 delete_session_dir_lock_file(sess, &lock_file_path);
323 sess.mark_incr_comp_session_as_invalid();
326 debug!("finalize_session_directory() - session directory: {}",
327 incr_comp_session_dir.display());
329 let old_sub_dir_name = incr_comp_session_dir.file_name()
332 assert_no_characters_lost(&old_sub_dir_name);
334 // Keep the 's-{timestamp}-{random-number}' prefix, but replace the
335 // '-working' part with the SVH of the crate
336 let dash_indices: Vec<_> = old_sub_dir_name.match_indices("-")
339 if dash_indices.len() != 3 {
340 bug!("Encountered incremental compilation session directory with \
342 incr_comp_session_dir.display())
345 // State: "s-{timestamp}-{random-number}-"
346 let mut new_sub_dir_name = String::from(&old_sub_dir_name[..= dash_indices[2]]);
349 base_n::push_str(svh.as_u64() as u128, INT_ENCODE_BASE, &mut new_sub_dir_name);
351 // Create the full path
352 let new_path = incr_comp_session_dir.parent().unwrap().join(new_sub_dir_name);
353 debug!("finalize_session_directory() - new path: {}", new_path.display());
355 match std_fs::rename(&*incr_comp_session_dir, &new_path) {
357 debug!("finalize_session_directory() - directory renamed successfully");
359 // This unlocks the directory
360 sess.finalize_incr_comp_session(new_path);
363 // Warn about the error. However, no need to abort compilation now.
364 sess.warn(&format!("Error finalizing incremental compilation \
365 session directory `{}`: {}",
366 incr_comp_session_dir.display(),
369 debug!("finalize_session_directory() - error, marking as invalid");
370 // Drop the file lock, so we can garage collect
371 sess.mark_incr_comp_session_as_invalid();
375 let _ = garbage_collect_session_directories(sess);
378 pub fn delete_all_session_dir_contents(sess: &Session) -> io::Result<()> {
379 let sess_dir_iterator = sess.incr_comp_session_dir().read_dir()?;
380 for entry in sess_dir_iterator {
382 safe_remove_file(&entry.path())?
387 fn copy_files(sess: &Session,
390 -> Result<bool, ()> {
391 // We acquire a shared lock on the lock file of the directory, so that
392 // nobody deletes it out from under us while we are reading from it.
393 let lock_file_path = lock_file_path(source_dir);
394 let _lock = if let Ok(lock) = flock::Lock::new(&lock_file_path,
395 false, // don't wait,
396 false, // don't create
397 false) { // not exclusive
400 // Could not acquire the lock, don't try to copy from here
404 let source_dir_iterator = match source_dir.read_dir() {
406 Err(_) => return Err(())
409 let mut files_linked = 0;
410 let mut files_copied = 0;
412 for entry in source_dir_iterator {
415 let file_name = entry.file_name();
417 let target_file_path = target_dir.join(file_name);
418 let source_path = entry.path();
420 debug!("copying into session dir: {}", source_path.display());
421 match link_or_copy(source_path, target_file_path) {
422 Ok(LinkOrCopy::Link) => {
425 Ok(LinkOrCopy::Copy) => {
428 Err(_) => return Err(())
437 if sess.opts.debugging_opts.incremental_info {
438 println!("[incremental] session directory: \
439 {} files hard-linked", files_linked);
440 println!("[incremental] session directory: \
441 {} files copied", files_copied);
444 Ok(files_linked > 0 || files_copied == 0)
447 /// Generate unique directory path of the form:
448 /// {crate_dir}/s-{timestamp}-{random-number}-working
449 fn generate_session_dir_path(crate_dir: &Path) -> PathBuf {
450 let timestamp = timestamp_to_string(SystemTime::now());
451 debug!("generate_session_dir_path: timestamp = {}", timestamp);
452 let random_number = thread_rng().next_u32();
453 debug!("generate_session_dir_path: random_number = {}", random_number);
455 let directory_name = format!("s-{}-{}-working",
457 base_n::encode(random_number as u128,
459 debug!("generate_session_dir_path: directory_name = {}", directory_name);
460 let directory_path = crate_dir.join(directory_name);
461 debug!("generate_session_dir_path: directory_path = {}", directory_path.display());
465 fn create_dir(sess: &Session, path: &Path, dir_tag: &str) -> Result<(),()> {
466 match std_fs::create_dir_all(path) {
468 debug!("{} directory created successfully", dir_tag);
472 sess.err(&format!("Could not create incremental compilation {} \
482 /// Allocate the lock-file and lock it.
483 fn lock_directory(sess: &Session,
485 -> Result<(flock::Lock, PathBuf), ()> {
486 let lock_file_path = lock_file_path(session_dir);
487 debug!("lock_directory() - lock_file: {}", lock_file_path.display());
489 match flock::Lock::new(&lock_file_path,
491 true, // create the lock file
492 true) { // the lock should be exclusive
493 Ok(lock) => Ok((lock, lock_file_path)),
495 sess.err(&format!("incremental compilation: could not create \
496 session directory lock file: {}", err));
502 fn delete_session_dir_lock_file(sess: &Session,
503 lock_file_path: &Path) {
504 if let Err(err) = safe_remove_file(&lock_file_path) {
505 sess.warn(&format!("Error deleting lock file for incremental \
506 compilation session directory `{}`: {}",
507 lock_file_path.display(),
512 /// Find the most recent published session directory that is not in the
514 fn find_source_directory(crate_dir: &Path,
515 source_directories_already_tried: &FxHashSet<PathBuf>)
517 let iter = crate_dir.read_dir()
519 .filter_map(|e| e.ok().map(|e| e.path()));
521 find_source_directory_in_iter(iter, source_directories_already_tried)
524 fn find_source_directory_in_iter<I>(iter: I,
525 source_directories_already_tried: &FxHashSet<PathBuf>)
527 where I: Iterator<Item=PathBuf>
529 let mut best_candidate = (UNIX_EPOCH, None);
531 for session_dir in iter {
532 debug!("find_source_directory_in_iter - inspecting `{}`",
533 session_dir.display());
535 let directory_name = session_dir.file_name().unwrap().to_string_lossy();
536 assert_no_characters_lost(&directory_name);
538 if source_directories_already_tried.contains(&session_dir) ||
539 !is_session_directory(&directory_name) ||
540 !is_finalized(&directory_name) {
541 debug!("find_source_directory_in_iter - ignoring.");
545 let timestamp = extract_timestamp_from_session_dir(&directory_name)
546 .unwrap_or_else(|_| {
547 bug!("unexpected incr-comp session dir: {}", session_dir.display())
550 if timestamp > best_candidate.0 {
551 best_candidate = (timestamp, Some(session_dir.clone()));
558 fn is_finalized(directory_name: &str) -> bool {
559 !directory_name.ends_with("-working")
562 fn is_session_directory(directory_name: &str) -> bool {
563 directory_name.starts_with("s-") &&
564 !directory_name.ends_with(LOCK_FILE_EXT)
567 fn is_session_directory_lock_file(file_name: &str) -> bool {
568 file_name.starts_with("s-") && file_name.ends_with(LOCK_FILE_EXT)
571 fn extract_timestamp_from_session_dir(directory_name: &str)
572 -> Result<SystemTime, ()> {
573 if !is_session_directory(directory_name) {
577 let dash_indices: Vec<_> = directory_name.match_indices("-")
580 if dash_indices.len() != 3 {
584 string_to_timestamp(&directory_name[dash_indices[0]+1 .. dash_indices[1]])
587 fn timestamp_to_string(timestamp: SystemTime) -> String {
588 let duration = timestamp.duration_since(UNIX_EPOCH).unwrap();
589 let micros = duration.as_secs() * 1_000_000 +
590 (duration.subsec_nanos() as u64) / 1000;
591 base_n::encode(micros as u128, INT_ENCODE_BASE)
594 fn string_to_timestamp(s: &str) -> Result<SystemTime, ()> {
595 let micros_since_unix_epoch = u64::from_str_radix(s, INT_ENCODE_BASE as u32);
597 if micros_since_unix_epoch.is_err() {
601 let micros_since_unix_epoch = micros_since_unix_epoch.unwrap();
603 let duration = Duration::new(micros_since_unix_epoch / 1_000_000,
604 1000 * (micros_since_unix_epoch % 1_000_000) as u32);
605 Ok(UNIX_EPOCH + duration)
608 fn crate_path(sess: &Session,
610 crate_disambiguator: CrateDisambiguator)
613 let incr_dir = sess.opts.incremental.as_ref().unwrap().clone();
615 // The full crate disambiguator is really long. 64 bits of it should be
617 let crate_disambiguator = crate_disambiguator.to_fingerprint().to_smaller_hash();
618 let crate_disambiguator = base_n::encode(crate_disambiguator as u128,
621 let crate_name = format!("{}-{}", crate_name, crate_disambiguator);
622 incr_dir.join(crate_name)
625 fn assert_no_characters_lost(s: &str) {
626 if s.contains('\u{FFFD}') {
627 bug!("Could not losslessly convert '{}'.", s)
631 fn is_old_enough_to_be_collected(timestamp: SystemTime) -> bool {
632 timestamp < SystemTime::now() - Duration::from_secs(10)
635 pub fn garbage_collect_session_directories(sess: &Session) -> io::Result<()> {
636 debug!("garbage_collect_session_directories() - begin");
638 let session_directory = sess.incr_comp_session_dir();
639 debug!("garbage_collect_session_directories() - session directory: {}",
640 session_directory.display());
642 let crate_directory = session_directory.parent().unwrap();
643 debug!("garbage_collect_session_directories() - crate directory: {}",
644 crate_directory.display());
646 // First do a pass over the crate directory, collecting lock files and
647 // session directories
648 let mut session_directories = FxHashSet::default();
649 let mut lock_files = FxHashSet::default();
651 for dir_entry in crate_directory.read_dir()? {
652 let dir_entry = match dir_entry {
653 Ok(dir_entry) => dir_entry,
660 let entry_name = dir_entry.file_name();
661 let entry_name = entry_name.to_string_lossy();
663 if is_session_directory_lock_file(&entry_name) {
664 assert_no_characters_lost(&entry_name);
665 lock_files.insert(entry_name.into_owned());
666 } else if is_session_directory(&entry_name) {
667 assert_no_characters_lost(&entry_name);
668 session_directories.insert(entry_name.into_owned());
670 // This is something we don't know, leave it alone
674 // Now map from lock files to session directories
675 let lock_file_to_session_dir: FxHashMap<String, Option<String>> =
676 lock_files.into_iter()
677 .map(|lock_file_name| {
678 assert!(lock_file_name.ends_with(LOCK_FILE_EXT));
679 let dir_prefix_end = lock_file_name.len() - LOCK_FILE_EXT.len();
681 let dir_prefix = &lock_file_name[0 .. dir_prefix_end];
682 session_directories.iter()
683 .find(|dir_name| dir_name.starts_with(dir_prefix))
685 (lock_file_name, session_dir.map(String::clone))
689 // Delete all lock files, that don't have an associated directory. They must
690 // be some kind of leftover
691 for (lock_file_name, directory_name) in &lock_file_to_session_dir {
692 if directory_name.is_none() {
693 let timestamp = match extract_timestamp_from_session_dir(lock_file_name) {
694 Ok(timestamp) => timestamp,
696 debug!("Found lock-file with malformed timestamp: {}",
697 crate_directory.join(&lock_file_name).display());
703 let lock_file_path = crate_directory.join(&**lock_file_name);
705 if is_old_enough_to_be_collected(timestamp) {
706 debug!("garbage_collect_session_directories() - deleting \
707 garbage lock file: {}", lock_file_path.display());
708 delete_session_dir_lock_file(sess, &lock_file_path);
710 debug!("garbage_collect_session_directories() - lock file with \
711 no session dir not old enough to be collected: {}",
712 lock_file_path.display());
717 // Filter out `None` directories
718 let lock_file_to_session_dir: FxHashMap<String, String> =
719 lock_file_to_session_dir.into_iter()
720 .filter_map(|(lock_file_name, directory_name)| {
721 directory_name.map(|n| (lock_file_name, n))
725 // Delete all session directories that don't have a lock file.
726 for directory_name in session_directories {
727 if !lock_file_to_session_dir.values().any(|dir| *dir == directory_name) {
728 let path = crate_directory.join(directory_name);
729 if let Err(err) = safe_remove_dir_all(&path) {
730 sess.warn(&format!("Failed to garbage collect invalid incremental \
731 compilation session directory `{}`: {}",
738 // Now garbage collect the valid session directories.
739 let mut deletion_candidates = vec![];
740 let mut definitely_delete = vec![];
742 for (lock_file_name, directory_name) in &lock_file_to_session_dir {
743 debug!("garbage_collect_session_directories() - inspecting: {}",
746 let timestamp = match extract_timestamp_from_session_dir(directory_name) {
747 Ok(timestamp) => timestamp,
749 debug!("Found session-dir with malformed timestamp: {}",
750 crate_directory.join(directory_name).display());
756 if is_finalized(directory_name) {
757 let lock_file_path = crate_directory.join(lock_file_name);
758 match flock::Lock::new(&lock_file_path,
760 false, // don't create the lock-file
761 true) { // get an exclusive lock
763 debug!("garbage_collect_session_directories() - \
764 successfully acquired lock");
765 debug!("garbage_collect_session_directories() - adding \
766 deletion candidate: {}", directory_name);
768 // Note that we are holding on to the lock
769 deletion_candidates.push((timestamp,
770 crate_directory.join(directory_name),
774 debug!("garbage_collect_session_directories() - \
775 not collecting, still in use");
778 } else if is_old_enough_to_be_collected(timestamp) {
779 // When cleaning out "-working" session directories, i.e.
780 // session directories that might still be in use by another
781 // compiler instance, we only look a directories that are
782 // at least ten seconds old. This is supposed to reduce the
783 // chance of deleting a directory in the time window where
784 // the process has allocated the directory but has not yet
785 // acquired the file-lock on it.
787 // Try to acquire the directory lock. If we can't, it
788 // means that the owning process is still alive and we
789 // leave this directory alone.
790 let lock_file_path = crate_directory.join(lock_file_name);
791 match flock::Lock::new(&lock_file_path,
793 false, // don't create the lock-file
794 true) { // get an exclusive lock
796 debug!("garbage_collect_session_directories() - \
797 successfully acquired lock");
799 // Note that we are holding on to the lock
800 definitely_delete.push((crate_directory.join(directory_name),
804 debug!("garbage_collect_session_directories() - \
805 not collecting, still in use");
809 debug!("garbage_collect_session_directories() - not finalized, not \
814 // Delete all but the most recent of the candidates
815 for (path, lock) in all_except_most_recent(deletion_candidates) {
816 debug!("garbage_collect_session_directories() - deleting `{}`",
819 if let Err(err) = safe_remove_dir_all(&path) {
820 sess.warn(&format!("Failed to garbage collect finalized incremental \
821 compilation session directory `{}`: {}",
825 delete_session_dir_lock_file(sess, &lock_file_path(&path));
829 // Let's make it explicit that the file lock is released at this point,
830 // or rather, that we held on to it until here
834 for (path, lock) in definitely_delete {
835 debug!("garbage_collect_session_directories() - deleting `{}`",
838 if let Err(err) = safe_remove_dir_all(&path) {
839 sess.warn(&format!("Failed to garbage collect incremental \
840 compilation session directory `{}`: {}",
844 delete_session_dir_lock_file(sess, &lock_file_path(&path));
847 // Let's make it explicit that the file lock is released at this point,
848 // or rather, that we held on to it until here
855 fn all_except_most_recent(deletion_candidates: Vec<(SystemTime, PathBuf, Option<flock::Lock>)>)
856 -> FxHashMap<PathBuf, Option<flock::Lock>> {
857 let most_recent = deletion_candidates.iter()
858 .map(|&(timestamp, ..)| timestamp)
861 if let Some(most_recent) = most_recent {
862 deletion_candidates.into_iter()
863 .filter(|&(timestamp, ..)| timestamp != most_recent)
864 .map(|(_, path, lock)| (path, lock))
871 /// Since paths of artifacts within session directories can get quite long, we
872 /// need to support deleting files with very long paths. The regular
873 /// WinApi functions only support paths up to 260 characters, however. In order
874 /// to circumvent this limitation, we canonicalize the path of the directory
875 /// before passing it to std::fs::remove_dir_all(). This will convert the path
876 /// into the '\\?\' format, which supports much longer paths.
877 fn safe_remove_dir_all(p: &Path) -> io::Result<()> {
879 let canonicalized = p.canonicalize()?;
880 std_fs::remove_dir_all(canonicalized)
886 fn safe_remove_file(p: &Path) -> io::Result<()> {
888 let canonicalized = p.canonicalize()?;
889 std_fs::remove_file(canonicalized)
896 fn test_all_except_most_recent() {
897 assert_eq!(all_except_most_recent(
899 (UNIX_EPOCH + Duration::new(4, 0), PathBuf::from("4"), None),
900 (UNIX_EPOCH + Duration::new(1, 0), PathBuf::from("1"), None),
901 (UNIX_EPOCH + Duration::new(5, 0), PathBuf::from("5"), None),
902 (UNIX_EPOCH + Duration::new(3, 0), PathBuf::from("3"), None),
903 (UNIX_EPOCH + Duration::new(2, 0), PathBuf::from("2"), None),
904 ]).keys().cloned().collect::<FxHashSet<PathBuf>>(),
910 ].into_iter().collect::<FxHashSet<PathBuf>>()
913 assert_eq!(all_except_most_recent(
915 ]).keys().cloned().collect::<FxHashSet<PathBuf>>(),
921 fn test_timestamp_serialization() {
922 for i in 0 .. 1_000u64 {
923 let time = UNIX_EPOCH + Duration::new(i * 1_434_578, (i as u32) * 239_000);
924 let s = timestamp_to_string(time);
925 assert_eq!(Ok(time), string_to_timestamp(&s));
930 fn test_find_source_directory_in_iter() {
931 let already_visited = FxHashSet::default();
934 assert_eq!(find_source_directory_in_iter(
935 vec![PathBuf::from("crate-dir/s-3234-0000-svh"),
936 PathBuf::from("crate-dir/s-2234-0000-svh"),
937 PathBuf::from("crate-dir/s-1234-0000-svh")].into_iter(), &already_visited),
938 Some(PathBuf::from("crate-dir/s-3234-0000-svh")));
940 // Filter out "-working"
941 assert_eq!(find_source_directory_in_iter(
942 vec![PathBuf::from("crate-dir/s-3234-0000-working"),
943 PathBuf::from("crate-dir/s-2234-0000-svh"),
944 PathBuf::from("crate-dir/s-1234-0000-svh")].into_iter(), &already_visited),
945 Some(PathBuf::from("crate-dir/s-2234-0000-svh")));
948 assert_eq!(find_source_directory_in_iter(vec![].into_iter(), &already_visited),
951 // Handle only working
952 assert_eq!(find_source_directory_in_iter(
953 vec![PathBuf::from("crate-dir/s-3234-0000-working"),
954 PathBuf::from("crate-dir/s-2234-0000-working"),
955 PathBuf::from("crate-dir/s-1234-0000-working")].into_iter(), &already_visited),