1 // Copyright (c) 2011 The LevelDB Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file. See the AUTHORS file for names of contributors.
5 #include "db/version_set.h"
9 #include "db/filename.h"
10 #include "db/log_reader.h"
11 #include "db/log_writer.h"
12 #include "db/memtable.h"
13 #include "db/table_cache.h"
14 #include "leveldb/env.h"
15 #include "leveldb/table_builder.h"
16 #include "table/merger.h"
17 #include "table/two_level_iterator.h"
18 #include "util/coding.h"
19 #include "util/logging.h"
23 static const int kTargetFileSize = 2 * 1048576;
25 // Maximum bytes of overlaps in grandparent (i.e., level+2) before we
26 // stop building a single file in a level->level+1 compaction.
27 static const int64_t kMaxGrandParentOverlapBytes = 10 * kTargetFileSize;
29 static double MaxBytesForLevel(int level) {
30 // Note: the result for level zero is not really used since we set
31 // the level-0 compaction threshold based on number of files.
32 double result = 10 * 1048576.0; // Result for both level-0 and level-1
40 static uint64_t MaxFileSizeForLevel(int level) {
41 return kTargetFileSize; // We could vary per level to reduce number of files?
44 static int64_t TotalFileSize(const std::vector<FileMetaData*>& files) {
46 for (size_t i = 0; i < files.size(); i++) {
47 sum += files[i]->file_size;
53 std::string IntSetToString(const std::set<uint64_t>& s) {
54 std::string result = "{";
55 for (std::set<uint64_t>::const_iterator it = s.begin();
58 result += (result.size() > 1) ? "," : "";
59 result += NumberToString(*it);
69 // Remove from linked list
73 // Drop references to files
74 for (int level = 0; level < config::kNumLevels; level++) {
75 for (size_t i = 0; i < files_[level].size(); i++) {
76 FileMetaData* f = files_[level][i];
86 int FindFile(const InternalKeyComparator& icmp,
87 const std::vector<FileMetaData*>& files,
90 uint32_t right = files.size();
91 while (left < right) {
92 uint32_t mid = (left + right) / 2;
93 const FileMetaData* f = files[mid];
94 if (icmp.InternalKeyComparator::Compare(f->largest.Encode(), key) < 0) {
95 // Key at "mid.largest" is < "target". Therefore all
96 // files at or before "mid" are uninteresting.
99 // Key at "mid.largest" is >= "target". Therefore all files
100 // after "mid" are uninteresting.
107 static bool AfterFile(const Comparator* ucmp,
108 const Slice* user_key, const FileMetaData* f) {
109 // NULL user_key occurs before all keys and is therefore never after *f
110 return (user_key != NULL &&
111 ucmp->Compare(*user_key, f->largest.user_key()) > 0);
114 static bool BeforeFile(const Comparator* ucmp,
115 const Slice* user_key, const FileMetaData* f) {
116 // NULL user_key occurs after all keys and is therefore never before *f
117 return (user_key != NULL &&
118 ucmp->Compare(*user_key, f->smallest.user_key()) < 0);
121 bool SomeFileOverlapsRange(
122 const InternalKeyComparator& icmp,
123 bool disjoint_sorted_files,
124 const std::vector<FileMetaData*>& files,
125 const Slice* smallest_user_key,
126 const Slice* largest_user_key) {
127 const Comparator* ucmp = icmp.user_comparator();
128 if (!disjoint_sorted_files) {
129 // Need to check against all files
130 for (int i = 0; i < files.size(); i++) {
131 const FileMetaData* f = files[i];
132 if (AfterFile(ucmp, smallest_user_key, f) ||
133 BeforeFile(ucmp, largest_user_key, f)) {
136 return true; // Overlap
142 // Binary search over file list
144 if (smallest_user_key != NULL) {
145 // Find the earliest possible internal key for smallest_user_key
146 InternalKey small(*smallest_user_key, kMaxSequenceNumber,kValueTypeForSeek);
147 index = FindFile(icmp, files, small.Encode());
150 if (index >= files.size()) {
151 // beginning of range is after all files, so no overlap.
155 return !BeforeFile(ucmp, largest_user_key, files[index]);
158 // An internal iterator. For a given version/level pair, yields
159 // information about the files in the level. For a given entry, key()
160 // is the largest key that occurs in the file, and value() is an
161 // 16-byte value containing the file number and file size, both
162 // encoded using EncodeFixed64.
163 class Version::LevelFileNumIterator : public Iterator {
165 LevelFileNumIterator(const InternalKeyComparator& icmp,
166 const std::vector<FileMetaData*>* flist)
169 index_(flist->size()) { // Marks as invalid
171 virtual bool Valid() const {
172 return index_ < flist_->size();
174 virtual void Seek(const Slice& target) {
175 index_ = FindFile(icmp_, *flist_, target);
177 virtual void SeekToFirst() { index_ = 0; }
178 virtual void SeekToLast() {
179 index_ = flist_->empty() ? 0 : flist_->size() - 1;
181 virtual void Next() {
185 virtual void Prev() {
188 index_ = flist_->size(); // Marks as invalid
195 return (*flist_)[index_]->largest.Encode();
197 Slice value() const {
199 EncodeFixed64(value_buf_, (*flist_)[index_]->number);
200 EncodeFixed64(value_buf_+8, (*flist_)[index_]->file_size);
201 return Slice(value_buf_, sizeof(value_buf_));
203 virtual Status status() const { return Status::OK(); }
205 const InternalKeyComparator icmp_;
206 const std::vector<FileMetaData*>* const flist_;
209 // Backing store for value(). Holds the file number and size.
210 mutable char value_buf_[16];
213 static Iterator* GetFileIterator(void* arg,
214 const ReadOptions& options,
215 const Slice& file_value) {
216 TableCache* cache = reinterpret_cast<TableCache*>(arg);
217 if (file_value.size() != 16) {
218 return NewErrorIterator(
219 Status::Corruption("FileReader invoked with unexpected value"));
221 return cache->NewIterator(options,
222 DecodeFixed64(file_value.data()),
223 DecodeFixed64(file_value.data() + 8));
227 Iterator* Version::NewConcatenatingIterator(const ReadOptions& options,
229 return NewTwoLevelIterator(
230 new LevelFileNumIterator(vset_->icmp_, &files_[level]),
231 &GetFileIterator, vset_->table_cache_, options);
234 void Version::AddIterators(const ReadOptions& options,
235 std::vector<Iterator*>* iters) {
236 // Merge all level zero files together since they may overlap
237 for (size_t i = 0; i < files_[0].size(); i++) {
239 vset_->table_cache_->NewIterator(
240 options, files_[0][i]->number, files_[0][i]->file_size));
243 // For levels > 0, we can use a concatenating iterator that sequentially
244 // walks through the non-overlapping files in the level, opening them
246 for (int level = 1; level < config::kNumLevels; level++) {
247 if (!files_[level].empty()) {
248 iters->push_back(NewConcatenatingIterator(options, level));
253 // If "*iter" points at a value or deletion for user_key, store
254 // either the value, or a NotFound error and return true.
255 // Else return false.
256 static bool GetValue(const Comparator* cmp,
257 Iterator* iter, const Slice& user_key,
260 if (!iter->Valid()) {
263 ParsedInternalKey parsed_key;
264 if (!ParseInternalKey(iter->key(), &parsed_key)) {
265 *s = Status::Corruption("corrupted key for ", user_key);
268 if (cmp->Compare(parsed_key.user_key, user_key) != 0) {
271 switch (parsed_key.type) {
273 *s = Status::NotFound(Slice()); // Use an empty error message for speed
276 Slice v = iter->value();
277 value->assign(v.data(), v.size());
284 static bool NewestFirst(FileMetaData* a, FileMetaData* b) {
285 return a->number > b->number;
288 Status Version::Get(const ReadOptions& options,
292 Slice ikey = k.internal_key();
293 Slice user_key = k.user_key();
294 const Comparator* ucmp = vset_->icmp_.user_comparator();
297 stats->seek_file = NULL;
298 stats->seek_file_level = -1;
299 FileMetaData* last_file_read = NULL;
300 int last_file_read_level = -1;
302 // We can search level-by-level since entries never hop across
303 // levels. Therefore we are guaranteed that if we find data
304 // in an smaller level, later levels are irrelevant.
305 std::vector<FileMetaData*> tmp;
307 for (int level = 0; level < config::kNumLevels; level++) {
308 size_t num_files = files_[level].size();
309 if (num_files == 0) continue;
311 // Get the list of files to search in this level
312 FileMetaData* const* files = &files_[level][0];
314 // Level-0 files may overlap each other. Find all files that
315 // overlap user_key and process them in order from newest to oldest.
316 tmp.reserve(num_files);
317 for (uint32_t i = 0; i < num_files; i++) {
318 FileMetaData* f = files[i];
319 if (ucmp->Compare(user_key, f->smallest.user_key()) >= 0 &&
320 ucmp->Compare(user_key, f->largest.user_key()) <= 0) {
324 if (tmp.empty()) continue;
326 std::sort(tmp.begin(), tmp.end(), NewestFirst);
328 num_files = tmp.size();
330 // Binary search to find earliest index whose largest key >= ikey.
331 uint32_t index = FindFile(vset_->icmp_, files_[level], ikey);
332 if (index >= num_files) {
337 if (ucmp->Compare(user_key, tmp2->smallest.user_key()) < 0) {
338 // All of "tmp2" is past any data for user_key
348 for (uint32_t i = 0; i < num_files; ++i) {
349 if (last_file_read != NULL && stats->seek_file == NULL) {
350 // We have had more than one seek for this read. Charge the 1st file.
351 stats->seek_file = last_file_read;
352 stats->seek_file_level = last_file_read_level;
355 FileMetaData* f = files[i];
357 last_file_read_level = level;
359 Iterator* iter = vset_->table_cache_->NewIterator(
364 const bool done = GetValue(ucmp, iter, user_key, value, &s);
365 if (!iter->status().ok()) {
378 return Status::NotFound(Slice()); // Use an empty error message for speed
381 bool Version::UpdateStats(const GetStats& stats) {
382 FileMetaData* f = stats.seek_file;
385 if (f->allowed_seeks <= 0 && file_to_compact_ == NULL) {
386 file_to_compact_ = f;
387 file_to_compact_level_ = stats.seek_file_level;
394 void Version::Ref() {
398 void Version::Unref() {
399 assert(this != &vset_->dummy_versions_);
407 bool Version::OverlapInLevel(int level,
408 const Slice* smallest_user_key,
409 const Slice* largest_user_key) {
410 return SomeFileOverlapsRange(vset_->icmp_, (level > 0), files_[level],
411 smallest_user_key, largest_user_key);
414 int Version::PickLevelForMemTableOutput(
415 const Slice& smallest_user_key,
416 const Slice& largest_user_key) {
418 if (!OverlapInLevel(0, &smallest_user_key, &largest_user_key)) {
419 // Push to next level if there is no overlap in next level,
420 // and the #bytes overlapping in the level after that are limited.
421 InternalKey start(smallest_user_key, kMaxSequenceNumber, kValueTypeForSeek);
422 InternalKey limit(largest_user_key, 0, static_cast<ValueType>(0));
423 std::vector<FileMetaData*> overlaps;
424 while (level < config::kMaxMemCompactLevel) {
425 if (OverlapInLevel(level + 1, &smallest_user_key, &largest_user_key)) {
428 GetOverlappingInputs(level + 2, &start, &limit, &overlaps);
429 const int64_t sum = TotalFileSize(overlaps);
430 if (sum > kMaxGrandParentOverlapBytes) {
439 // Store in "*inputs" all files in "level" that overlap [begin,end]
440 void Version::GetOverlappingInputs(
442 const InternalKey* begin,
443 const InternalKey* end,
444 std::vector<FileMetaData*>* inputs) {
446 Slice user_begin, user_end;
448 user_begin = begin->user_key();
451 user_end = end->user_key();
453 const Comparator* user_cmp = vset_->icmp_.user_comparator();
454 for (size_t i = 0; i < files_[level].size(); ) {
455 FileMetaData* f = files_[level][i++];
456 const Slice file_start = f->smallest.user_key();
457 const Slice file_limit = f->largest.user_key();
458 if (begin != NULL && user_cmp->Compare(file_limit, user_begin) < 0) {
459 // "f" is completely before specified range; skip it
460 } else if (end != NULL && user_cmp->Compare(file_start, user_end) > 0) {
461 // "f" is completely after specified range; skip it
463 inputs->push_back(f);
465 // Level-0 files may overlap each other. So check if the newly
466 // added file has expanded the range. If so, restart search.
467 if (begin != NULL && user_cmp->Compare(file_start, user_begin) < 0) {
468 user_begin = file_start;
471 } else if (end != NULL && user_cmp->Compare(file_limit, user_end) > 0) {
472 user_end = file_limit;
481 std::string Version::DebugString() const {
483 for (int level = 0; level < config::kNumLevels; level++) {
486 // 17:123['a' .. 'd']
488 r.append("--- level ");
489 AppendNumberTo(&r, level);
491 const std::vector<FileMetaData*>& files = files_[level];
492 for (size_t i = 0; i < files.size(); i++) {
494 AppendNumberTo(&r, files[i]->number);
496 AppendNumberTo(&r, files[i]->file_size);
498 r.append(files[i]->smallest.DebugString());
500 r.append(files[i]->largest.DebugString());
507 // A helper class so we can efficiently apply a whole sequence
508 // of edits to a particular state without creating intermediate
509 // Versions that contain full copies of the intermediate state.
510 class VersionSet::Builder {
512 // Helper to sort by v->files_[file_number].smallest
513 struct BySmallestKey {
514 const InternalKeyComparator* internal_comparator;
516 bool operator()(FileMetaData* f1, FileMetaData* f2) const {
517 int r = internal_comparator->Compare(f1->smallest, f2->smallest);
521 // Break ties by file number
522 return (f1->number < f2->number);
527 typedef std::set<FileMetaData*, BySmallestKey> FileSet;
529 std::set<uint64_t> deleted_files;
530 FileSet* added_files;
535 LevelState levels_[config::kNumLevels];
538 // Initialize a builder with the files from *base and other info from *vset
539 Builder(VersionSet* vset, Version* base)
544 cmp.internal_comparator = &vset_->icmp_;
545 for (int level = 0; level < config::kNumLevels; level++) {
546 levels_[level].added_files = new FileSet(cmp);
551 for (int level = 0; level < config::kNumLevels; level++) {
552 const FileSet* added = levels_[level].added_files;
553 std::vector<FileMetaData*> to_unref;
554 to_unref.reserve(added->size());
555 for (FileSet::const_iterator it = added->begin();
556 it != added->end(); ++it) {
557 to_unref.push_back(*it);
560 for (uint32_t i = 0; i < to_unref.size(); i++) {
561 FileMetaData* f = to_unref[i];
571 // Apply all of the edits in *edit to the current state.
572 void Apply(VersionEdit* edit) {
573 // Update compaction pointers
574 for (size_t i = 0; i < edit->compact_pointers_.size(); i++) {
575 const int level = edit->compact_pointers_[i].first;
576 vset_->compact_pointer_[level] =
577 edit->compact_pointers_[i].second.Encode().ToString();
581 const VersionEdit::DeletedFileSet& del = edit->deleted_files_;
582 for (VersionEdit::DeletedFileSet::const_iterator iter = del.begin();
585 const int level = iter->first;
586 const uint64_t number = iter->second;
587 levels_[level].deleted_files.insert(number);
591 for (size_t i = 0; i < edit->new_files_.size(); i++) {
592 const int level = edit->new_files_[i].first;
593 FileMetaData* f = new FileMetaData(edit->new_files_[i].second);
596 // We arrange to automatically compact this file after
597 // a certain number of seeks. Let's assume:
598 // (1) One seek costs 10ms
599 // (2) Writing or reading 1MB costs 10ms (100MB/s)
600 // (3) A compaction of 1MB does 25MB of IO:
601 // 1MB read from this level
602 // 10-12MB read from next level (boundaries may be misaligned)
603 // 10-12MB written to next level
604 // This implies that 25 seeks cost the same as the compaction
605 // of 1MB of data. I.e., one seek costs approximately the
606 // same as the compaction of 40KB of data. We are a little
607 // conservative and allow approximately one seek for every 16KB
608 // of data before triggering a compaction.
609 f->allowed_seeks = (f->file_size / 16384);
610 if (f->allowed_seeks < 100) f->allowed_seeks = 100;
612 levels_[level].deleted_files.erase(f->number);
613 levels_[level].added_files->insert(f);
617 // Save the current state in *v.
618 void SaveTo(Version* v) {
620 cmp.internal_comparator = &vset_->icmp_;
621 for (int level = 0; level < config::kNumLevels; level++) {
622 // Merge the set of added files with the set of pre-existing files.
623 // Drop any deleted files. Store the result in *v.
624 const std::vector<FileMetaData*>& base_files = base_->files_[level];
625 std::vector<FileMetaData*>::const_iterator base_iter = base_files.begin();
626 std::vector<FileMetaData*>::const_iterator base_end = base_files.end();
627 const FileSet* added = levels_[level].added_files;
628 v->files_[level].reserve(base_files.size() + added->size());
629 for (FileSet::const_iterator added_iter = added->begin();
630 added_iter != added->end();
632 // Add all smaller files listed in base_
633 for (std::vector<FileMetaData*>::const_iterator bpos
634 = std::upper_bound(base_iter, base_end, *added_iter, cmp);
637 MaybeAddFile(v, level, *base_iter);
640 MaybeAddFile(v, level, *added_iter);
643 // Add remaining base files
644 for (; base_iter != base_end; ++base_iter) {
645 MaybeAddFile(v, level, *base_iter);
649 // Make sure there is no overlap in levels > 0
651 for (uint32_t i = 1; i < v->files_[level].size(); i++) {
652 const InternalKey& prev_end = v->files_[level][i-1]->largest;
653 const InternalKey& this_begin = v->files_[level][i]->smallest;
654 if (vset_->icmp_.Compare(prev_end, this_begin) >= 0) {
655 fprintf(stderr, "overlapping ranges in same level %s vs. %s\n",
656 prev_end.DebugString().c_str(),
657 this_begin.DebugString().c_str());
666 void MaybeAddFile(Version* v, int level, FileMetaData* f) {
667 if (levels_[level].deleted_files.count(f->number) > 0) {
668 // File is deleted: do nothing
670 std::vector<FileMetaData*>* files = &v->files_[level];
671 if (level > 0 && !files->empty()) {
673 assert(vset_->icmp_.Compare((*files)[files->size()-1]->largest,
682 VersionSet::VersionSet(const std::string& dbname,
683 const Options* options,
684 TableCache* table_cache,
685 const InternalKeyComparator* cmp)
686 : env_(options->env),
689 table_cache_(table_cache),
691 next_file_number_(2),
692 manifest_file_number_(0), // Filled by Recover()
696 descriptor_file_(NULL),
697 descriptor_log_(NULL),
698 dummy_versions_(this),
700 AppendVersion(new Version(this));
703 VersionSet::~VersionSet() {
705 assert(dummy_versions_.next_ == &dummy_versions_); // List must be empty
706 delete descriptor_log_;
707 delete descriptor_file_;
710 void VersionSet::AppendVersion(Version* v) {
712 assert(v->refs_ == 0);
713 assert(v != current_);
714 if (current_ != NULL) {
720 // Append to linked list
721 v->prev_ = dummy_versions_.prev_;
722 v->next_ = &dummy_versions_;
727 Status VersionSet::LogAndApply(VersionEdit* edit, port::Mutex* mu) {
728 if (edit->has_log_number_) {
729 assert(edit->log_number_ >= log_number_);
730 assert(edit->log_number_ < next_file_number_);
732 edit->SetLogNumber(log_number_);
735 if (!edit->has_prev_log_number_) {
736 edit->SetPrevLogNumber(prev_log_number_);
739 edit->SetNextFile(next_file_number_);
740 edit->SetLastSequence(last_sequence_);
742 Version* v = new Version(this);
744 Builder builder(this, current_);
750 // Initialize new descriptor log file if necessary by creating
751 // a temporary file that contains a snapshot of the current version.
752 std::string new_manifest_file;
754 if (descriptor_log_ == NULL) {
755 // No reason to unlock *mu here since we only hit this path in the
756 // first call to LogAndApply (when opening the database).
757 assert(descriptor_file_ == NULL);
758 new_manifest_file = DescriptorFileName(dbname_, manifest_file_number_);
759 edit->SetNextFile(next_file_number_);
760 s = env_->NewWritableFile(new_manifest_file, &descriptor_file_);
762 descriptor_log_ = new log::Writer(descriptor_file_);
763 s = WriteSnapshot(descriptor_log_);
767 // Unlock during expensive MANIFEST log write
771 // Write new record to MANIFEST log
774 edit->EncodeTo(&record);
775 s = descriptor_log_->AddRecord(record);
777 s = descriptor_file_->Sync();
781 // If we just created a new descriptor file, install it by writing a
782 // new CURRENT file that points to it.
783 if (s.ok() && !new_manifest_file.empty()) {
784 s = SetCurrentFile(env_, dbname_, manifest_file_number_);
790 // Install the new version
793 log_number_ = edit->log_number_;
794 prev_log_number_ = edit->prev_log_number_;
797 if (!new_manifest_file.empty()) {
798 delete descriptor_log_;
799 delete descriptor_file_;
800 descriptor_log_ = NULL;
801 descriptor_file_ = NULL;
802 env_->DeleteFile(new_manifest_file);
809 Status VersionSet::Recover() {
810 struct LogReporter : public log::Reader::Reporter {
812 virtual void Corruption(size_t bytes, const Status& s) {
813 if (this->status->ok()) *this->status = s;
817 // Read "CURRENT" file, which contains a pointer to the current manifest file
819 Status s = ReadFileToString(env_, CurrentFileName(dbname_), ¤t);
823 if (current.empty() || current[current.size()-1] != '\n') {
824 return Status::Corruption("CURRENT file does not end with newline");
826 current.resize(current.size() - 1);
828 std::string dscname = dbname_ + "/" + current;
829 SequentialFile* file;
830 s = env_->NewSequentialFile(dscname, &file);
835 bool have_log_number = false;
836 bool have_prev_log_number = false;
837 bool have_next_file = false;
838 bool have_last_sequence = false;
839 uint64_t next_file = 0;
840 uint64_t last_sequence = 0;
841 uint64_t log_number = 0;
842 uint64_t prev_log_number = 0;
843 Builder builder(this, current_);
846 LogReporter reporter;
847 reporter.status = &s;
848 log::Reader reader(file, &reporter, true/*checksum*/, 0/*initial_offset*/);
851 while (reader.ReadRecord(&record, &scratch) && s.ok()) {
853 s = edit.DecodeFrom(record);
855 if (edit.has_comparator_ &&
856 edit.comparator_ != icmp_.user_comparator()->Name()) {
857 s = Status::InvalidArgument(
858 edit.comparator_ + "does not match existing comparator ",
859 icmp_.user_comparator()->Name());
864 builder.Apply(&edit);
867 if (edit.has_log_number_) {
868 log_number = edit.log_number_;
869 have_log_number = true;
872 if (edit.has_prev_log_number_) {
873 prev_log_number = edit.prev_log_number_;
874 have_prev_log_number = true;
877 if (edit.has_next_file_number_) {
878 next_file = edit.next_file_number_;
879 have_next_file = true;
882 if (edit.has_last_sequence_) {
883 last_sequence = edit.last_sequence_;
884 have_last_sequence = true;
892 if (!have_next_file) {
893 s = Status::Corruption("no meta-nextfile entry in descriptor");
894 } else if (!have_log_number) {
895 s = Status::Corruption("no meta-lognumber entry in descriptor");
896 } else if (!have_last_sequence) {
897 s = Status::Corruption("no last-sequence-number entry in descriptor");
900 if (!have_prev_log_number) {
904 MarkFileNumberUsed(prev_log_number);
905 MarkFileNumberUsed(log_number);
909 Version* v = new Version(this);
911 // Install recovered version
914 manifest_file_number_ = next_file;
915 next_file_number_ = next_file + 1;
916 last_sequence_ = last_sequence;
917 log_number_ = log_number;
918 prev_log_number_ = prev_log_number;
924 void VersionSet::MarkFileNumberUsed(uint64_t number) {
925 if (next_file_number_ <= number) {
926 next_file_number_ = number + 1;
930 void VersionSet::Finalize(Version* v) {
931 // Precomputed best level for next compaction
933 double best_score = -1;
935 for (int level = 0; level < config::kNumLevels-1; level++) {
938 // We treat level-0 specially by bounding the number of files
939 // instead of number of bytes for two reasons:
941 // (1) With larger write-buffer sizes, it is nice not to do too
942 // many level-0 compactions.
944 // (2) The files in level-0 are merged on every read and
945 // therefore we wish to avoid too many files when the individual
946 // file size is small (perhaps because of a small write-buffer
947 // setting, or very high compression ratios, or lots of
948 // overwrites/deletions).
949 score = v->files_[level].size() /
950 static_cast<double>(config::kL0_CompactionTrigger);
952 // Compute the ratio of current size to size limit.
953 const uint64_t level_bytes = TotalFileSize(v->files_[level]);
954 score = static_cast<double>(level_bytes) / MaxBytesForLevel(level);
957 if (score > best_score) {
963 v->compaction_level_ = best_level;
964 v->compaction_score_ = best_score;
967 Status VersionSet::WriteSnapshot(log::Writer* log) {
968 // TODO: Break up into multiple records to reduce memory usage on recovery?
972 edit.SetComparatorName(icmp_.user_comparator()->Name());
974 // Save compaction pointers
975 for (int level = 0; level < config::kNumLevels; level++) {
976 if (!compact_pointer_[level].empty()) {
978 key.DecodeFrom(compact_pointer_[level]);
979 edit.SetCompactPointer(level, key);
984 for (int level = 0; level < config::kNumLevels; level++) {
985 const std::vector<FileMetaData*>& files = current_->files_[level];
986 for (size_t i = 0; i < files.size(); i++) {
987 const FileMetaData* f = files[i];
988 edit.AddFile(level, f->number, f->file_size, f->smallest, f->largest);
993 edit.EncodeTo(&record);
994 return log->AddRecord(record);
997 int VersionSet::NumLevelFiles(int level) const {
999 assert(level < config::kNumLevels);
1000 return current_->files_[level].size();
1003 const char* VersionSet::LevelSummary(LevelSummaryStorage* scratch) const {
1004 // Update code if kNumLevels changes
1005 assert(config::kNumLevels == 7);
1006 snprintf(scratch->buffer, sizeof(scratch->buffer),
1007 "files[ %d %d %d %d %d %d %d ]",
1008 int(current_->files_[0].size()),
1009 int(current_->files_[1].size()),
1010 int(current_->files_[2].size()),
1011 int(current_->files_[3].size()),
1012 int(current_->files_[4].size()),
1013 int(current_->files_[5].size()),
1014 int(current_->files_[6].size()));
1015 return scratch->buffer;
1018 uint64_t VersionSet::ApproximateOffsetOf(Version* v, const InternalKey& ikey) {
1019 uint64_t result = 0;
1020 for (int level = 0; level < config::kNumLevels; level++) {
1021 const std::vector<FileMetaData*>& files = v->files_[level];
1022 for (size_t i = 0; i < files.size(); i++) {
1023 if (icmp_.Compare(files[i]->largest, ikey) <= 0) {
1024 // Entire file is before "ikey", so just add the file size
1025 result += files[i]->file_size;
1026 } else if (icmp_.Compare(files[i]->smallest, ikey) > 0) {
1027 // Entire file is after "ikey", so ignore
1029 // Files other than level 0 are sorted by meta->smallest, so
1030 // no further files in this level will contain data for
1035 // "ikey" falls in the range for this table. Add the
1036 // approximate offset of "ikey" within the table.
1038 Iterator* iter = table_cache_->NewIterator(
1039 ReadOptions(), files[i]->number, files[i]->file_size, &tableptr);
1040 if (tableptr != NULL) {
1041 result += tableptr->ApproximateOffsetOf(ikey.Encode());
1050 void VersionSet::AddLiveFiles(std::set<uint64_t>* live) {
1051 for (Version* v = dummy_versions_.next_;
1052 v != &dummy_versions_;
1054 for (int level = 0; level < config::kNumLevels; level++) {
1055 const std::vector<FileMetaData*>& files = v->files_[level];
1056 for (size_t i = 0; i < files.size(); i++) {
1057 live->insert(files[i]->number);
1063 int64_t VersionSet::NumLevelBytes(int level) const {
1065 assert(level < config::kNumLevels);
1066 return TotalFileSize(current_->files_[level]);
1069 int64_t VersionSet::MaxNextLevelOverlappingBytes() {
1071 std::vector<FileMetaData*> overlaps;
1072 for (int level = 1; level < config::kNumLevels - 1; level++) {
1073 for (size_t i = 0; i < current_->files_[level].size(); i++) {
1074 const FileMetaData* f = current_->files_[level][i];
1075 current_->GetOverlappingInputs(level+1, &f->smallest, &f->largest,
1077 const int64_t sum = TotalFileSize(overlaps);
1086 // Stores the minimal range that covers all entries in inputs in
1087 // *smallest, *largest.
1088 // REQUIRES: inputs is not empty
1089 void VersionSet::GetRange(const std::vector<FileMetaData*>& inputs,
1090 InternalKey* smallest,
1091 InternalKey* largest) {
1092 assert(!inputs.empty());
1095 for (size_t i = 0; i < inputs.size(); i++) {
1096 FileMetaData* f = inputs[i];
1098 *smallest = f->smallest;
1099 *largest = f->largest;
1101 if (icmp_.Compare(f->smallest, *smallest) < 0) {
1102 *smallest = f->smallest;
1104 if (icmp_.Compare(f->largest, *largest) > 0) {
1105 *largest = f->largest;
1111 // Stores the minimal range that covers all entries in inputs1 and inputs2
1112 // in *smallest, *largest.
1113 // REQUIRES: inputs is not empty
1114 void VersionSet::GetRange2(const std::vector<FileMetaData*>& inputs1,
1115 const std::vector<FileMetaData*>& inputs2,
1116 InternalKey* smallest,
1117 InternalKey* largest) {
1118 std::vector<FileMetaData*> all = inputs1;
1119 all.insert(all.end(), inputs2.begin(), inputs2.end());
1120 GetRange(all, smallest, largest);
1123 Iterator* VersionSet::MakeInputIterator(Compaction* c) {
1124 ReadOptions options;
1125 options.verify_checksums = options_->paranoid_checks;
1126 options.fill_cache = false;
1128 // Level-0 files have to be merged together. For other levels,
1129 // we will make a concatenating iterator per level.
1130 // TODO(opt): use concatenating iterator for level-0 if there is no overlap
1131 const int space = (c->level() == 0 ? c->inputs_[0].size() + 1 : 2);
1132 Iterator** list = new Iterator*[space];
1134 for (int which = 0; which < 2; which++) {
1135 if (!c->inputs_[which].empty()) {
1136 if (c->level() + which == 0) {
1137 const std::vector<FileMetaData*>& files = c->inputs_[which];
1138 for (size_t i = 0; i < files.size(); i++) {
1139 list[num++] = table_cache_->NewIterator(
1140 options, files[i]->number, files[i]->file_size);
1143 // Create concatenating iterator for the files from this level
1144 list[num++] = NewTwoLevelIterator(
1145 new Version::LevelFileNumIterator(icmp_, &c->inputs_[which]),
1146 &GetFileIterator, table_cache_, options);
1150 assert(num <= space);
1151 Iterator* result = NewMergingIterator(&icmp_, list, num);
1156 Compaction* VersionSet::PickCompaction() {
1160 // We prefer compactions triggered by too much data in a level over
1161 // the compactions triggered by seeks.
1162 const bool size_compaction = (current_->compaction_score_ >= 1);
1163 const bool seek_compaction = (current_->file_to_compact_ != NULL);
1164 if (size_compaction) {
1165 level = current_->compaction_level_;
1167 assert(level+1 < config::kNumLevels);
1168 c = new Compaction(level);
1170 // Pick the first file that comes after compact_pointer_[level]
1171 for (size_t i = 0; i < current_->files_[level].size(); i++) {
1172 FileMetaData* f = current_->files_[level][i];
1173 if (compact_pointer_[level].empty() ||
1174 icmp_.Compare(f->largest.Encode(), compact_pointer_[level]) > 0) {
1175 c->inputs_[0].push_back(f);
1179 if (c->inputs_[0].empty()) {
1180 // Wrap-around to the beginning of the key space
1181 c->inputs_[0].push_back(current_->files_[level][0]);
1183 } else if (seek_compaction) {
1184 level = current_->file_to_compact_level_;
1185 c = new Compaction(level);
1186 c->inputs_[0].push_back(current_->file_to_compact_);
1191 c->input_version_ = current_;
1192 c->input_version_->Ref();
1194 // Files in level 0 may overlap each other, so pick up all overlapping ones
1196 InternalKey smallest, largest;
1197 GetRange(c->inputs_[0], &smallest, &largest);
1198 // Note that the next call will discard the file we placed in
1199 // c->inputs_[0] earlier and replace it with an overlapping set
1200 // which will include the picked file.
1201 current_->GetOverlappingInputs(0, &smallest, &largest, &c->inputs_[0]);
1202 assert(!c->inputs_[0].empty());
1205 SetupOtherInputs(c);
1210 void VersionSet::SetupOtherInputs(Compaction* c) {
1211 const int level = c->level();
1212 InternalKey smallest, largest;
1213 GetRange(c->inputs_[0], &smallest, &largest);
1215 current_->GetOverlappingInputs(level+1, &smallest, &largest, &c->inputs_[1]);
1217 // Get entire range covered by compaction
1218 InternalKey all_start, all_limit;
1219 GetRange2(c->inputs_[0], c->inputs_[1], &all_start, &all_limit);
1221 // See if we can grow the number of inputs in "level" without
1222 // changing the number of "level+1" files we pick up.
1223 if (!c->inputs_[1].empty()) {
1224 std::vector<FileMetaData*> expanded0;
1225 current_->GetOverlappingInputs(level, &all_start, &all_limit, &expanded0);
1226 if (expanded0.size() > c->inputs_[0].size()) {
1227 InternalKey new_start, new_limit;
1228 GetRange(expanded0, &new_start, &new_limit);
1229 std::vector<FileMetaData*> expanded1;
1230 current_->GetOverlappingInputs(level+1, &new_start, &new_limit,
1232 if (expanded1.size() == c->inputs_[1].size()) {
1233 Log(options_->info_log,
1234 "Expanding@%d %d+%d to %d+%d\n",
1236 int(c->inputs_[0].size()),
1237 int(c->inputs_[1].size()),
1238 int(expanded0.size()),
1239 int(expanded1.size()));
1240 smallest = new_start;
1241 largest = new_limit;
1242 c->inputs_[0] = expanded0;
1243 c->inputs_[1] = expanded1;
1244 GetRange2(c->inputs_[0], c->inputs_[1], &all_start, &all_limit);
1249 // Compute the set of grandparent files that overlap this compaction
1250 // (parent == level+1; grandparent == level+2)
1251 if (level + 2 < config::kNumLevels) {
1252 current_->GetOverlappingInputs(level + 2, &all_start, &all_limit,
1257 Log(options_->info_log, "Compacting %d '%s' .. '%s'",
1259 smallest.DebugString().c_str(),
1260 largest.DebugString().c_str());
1263 // Update the place where we will do the next compaction for this level.
1264 // We update this immediately instead of waiting for the VersionEdit
1265 // to be applied so that if the compaction fails, we will try a different
1266 // key range next time.
1267 compact_pointer_[level] = largest.Encode().ToString();
1268 c->edit_.SetCompactPointer(level, largest);
1271 Compaction* VersionSet::CompactRange(
1273 const InternalKey* begin,
1274 const InternalKey* end) {
1275 std::vector<FileMetaData*> inputs;
1276 current_->GetOverlappingInputs(level, begin, end, &inputs);
1277 if (inputs.empty()) {
1281 // Avoid compacting too much in one shot in case the range is large.
1282 const uint64_t limit = MaxFileSizeForLevel(level);
1284 for (int i = 0; i < inputs.size(); i++) {
1285 uint64_t s = inputs[i]->file_size;
1287 if (total >= limit) {
1288 inputs.resize(i + 1);
1293 Compaction* c = new Compaction(level);
1294 c->input_version_ = current_;
1295 c->input_version_->Ref();
1296 c->inputs_[0] = inputs;
1297 SetupOtherInputs(c);
1301 Compaction::Compaction(int level)
1303 max_output_file_size_(MaxFileSizeForLevel(level)),
1304 input_version_(NULL),
1305 grandparent_index_(0),
1307 overlapped_bytes_(0) {
1308 for (int i = 0; i < config::kNumLevels; i++) {
1313 Compaction::~Compaction() {
1314 if (input_version_ != NULL) {
1315 input_version_->Unref();
1319 bool Compaction::IsTrivialMove() const {
1320 // Avoid a move if there is lots of overlapping grandparent data.
1321 // Otherwise, the move could create a parent file that will require
1322 // a very expensive merge later on.
1323 return (num_input_files(0) == 1 &&
1324 num_input_files(1) == 0 &&
1325 TotalFileSize(grandparents_) <= kMaxGrandParentOverlapBytes);
1328 void Compaction::AddInputDeletions(VersionEdit* edit) {
1329 for (int which = 0; which < 2; which++) {
1330 for (size_t i = 0; i < inputs_[which].size(); i++) {
1331 edit->DeleteFile(level_ + which, inputs_[which][i]->number);
1336 bool Compaction::IsBaseLevelForKey(const Slice& user_key) {
1337 // Maybe use binary search to find right entry instead of linear search?
1338 const Comparator* user_cmp = input_version_->vset_->icmp_.user_comparator();
1339 for (int lvl = level_ + 2; lvl < config::kNumLevels; lvl++) {
1340 const std::vector<FileMetaData*>& files = input_version_->files_[lvl];
1341 for (; level_ptrs_[lvl] < files.size(); ) {
1342 FileMetaData* f = files[level_ptrs_[lvl]];
1343 if (user_cmp->Compare(user_key, f->largest.user_key()) <= 0) {
1344 // We've advanced far enough
1345 if (user_cmp->Compare(user_key, f->smallest.user_key()) >= 0) {
1346 // Key falls in this file's range, so definitely not base level
1357 bool Compaction::ShouldStopBefore(const Slice& internal_key) {
1358 // Scan to find earliest grandparent file that contains key.
1359 const InternalKeyComparator* icmp = &input_version_->vset_->icmp_;
1360 while (grandparent_index_ < grandparents_.size() &&
1361 icmp->Compare(internal_key,
1362 grandparents_[grandparent_index_]->largest.Encode()) > 0) {
1364 overlapped_bytes_ += grandparents_[grandparent_index_]->file_size;
1366 grandparent_index_++;
1370 if (overlapped_bytes_ > kMaxGrandParentOverlapBytes) {
1371 // Too much overlap for current output; start new output
1372 overlapped_bytes_ = 0;
1379 void Compaction::ReleaseInputs() {
1380 if (input_version_ != NULL) {
1381 input_version_->Unref();
1382 input_version_ = NULL;
1386 } // namespace leveldb