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.
4//
5// The representation of a DBImpl consists of a set of Versions.  The
6// newest version is called "current".  Older versions may be kept
7// around to provide a consistent view to live iterators.
8//
9// Each Version keeps track of a set of Table files per level.  The
10// entire set of versions is maintained in a VersionSet.
11//
12// Version,VersionSet are thread-compatible, but require external
13// synchronization on all accesses.
14
15#ifndef STORAGE_LEVELDB_DB_VERSION_SET_H_
16#define STORAGE_LEVELDB_DB_VERSION_SET_H_
17
18#include <map>
19#include <set>
20#include <vector>
21#include "db/dbformat.h"
22#include "db/version_edit.h"
23#include "port/port.h"
24#include "port/thread_annotations.h"
25
26namespace leveldb {
27
28namespace log { class Writer; }
29
30class Compaction;
31class Iterator;
32class MemTable;
33class TableBuilder;
34class TableCache;
35class Version;
36class VersionSet;
37class WritableFile;
38
39// Return the smallest index i such that files[i]->largest >= key.
40// Return files.size() if there is no such file.
41// REQUIRES: "files" contains a sorted list of non-overlapping files.
42extern int FindFile(const InternalKeyComparator& icmp,
43                    const std::vector<FileMetaData*>& files,
44                    const Slice& key);
45
46// Returns true iff some file in "files" overlaps the user key range
47// [*smallest,*largest].
48// smallest==NULL represents a key smaller than all keys in the DB.
49// largest==NULL represents a key largest than all keys in the DB.
50// REQUIRES: If disjoint_sorted_files, files[] contains disjoint ranges
51//           in sorted order.
52extern bool SomeFileOverlapsRange(
53    const InternalKeyComparator& icmp,
54    bool disjoint_sorted_files,
55    const std::vector<FileMetaData*>& files,
56    const Slice* smallest_user_key,
57    const Slice* largest_user_key);
58
59class Version {
60 public:
61  // Append to *iters a sequence of iterators that will
62  // yield the contents of this Version when merged together.
63  // REQUIRES: This version has been saved (see VersionSet::SaveTo)
64  void AddIterators(const ReadOptions&, std::vector<Iterator*>* iters);
65
66  // Lookup the value for key.  If found, store it in *val and
67  // return OK.  Else return a non-OK status.  Fills *stats.
68  // REQUIRES: lock is not held
69  struct GetStats {
70    FileMetaData* seek_file;
71    int seek_file_level;
72  };
73  Status Get(const ReadOptions&, const LookupKey& key, std::string* val,
74             GetStats* stats);
75
76  // Adds "stats" into the current state.  Returns true if a new
77  // compaction may need to be triggered, false otherwise.
78  // REQUIRES: lock is held
79  bool UpdateStats(const GetStats& stats);
80
81  // Record a sample of bytes read at the specified internal key.
82  // Samples are taken approximately once every config::kReadBytesPeriod
83  // bytes.  Returns true if a new compaction may need to be triggered.
84  // REQUIRES: lock is held
85  bool RecordReadSample(Slice key);
86
87  // Reference count management (so Versions do not disappear out from
88  // under live iterators)
89  void Ref();
90  void Unref();
91
92  void GetOverlappingInputs(
93      int level,
94      const InternalKey* begin,         // NULL means before all keys
95      const InternalKey* end,           // NULL means after all keys
96      std::vector<FileMetaData*>* inputs);
97
98  // Returns true iff some file in the specified level overlaps
99  // some part of [*smallest_user_key,*largest_user_key].
100  // smallest_user_key==NULL represents a key smaller than all keys in the DB.
101  // largest_user_key==NULL represents a key largest than all keys in the DB.
102  bool OverlapInLevel(int level,
103                      const Slice* smallest_user_key,
104                      const Slice* largest_user_key);
105
106  // Return the level at which we should place a new memtable compaction
107  // result that covers the range [smallest_user_key,largest_user_key].
108  int PickLevelForMemTableOutput(const Slice& smallest_user_key,
109                                 const Slice& largest_user_key);
110
111  int NumFiles(int level) const { return files_[level].size(); }
112
113  // Return a human readable string that describes this version's contents.
114  std::string DebugString() const;
115
116 private:
117  friend class Compaction;
118  friend class VersionSet;
119
120  class LevelFileNumIterator;
121  Iterator* NewConcatenatingIterator(const ReadOptions&, int level) const;
122
123  // Call func(arg, level, f) for every file that overlaps user_key in
124  // order from newest to oldest.  If an invocation of func returns
125  // false, makes no more calls.
126  //
127  // REQUIRES: user portion of internal_key == user_key.
128  void ForEachOverlapping(Slice user_key, Slice internal_key,
129                          void* arg,
130                          bool (*func)(void*, int, FileMetaData*));
131
132  VersionSet* vset_;            // VersionSet to which this Version belongs
133  Version* next_;               // Next version in linked list
134  Version* prev_;               // Previous version in linked list
135  int refs_;                    // Number of live refs to this version
136
137  // List of files per level
138  std::vector<FileMetaData*> files_[config::kNumLevels];
139
140  // Next file to compact based on seek stats.
141  FileMetaData* file_to_compact_;
142  int file_to_compact_level_;
143
144  // Level that should be compacted next and its compaction score.
145  // Score < 1 means compaction is not strictly needed.  These fields
146  // are initialized by Finalize().
147  double compaction_score_;
148  int compaction_level_;
149
150  explicit Version(VersionSet* vset)
151      : vset_(vset), next_(this), prev_(this), refs_(0),
152        file_to_compact_(NULL),
153        file_to_compact_level_(-1),
154        compaction_score_(-1),
155        compaction_level_(-1) {
156  }
157
158  ~Version();
159
160  // No copying allowed
161  Version(const Version&);
162  void operator=(const Version&);
163};
164
165class VersionSet {
166 public:
167  VersionSet(const std::string& dbname,
168             const Options* options,
169             TableCache* table_cache,
170             const InternalKeyComparator*);
171  ~VersionSet();
172
173  // Apply *edit to the current version to form a new descriptor that
174  // is both saved to persistent state and installed as the new
175  // current version.  Will release *mu while actually writing to the file.
176  // REQUIRES: *mu is held on entry.
177  // REQUIRES: no other thread concurrently calls LogAndApply()
178  Status LogAndApply(VersionEdit* edit, port::Mutex* mu)
179      EXCLUSIVE_LOCKS_REQUIRED(mu);
180
181  // Recover the last saved descriptor from persistent storage.
182  Status Recover();
183
184  // Return the current version.
185  Version* current() const { return current_; }
186
187  // Return the current manifest file number
188  uint64_t ManifestFileNumber() const { return manifest_file_number_; }
189
190  // Allocate and return a new file number
191  uint64_t NewFileNumber() { return next_file_number_++; }
192
193  // Arrange to reuse "file_number" unless a newer file number has
194  // already been allocated.
195  // REQUIRES: "file_number" was returned by a call to NewFileNumber().
196  void ReuseFileNumber(uint64_t file_number) {
197    if (next_file_number_ == file_number + 1) {
198      next_file_number_ = file_number;
199    }
200  }
201
202  // Return the number of Table files at the specified level.
203  int NumLevelFiles(int level) const;
204
205  // Return the combined file size of all files at the specified level.
206  int64_t NumLevelBytes(int level) const;
207
208  // Return the last sequence number.
209  uint64_t LastSequence() const { return last_sequence_; }
210
211  // Set the last sequence number to s.
212  void SetLastSequence(uint64_t s) {
213    assert(s >= last_sequence_);
214    last_sequence_ = s;
215  }
216
217  // Mark the specified file number as used.
218  void MarkFileNumberUsed(uint64_t number);
219
220  // Return the current log file number.
221  uint64_t LogNumber() const { return log_number_; }
222
223  // Return the log file number for the log file that is currently
224  // being compacted, or zero if there is no such log file.
225  uint64_t PrevLogNumber() const { return prev_log_number_; }
226
227  // Pick level and inputs for a new compaction.
228  // Returns NULL if there is no compaction to be done.
229  // Otherwise returns a pointer to a heap-allocated object that
230  // describes the compaction.  Caller should delete the result.
231  Compaction* PickCompaction();
232
233  // Return a compaction object for compacting the range [begin,end] in
234  // the specified level.  Returns NULL if there is nothing in that
235  // level that overlaps the specified range.  Caller should delete
236  // the result.
237  Compaction* CompactRange(
238      int level,
239      const InternalKey* begin,
240      const InternalKey* end);
241
242  // Return the maximum overlapping data (in bytes) at next level for any
243  // file at a level >= 1.
244  int64_t MaxNextLevelOverlappingBytes();
245
246  // Create an iterator that reads over the compaction inputs for "*c".
247  // The caller should delete the iterator when no longer needed.
248  Iterator* MakeInputIterator(Compaction* c);
249
250  // Returns true iff some level needs a compaction.
251  bool NeedsCompaction() const {
252    Version* v = current_;
253    return (v->compaction_score_ >= 1) || (v->file_to_compact_ != NULL);
254  }
255
256  // Add all files listed in any live version to *live.
257  // May also mutate some internal state.
258  void AddLiveFiles(std::set<uint64_t>* live);
259
260  // Return the approximate offset in the database of the data for
261  // "key" as of version "v".
262  uint64_t ApproximateOffsetOf(Version* v, const InternalKey& key);
263
264  // Return a human-readable short (single-line) summary of the number
265  // of files per level.  Uses *scratch as backing store.
266  struct LevelSummaryStorage {
267    char buffer[100];
268  };
269  const char* LevelSummary(LevelSummaryStorage* scratch) const;
270
271 private:
272  class Builder;
273
274  friend class Compaction;
275  friend class Version;
276
277  void Finalize(Version* v);
278
279  void GetRange(const std::vector<FileMetaData*>& inputs,
280                InternalKey* smallest,
281                InternalKey* largest);
282
283  void GetRange2(const std::vector<FileMetaData*>& inputs1,
284                 const std::vector<FileMetaData*>& inputs2,
285                 InternalKey* smallest,
286                 InternalKey* largest);
287
288  void SetupOtherInputs(Compaction* c);
289
290  // Save current contents to *log
291  Status WriteSnapshot(log::Writer* log);
292
293  void AppendVersion(Version* v);
294
295  Env* const env_;
296  const std::string dbname_;
297  const Options* const options_;
298  TableCache* const table_cache_;
299  const InternalKeyComparator icmp_;
300  uint64_t next_file_number_;
301  uint64_t manifest_file_number_;
302  uint64_t last_sequence_;
303  uint64_t log_number_;
304  uint64_t prev_log_number_;  // 0 or backing store for memtable being compacted
305
306  // Opened lazily
307  WritableFile* descriptor_file_;
308  log::Writer* descriptor_log_;
309  Version dummy_versions_;  // Head of circular doubly-linked list of versions.
310  Version* current_;        // == dummy_versions_.prev_
311
312  // Per-level key at which the next compaction at that level should start.
313  // Either an empty string, or a valid InternalKey.
314  std::string compact_pointer_[config::kNumLevels];
315
316  // No copying allowed
317  VersionSet(const VersionSet&);
318  void operator=(const VersionSet&);
319};
320
321// A Compaction encapsulates information about a compaction.
322class Compaction {
323 public:
324  ~Compaction();
325
326  // Return the level that is being compacted.  Inputs from "level"
327  // and "level+1" will be merged to produce a set of "level+1" files.
328  int level() const { return level_; }
329
330  // Return the object that holds the edits to the descriptor done
331  // by this compaction.
332  VersionEdit* edit() { return &edit_; }
333
334  // "which" must be either 0 or 1
335  int num_input_files(int which) const { return inputs_[which].size(); }
336
337  // Return the ith input file at "level()+which" ("which" must be 0 or 1).
338  FileMetaData* input(int which, int i) const { return inputs_[which][i]; }
339
340  // Maximum size of files to build during this compaction.
341  uint64_t MaxOutputFileSize() const { return max_output_file_size_; }
342
343  // Is this a trivial compaction that can be implemented by just
344  // moving a single input file to the next level (no merging or splitting)
345  bool IsTrivialMove() const;
346
347  // Add all inputs to this compaction as delete operations to *edit.
348  void AddInputDeletions(VersionEdit* edit);
349
350  // Returns true if the information we have available guarantees that
351  // the compaction is producing data in "level+1" for which no data exists
352  // in levels greater than "level+1".
353  bool IsBaseLevelForKey(const Slice& user_key);
354
355  // Returns true iff we should stop building the current output
356  // before processing "internal_key".
357  bool ShouldStopBefore(const Slice& internal_key);
358
359  // Release the input version for the compaction, once the compaction
360  // is successful.
361  void ReleaseInputs();
362
363 private:
364  friend class Version;
365  friend class VersionSet;
366
367  explicit Compaction(int level);
368
369  int level_;
370  uint64_t max_output_file_size_;
371  Version* input_version_;
372  VersionEdit edit_;
373
374  // Each compaction reads inputs from "level_" and "level_+1"
375  std::vector<FileMetaData*> inputs_[2];      // The two sets of inputs
376
377  // State used to check for number of of overlapping grandparent files
378  // (parent == level_ + 1, grandparent == level_ + 2)
379  std::vector<FileMetaData*> grandparents_;
380  size_t grandparent_index_;  // Index in grandparent_starts_
381  bool seen_key_;             // Some output key has been seen
382  int64_t overlapped_bytes_;  // Bytes of overlap between current output
383                              // and grandparent files
384
385  // State for implementing IsBaseLevelForKey
386
387  // level_ptrs_ holds indices into input_version_->levels_: our state
388  // is that we are positioned at one of the file ranges for each
389  // higher level than the ones involved in this compaction (i.e. for
390  // all L >= level_ + 2).
391  size_t level_ptrs_[config::kNumLevels];
392};
393
394}  // namespace leveldb
395
396#endif  // STORAGE_LEVELDB_DB_VERSION_SET_H_
397