leveldb-ruby 0.1
Sign up to get free protection for your applications and to get access to all the features.
- data/README +17 -0
- data/ext/leveldb/extconf.rb +10 -0
- data/ext/leveldb/leveldb.cc +181 -0
- data/leveldb/Makefile +172 -0
- data/leveldb/db/builder.cc +90 -0
- data/leveldb/db/builder.h +36 -0
- data/leveldb/db/corruption_test.cc +354 -0
- data/leveldb/db/db_bench.cc +677 -0
- data/leveldb/db/db_impl.cc +1236 -0
- data/leveldb/db/db_impl.h +180 -0
- data/leveldb/db/db_iter.cc +298 -0
- data/leveldb/db/db_iter.h +26 -0
- data/leveldb/db/db_test.cc +1192 -0
- data/leveldb/db/dbformat.cc +87 -0
- data/leveldb/db/dbformat.h +165 -0
- data/leveldb/db/dbformat_test.cc +112 -0
- data/leveldb/db/filename.cc +135 -0
- data/leveldb/db/filename.h +80 -0
- data/leveldb/db/filename_test.cc +122 -0
- data/leveldb/db/log_format.h +35 -0
- data/leveldb/db/log_reader.cc +254 -0
- data/leveldb/db/log_reader.h +108 -0
- data/leveldb/db/log_test.cc +500 -0
- data/leveldb/db/log_writer.cc +103 -0
- data/leveldb/db/log_writer.h +48 -0
- data/leveldb/db/memtable.cc +108 -0
- data/leveldb/db/memtable.h +85 -0
- data/leveldb/db/repair.cc +384 -0
- data/leveldb/db/skiplist.h +378 -0
- data/leveldb/db/skiplist_test.cc +378 -0
- data/leveldb/db/snapshot.h +66 -0
- data/leveldb/db/table_cache.cc +95 -0
- data/leveldb/db/table_cache.h +50 -0
- data/leveldb/db/version_edit.cc +268 -0
- data/leveldb/db/version_edit.h +106 -0
- data/leveldb/db/version_edit_test.cc +46 -0
- data/leveldb/db/version_set.cc +1060 -0
- data/leveldb/db/version_set.h +306 -0
- data/leveldb/db/write_batch.cc +138 -0
- data/leveldb/db/write_batch_internal.h +45 -0
- data/leveldb/db/write_batch_test.cc +89 -0
- data/leveldb/include/leveldb/cache.h +99 -0
- data/leveldb/include/leveldb/comparator.h +63 -0
- data/leveldb/include/leveldb/db.h +148 -0
- data/leveldb/include/leveldb/env.h +302 -0
- data/leveldb/include/leveldb/iterator.h +100 -0
- data/leveldb/include/leveldb/options.h +198 -0
- data/leveldb/include/leveldb/slice.h +109 -0
- data/leveldb/include/leveldb/status.h +100 -0
- data/leveldb/include/leveldb/table.h +70 -0
- data/leveldb/include/leveldb/table_builder.h +91 -0
- data/leveldb/include/leveldb/write_batch.h +64 -0
- data/leveldb/port/port.h +23 -0
- data/leveldb/port/port_android.cc +64 -0
- data/leveldb/port/port_android.h +150 -0
- data/leveldb/port/port_chromium.cc +80 -0
- data/leveldb/port/port_chromium.h +97 -0
- data/leveldb/port/port_example.h +115 -0
- data/leveldb/port/port_osx.cc +50 -0
- data/leveldb/port/port_osx.h +125 -0
- data/leveldb/port/port_posix.cc +50 -0
- data/leveldb/port/port_posix.h +94 -0
- data/leveldb/port/sha1_portable.cc +298 -0
- data/leveldb/port/sha1_portable.h +25 -0
- data/leveldb/port/sha1_test.cc +39 -0
- data/leveldb/port/win/stdint.h +24 -0
- data/leveldb/table/block.cc +263 -0
- data/leveldb/table/block.h +43 -0
- data/leveldb/table/block_builder.cc +109 -0
- data/leveldb/table/block_builder.h +57 -0
- data/leveldb/table/format.cc +131 -0
- data/leveldb/table/format.h +103 -0
- data/leveldb/table/iterator.cc +67 -0
- data/leveldb/table/iterator_wrapper.h +63 -0
- data/leveldb/table/merger.cc +197 -0
- data/leveldb/table/merger.h +26 -0
- data/leveldb/table/table.cc +175 -0
- data/leveldb/table/table_builder.cc +227 -0
- data/leveldb/table/table_test.cc +845 -0
- data/leveldb/table/two_level_iterator.cc +182 -0
- data/leveldb/table/two_level_iterator.h +34 -0
- data/leveldb/util/arena.cc +68 -0
- data/leveldb/util/arena.h +68 -0
- data/leveldb/util/arena_test.cc +68 -0
- data/leveldb/util/cache.cc +255 -0
- data/leveldb/util/cache_test.cc +169 -0
- data/leveldb/util/coding.cc +194 -0
- data/leveldb/util/coding.h +104 -0
- data/leveldb/util/coding_test.cc +173 -0
- data/leveldb/util/comparator.cc +72 -0
- data/leveldb/util/crc32c.cc +332 -0
- data/leveldb/util/crc32c.h +45 -0
- data/leveldb/util/crc32c_test.cc +72 -0
- data/leveldb/util/env.cc +77 -0
- data/leveldb/util/env_chromium.cc +612 -0
- data/leveldb/util/env_posix.cc +606 -0
- data/leveldb/util/env_test.cc +102 -0
- data/leveldb/util/hash.cc +45 -0
- data/leveldb/util/hash.h +19 -0
- data/leveldb/util/histogram.cc +128 -0
- data/leveldb/util/histogram.h +41 -0
- data/leveldb/util/logging.cc +81 -0
- data/leveldb/util/logging.h +47 -0
- data/leveldb/util/mutexlock.h +39 -0
- data/leveldb/util/options.cc +28 -0
- data/leveldb/util/random.h +59 -0
- data/leveldb/util/status.cc +75 -0
- data/leveldb/util/testharness.cc +65 -0
- data/leveldb/util/testharness.h +129 -0
- data/leveldb/util/testutil.cc +51 -0
- data/leveldb/util/testutil.h +53 -0
- data/lib/leveldb.rb +36 -0
- metadata +183 -0
@@ -0,0 +1,1236 @@
|
|
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
|
+
#include "db/db_impl.h"
|
6
|
+
|
7
|
+
#include <algorithm>
|
8
|
+
#include <set>
|
9
|
+
#include <string>
|
10
|
+
#include <stdint.h>
|
11
|
+
#include <stdio.h>
|
12
|
+
#include <vector>
|
13
|
+
#include "db/builder.h"
|
14
|
+
#include "db/db_iter.h"
|
15
|
+
#include "db/dbformat.h"
|
16
|
+
#include "db/filename.h"
|
17
|
+
#include "db/log_reader.h"
|
18
|
+
#include "db/log_writer.h"
|
19
|
+
#include "db/memtable.h"
|
20
|
+
#include "db/table_cache.h"
|
21
|
+
#include "db/version_set.h"
|
22
|
+
#include "db/write_batch_internal.h"
|
23
|
+
#include "leveldb/db.h"
|
24
|
+
#include "leveldb/env.h"
|
25
|
+
#include "leveldb/status.h"
|
26
|
+
#include "leveldb/table.h"
|
27
|
+
#include "leveldb/table_builder.h"
|
28
|
+
#include "port/port.h"
|
29
|
+
#include "table/block.h"
|
30
|
+
#include "table/merger.h"
|
31
|
+
#include "table/two_level_iterator.h"
|
32
|
+
#include "util/coding.h"
|
33
|
+
#include "util/logging.h"
|
34
|
+
#include "util/mutexlock.h"
|
35
|
+
|
36
|
+
namespace leveldb {
|
37
|
+
|
38
|
+
struct DBImpl::CompactionState {
|
39
|
+
Compaction* const compaction;
|
40
|
+
|
41
|
+
// Sequence numbers < smallest_snapshot are not significant since we
|
42
|
+
// will never have to service a snapshot below smallest_snapshot.
|
43
|
+
// Therefore if we have seen a sequence number S <= smallest_snapshot,
|
44
|
+
// we can drop all entries for the same key with sequence numbers < S.
|
45
|
+
SequenceNumber smallest_snapshot;
|
46
|
+
|
47
|
+
// Files produced by compaction
|
48
|
+
struct Output {
|
49
|
+
uint64_t number;
|
50
|
+
uint64_t file_size;
|
51
|
+
InternalKey smallest, largest;
|
52
|
+
};
|
53
|
+
std::vector<Output> outputs;
|
54
|
+
|
55
|
+
// State kept for output being generated
|
56
|
+
WritableFile* outfile;
|
57
|
+
TableBuilder* builder;
|
58
|
+
|
59
|
+
uint64_t total_bytes;
|
60
|
+
|
61
|
+
Output* current_output() { return &outputs[outputs.size()-1]; }
|
62
|
+
|
63
|
+
explicit CompactionState(Compaction* c)
|
64
|
+
: compaction(c),
|
65
|
+
outfile(NULL),
|
66
|
+
builder(NULL),
|
67
|
+
total_bytes(0) {
|
68
|
+
}
|
69
|
+
};
|
70
|
+
|
71
|
+
namespace {
|
72
|
+
class NullWritableFile : public WritableFile {
|
73
|
+
public:
|
74
|
+
virtual Status Append(const Slice& data) { return Status::OK(); }
|
75
|
+
virtual Status Close() { return Status::OK(); }
|
76
|
+
virtual Status Flush() { return Status::OK(); }
|
77
|
+
virtual Status Sync() { return Status::OK(); }
|
78
|
+
};
|
79
|
+
}
|
80
|
+
|
81
|
+
// Fix user-supplied options to be reasonable
|
82
|
+
template <class T,class V>
|
83
|
+
static void ClipToRange(T* ptr, V minvalue, V maxvalue) {
|
84
|
+
if (static_cast<V>(*ptr) > maxvalue) *ptr = maxvalue;
|
85
|
+
if (static_cast<V>(*ptr) < minvalue) *ptr = minvalue;
|
86
|
+
}
|
87
|
+
Options SanitizeOptions(const std::string& dbname,
|
88
|
+
const InternalKeyComparator* icmp,
|
89
|
+
const Options& src) {
|
90
|
+
Options result = src;
|
91
|
+
result.comparator = icmp;
|
92
|
+
ClipToRange(&result.max_open_files, 20, 50000);
|
93
|
+
ClipToRange(&result.write_buffer_size, 64<<10, 1<<30);
|
94
|
+
ClipToRange(&result.block_size, 1<<10, 4<<20);
|
95
|
+
if (result.info_log == NULL) {
|
96
|
+
// Open a log file in the same directory as the db
|
97
|
+
src.env->CreateDir(dbname); // In case it does not exist
|
98
|
+
src.env->RenameFile(InfoLogFileName(dbname), OldInfoLogFileName(dbname));
|
99
|
+
Status s = src.env->NewWritableFile(InfoLogFileName(dbname),
|
100
|
+
&result.info_log);
|
101
|
+
if (!s.ok()) {
|
102
|
+
// No place suitable for logging
|
103
|
+
result.info_log = new NullWritableFile;
|
104
|
+
}
|
105
|
+
}
|
106
|
+
if (result.block_cache == NULL) {
|
107
|
+
result.block_cache = NewLRUCache(8 << 20);
|
108
|
+
}
|
109
|
+
return result;
|
110
|
+
}
|
111
|
+
|
112
|
+
DBImpl::DBImpl(const Options& options, const std::string& dbname)
|
113
|
+
: env_(options.env),
|
114
|
+
internal_comparator_(options.comparator),
|
115
|
+
options_(SanitizeOptions(dbname, &internal_comparator_, options)),
|
116
|
+
owns_info_log_(options_.info_log != options.info_log),
|
117
|
+
owns_cache_(options_.block_cache != options.block_cache),
|
118
|
+
dbname_(dbname),
|
119
|
+
db_lock_(NULL),
|
120
|
+
shutting_down_(NULL),
|
121
|
+
bg_cv_(&mutex_),
|
122
|
+
compacting_cv_(&mutex_),
|
123
|
+
mem_(new MemTable(internal_comparator_)),
|
124
|
+
imm_(NULL),
|
125
|
+
logfile_(NULL),
|
126
|
+
log_(NULL),
|
127
|
+
bg_compaction_scheduled_(false),
|
128
|
+
compacting_(false) {
|
129
|
+
mem_->Ref();
|
130
|
+
has_imm_.Release_Store(NULL);
|
131
|
+
|
132
|
+
// Reserve ten files or so for other uses and give the rest to TableCache.
|
133
|
+
const int table_cache_size = options.max_open_files - 10;
|
134
|
+
table_cache_ = new TableCache(dbname_, &options_, table_cache_size);
|
135
|
+
|
136
|
+
versions_ = new VersionSet(dbname_, &options_, table_cache_,
|
137
|
+
&internal_comparator_);
|
138
|
+
}
|
139
|
+
|
140
|
+
DBImpl::~DBImpl() {
|
141
|
+
// Wait for background work to finish
|
142
|
+
mutex_.Lock();
|
143
|
+
shutting_down_.Release_Store(this); // Any non-NULL value is ok
|
144
|
+
if (bg_compaction_scheduled_) {
|
145
|
+
while (bg_compaction_scheduled_) {
|
146
|
+
bg_cv_.Wait();
|
147
|
+
}
|
148
|
+
}
|
149
|
+
mutex_.Unlock();
|
150
|
+
|
151
|
+
if (db_lock_ != NULL) {
|
152
|
+
env_->UnlockFile(db_lock_);
|
153
|
+
}
|
154
|
+
|
155
|
+
delete versions_;
|
156
|
+
if (mem_ != NULL) mem_->Unref();
|
157
|
+
if (imm_ != NULL) imm_->Unref();
|
158
|
+
delete log_;
|
159
|
+
delete logfile_;
|
160
|
+
delete table_cache_;
|
161
|
+
|
162
|
+
if (owns_info_log_) {
|
163
|
+
delete options_.info_log;
|
164
|
+
}
|
165
|
+
if (owns_cache_) {
|
166
|
+
delete options_.block_cache;
|
167
|
+
}
|
168
|
+
}
|
169
|
+
|
170
|
+
Status DBImpl::NewDB() {
|
171
|
+
VersionEdit new_db;
|
172
|
+
new_db.SetComparatorName(user_comparator()->Name());
|
173
|
+
new_db.SetLogNumber(0);
|
174
|
+
new_db.SetNextFile(2);
|
175
|
+
new_db.SetLastSequence(0);
|
176
|
+
|
177
|
+
const std::string manifest = DescriptorFileName(dbname_, 1);
|
178
|
+
WritableFile* file;
|
179
|
+
Status s = env_->NewWritableFile(manifest, &file);
|
180
|
+
if (!s.ok()) {
|
181
|
+
return s;
|
182
|
+
}
|
183
|
+
{
|
184
|
+
log::Writer log(file);
|
185
|
+
std::string record;
|
186
|
+
new_db.EncodeTo(&record);
|
187
|
+
s = log.AddRecord(record);
|
188
|
+
if (s.ok()) {
|
189
|
+
s = file->Close();
|
190
|
+
}
|
191
|
+
}
|
192
|
+
delete file;
|
193
|
+
if (s.ok()) {
|
194
|
+
// Make "CURRENT" file that points to the new manifest file.
|
195
|
+
s = SetCurrentFile(env_, dbname_, 1);
|
196
|
+
} else {
|
197
|
+
env_->DeleteFile(manifest);
|
198
|
+
}
|
199
|
+
return s;
|
200
|
+
}
|
201
|
+
|
202
|
+
void DBImpl::MaybeIgnoreError(Status* s) const {
|
203
|
+
if (s->ok() || options_.paranoid_checks) {
|
204
|
+
// No change needed
|
205
|
+
} else {
|
206
|
+
Log(env_, options_.info_log, "Ignoring error %s", s->ToString().c_str());
|
207
|
+
*s = Status::OK();
|
208
|
+
}
|
209
|
+
}
|
210
|
+
|
211
|
+
void DBImpl::DeleteObsoleteFiles() {
|
212
|
+
// Make a set of all of the live files
|
213
|
+
std::set<uint64_t> live = pending_outputs_;
|
214
|
+
versions_->AddLiveFiles(&live);
|
215
|
+
|
216
|
+
std::vector<std::string> filenames;
|
217
|
+
env_->GetChildren(dbname_, &filenames); // Ignoring errors on purpose
|
218
|
+
uint64_t number;
|
219
|
+
FileType type;
|
220
|
+
for (size_t i = 0; i < filenames.size(); i++) {
|
221
|
+
if (ParseFileName(filenames[i], &number, &type)) {
|
222
|
+
bool keep = true;
|
223
|
+
switch (type) {
|
224
|
+
case kLogFile:
|
225
|
+
keep = ((number == versions_->LogNumber()) ||
|
226
|
+
(number == versions_->PrevLogNumber()));
|
227
|
+
break;
|
228
|
+
case kDescriptorFile:
|
229
|
+
// Keep my manifest file, and any newer incarnations'
|
230
|
+
// (in case there is a race that allows other incarnations)
|
231
|
+
keep = (number >= versions_->ManifestFileNumber());
|
232
|
+
break;
|
233
|
+
case kTableFile:
|
234
|
+
keep = (live.find(number) != live.end());
|
235
|
+
break;
|
236
|
+
case kTempFile:
|
237
|
+
// Any temp files that are currently being written to must
|
238
|
+
// be recorded in pending_outputs_, which is inserted into "live"
|
239
|
+
keep = (live.find(number) != live.end());
|
240
|
+
break;
|
241
|
+
case kCurrentFile:
|
242
|
+
case kDBLockFile:
|
243
|
+
case kInfoLogFile:
|
244
|
+
keep = true;
|
245
|
+
break;
|
246
|
+
}
|
247
|
+
|
248
|
+
if (!keep) {
|
249
|
+
if (type == kTableFile) {
|
250
|
+
table_cache_->Evict(number);
|
251
|
+
}
|
252
|
+
Log(env_, options_.info_log, "Delete type=%d #%lld\n",
|
253
|
+
int(type),
|
254
|
+
static_cast<unsigned long long>(number));
|
255
|
+
env_->DeleteFile(dbname_ + "/" + filenames[i]);
|
256
|
+
}
|
257
|
+
}
|
258
|
+
}
|
259
|
+
}
|
260
|
+
|
261
|
+
Status DBImpl::Recover(VersionEdit* edit) {
|
262
|
+
mutex_.AssertHeld();
|
263
|
+
|
264
|
+
// Ignore error from CreateDir since the creation of the DB is
|
265
|
+
// committed only when the descriptor is created, and this directory
|
266
|
+
// may already exist from a previous failed creation attempt.
|
267
|
+
env_->CreateDir(dbname_);
|
268
|
+
assert(db_lock_ == NULL);
|
269
|
+
Status s = env_->LockFile(LockFileName(dbname_), &db_lock_);
|
270
|
+
if (!s.ok()) {
|
271
|
+
return s;
|
272
|
+
}
|
273
|
+
|
274
|
+
if (!env_->FileExists(CurrentFileName(dbname_))) {
|
275
|
+
if (options_.create_if_missing) {
|
276
|
+
s = NewDB();
|
277
|
+
if (!s.ok()) {
|
278
|
+
return s;
|
279
|
+
}
|
280
|
+
} else {
|
281
|
+
return Status::InvalidArgument(
|
282
|
+
dbname_, "does not exist (create_if_missing is false)");
|
283
|
+
}
|
284
|
+
} else {
|
285
|
+
if (options_.error_if_exists) {
|
286
|
+
return Status::InvalidArgument(
|
287
|
+
dbname_, "exists (error_if_exists is true)");
|
288
|
+
}
|
289
|
+
}
|
290
|
+
|
291
|
+
s = versions_->Recover();
|
292
|
+
if (s.ok()) {
|
293
|
+
// Recover from the log files named in the descriptor
|
294
|
+
SequenceNumber max_sequence(0);
|
295
|
+
if (versions_->PrevLogNumber() != 0) { // log#==0 means no prev log
|
296
|
+
s = RecoverLogFile(versions_->PrevLogNumber(), edit, &max_sequence);
|
297
|
+
}
|
298
|
+
if (s.ok() && versions_->LogNumber() != 0) { // log#==0 for initial state
|
299
|
+
s = RecoverLogFile(versions_->LogNumber(), edit, &max_sequence);
|
300
|
+
}
|
301
|
+
if (s.ok()) {
|
302
|
+
if (versions_->LastSequence() < max_sequence) {
|
303
|
+
versions_->SetLastSequence(max_sequence);
|
304
|
+
}
|
305
|
+
}
|
306
|
+
}
|
307
|
+
|
308
|
+
return s;
|
309
|
+
}
|
310
|
+
|
311
|
+
Status DBImpl::RecoverLogFile(uint64_t log_number,
|
312
|
+
VersionEdit* edit,
|
313
|
+
SequenceNumber* max_sequence) {
|
314
|
+
struct LogReporter : public log::Reader::Reporter {
|
315
|
+
Env* env;
|
316
|
+
WritableFile* info_log;
|
317
|
+
const char* fname;
|
318
|
+
Status* status; // NULL if options_.paranoid_checks==false
|
319
|
+
virtual void Corruption(size_t bytes, const Status& s) {
|
320
|
+
Log(env, info_log, "%s%s: dropping %d bytes; %s",
|
321
|
+
(this->status == NULL ? "(ignoring error) " : ""),
|
322
|
+
fname, static_cast<int>(bytes), s.ToString().c_str());
|
323
|
+
if (this->status != NULL && this->status->ok()) *this->status = s;
|
324
|
+
}
|
325
|
+
};
|
326
|
+
|
327
|
+
mutex_.AssertHeld();
|
328
|
+
|
329
|
+
// Open the log file
|
330
|
+
std::string fname = LogFileName(dbname_, log_number);
|
331
|
+
SequentialFile* file;
|
332
|
+
Status status = env_->NewSequentialFile(fname, &file);
|
333
|
+
if (!status.ok()) {
|
334
|
+
MaybeIgnoreError(&status);
|
335
|
+
return status;
|
336
|
+
}
|
337
|
+
|
338
|
+
// Create the log reader.
|
339
|
+
LogReporter reporter;
|
340
|
+
reporter.env = env_;
|
341
|
+
reporter.info_log = options_.info_log;
|
342
|
+
reporter.fname = fname.c_str();
|
343
|
+
reporter.status = (options_.paranoid_checks ? &status : NULL);
|
344
|
+
// We intentially make log::Reader do checksumming even if
|
345
|
+
// paranoid_checks==false so that corruptions cause entire commits
|
346
|
+
// to be skipped instead of propagating bad information (like overly
|
347
|
+
// large sequence numbers).
|
348
|
+
log::Reader reader(file, &reporter, true/*checksum*/,
|
349
|
+
0/*initial_offset*/);
|
350
|
+
Log(env_, options_.info_log, "Recovering log #%llu",
|
351
|
+
(unsigned long long) log_number);
|
352
|
+
|
353
|
+
// Read all the records and add to a memtable
|
354
|
+
std::string scratch;
|
355
|
+
Slice record;
|
356
|
+
WriteBatch batch;
|
357
|
+
MemTable* mem = NULL;
|
358
|
+
while (reader.ReadRecord(&record, &scratch) &&
|
359
|
+
status.ok()) {
|
360
|
+
if (record.size() < 12) {
|
361
|
+
reporter.Corruption(
|
362
|
+
record.size(), Status::Corruption("log record too small"));
|
363
|
+
continue;
|
364
|
+
}
|
365
|
+
WriteBatchInternal::SetContents(&batch, record);
|
366
|
+
|
367
|
+
if (mem == NULL) {
|
368
|
+
mem = new MemTable(internal_comparator_);
|
369
|
+
mem->Ref();
|
370
|
+
}
|
371
|
+
status = WriteBatchInternal::InsertInto(&batch, mem);
|
372
|
+
MaybeIgnoreError(&status);
|
373
|
+
if (!status.ok()) {
|
374
|
+
break;
|
375
|
+
}
|
376
|
+
const SequenceNumber last_seq =
|
377
|
+
WriteBatchInternal::Sequence(&batch) +
|
378
|
+
WriteBatchInternal::Count(&batch) - 1;
|
379
|
+
if (last_seq > *max_sequence) {
|
380
|
+
*max_sequence = last_seq;
|
381
|
+
}
|
382
|
+
|
383
|
+
if (mem->ApproximateMemoryUsage() > options_.write_buffer_size) {
|
384
|
+
status = WriteLevel0Table(mem, edit);
|
385
|
+
if (!status.ok()) {
|
386
|
+
// Reflect errors immediately so that conditions like full
|
387
|
+
// file-systems cause the DB::Open() to fail.
|
388
|
+
break;
|
389
|
+
}
|
390
|
+
mem->Unref();
|
391
|
+
mem = NULL;
|
392
|
+
}
|
393
|
+
}
|
394
|
+
|
395
|
+
if (status.ok() && mem != NULL) {
|
396
|
+
status = WriteLevel0Table(mem, edit);
|
397
|
+
// Reflect errors immediately so that conditions like full
|
398
|
+
// file-systems cause the DB::Open() to fail.
|
399
|
+
}
|
400
|
+
|
401
|
+
if (mem != NULL) mem->Unref();
|
402
|
+
delete file;
|
403
|
+
return status;
|
404
|
+
}
|
405
|
+
|
406
|
+
Status DBImpl::WriteLevel0Table(MemTable* mem, VersionEdit* edit) {
|
407
|
+
mutex_.AssertHeld();
|
408
|
+
const uint64_t start_micros = env_->NowMicros();
|
409
|
+
FileMetaData meta;
|
410
|
+
meta.number = versions_->NewFileNumber();
|
411
|
+
pending_outputs_.insert(meta.number);
|
412
|
+
Iterator* iter = mem->NewIterator();
|
413
|
+
Log(env_, options_.info_log, "Level-0 table #%llu: started",
|
414
|
+
(unsigned long long) meta.number);
|
415
|
+
|
416
|
+
Status s;
|
417
|
+
{
|
418
|
+
mutex_.Unlock();
|
419
|
+
s = BuildTable(dbname_, env_, options_, table_cache_, iter, &meta, edit);
|
420
|
+
mutex_.Lock();
|
421
|
+
}
|
422
|
+
|
423
|
+
Log(env_, options_.info_log, "Level-0 table #%llu: %lld bytes %s",
|
424
|
+
(unsigned long long) meta.number,
|
425
|
+
(unsigned long long) meta.file_size,
|
426
|
+
s.ToString().c_str());
|
427
|
+
delete iter;
|
428
|
+
pending_outputs_.erase(meta.number);
|
429
|
+
|
430
|
+
CompactionStats stats;
|
431
|
+
stats.micros = env_->NowMicros() - start_micros;
|
432
|
+
stats.bytes_written = meta.file_size;
|
433
|
+
stats_[0].Add(stats);
|
434
|
+
return s;
|
435
|
+
}
|
436
|
+
|
437
|
+
Status DBImpl::CompactMemTable() {
|
438
|
+
mutex_.AssertHeld();
|
439
|
+
assert(imm_ != NULL);
|
440
|
+
assert(compacting_);
|
441
|
+
|
442
|
+
// Save the contents of the memtable as a new Table
|
443
|
+
VersionEdit edit;
|
444
|
+
Status s = WriteLevel0Table(imm_, &edit);
|
445
|
+
|
446
|
+
// Replace immutable memtable with the generated Table
|
447
|
+
if (s.ok()) {
|
448
|
+
edit.SetPrevLogNumber(0);
|
449
|
+
s = versions_->LogAndApply(&edit);
|
450
|
+
}
|
451
|
+
|
452
|
+
if (s.ok()) {
|
453
|
+
// Commit to the new state
|
454
|
+
imm_->Unref();
|
455
|
+
imm_ = NULL;
|
456
|
+
has_imm_.Release_Store(NULL);
|
457
|
+
DeleteObsoleteFiles();
|
458
|
+
}
|
459
|
+
|
460
|
+
compacting_cv_.SignalAll(); // Wake up waiter even if there was an error
|
461
|
+
return s;
|
462
|
+
}
|
463
|
+
|
464
|
+
void DBImpl::TEST_CompactRange(
|
465
|
+
int level,
|
466
|
+
const std::string& begin,
|
467
|
+
const std::string& end) {
|
468
|
+
MutexLock l(&mutex_);
|
469
|
+
while (compacting_) {
|
470
|
+
compacting_cv_.Wait();
|
471
|
+
}
|
472
|
+
Compaction* c = versions_->CompactRange(
|
473
|
+
level,
|
474
|
+
InternalKey(begin, kMaxSequenceNumber, kValueTypeForSeek),
|
475
|
+
InternalKey(end, 0, static_cast<ValueType>(0)));
|
476
|
+
|
477
|
+
if (c != NULL) {
|
478
|
+
CompactionState* compact = new CompactionState(c);
|
479
|
+
DoCompactionWork(compact); // Ignore error in test compaction
|
480
|
+
CleanupCompaction(compact);
|
481
|
+
}
|
482
|
+
|
483
|
+
// Start any background compaction that may have been delayed by this thread
|
484
|
+
MaybeScheduleCompaction();
|
485
|
+
}
|
486
|
+
|
487
|
+
Status DBImpl::TEST_CompactMemTable() {
|
488
|
+
MutexLock l(&mutex_);
|
489
|
+
Status s = MakeRoomForWrite(true /* force compaction */);
|
490
|
+
if (s.ok()) {
|
491
|
+
// Wait until the compaction completes
|
492
|
+
while (imm_ != NULL && bg_error_.ok()) {
|
493
|
+
compacting_cv_.Wait();
|
494
|
+
}
|
495
|
+
if (imm_ != NULL) {
|
496
|
+
s = bg_error_;
|
497
|
+
}
|
498
|
+
}
|
499
|
+
return s;
|
500
|
+
}
|
501
|
+
|
502
|
+
void DBImpl::MaybeScheduleCompaction() {
|
503
|
+
mutex_.AssertHeld();
|
504
|
+
if (bg_compaction_scheduled_) {
|
505
|
+
// Already scheduled
|
506
|
+
} else if (compacting_) {
|
507
|
+
// Some other thread is running a compaction. Do not conflict with it.
|
508
|
+
} else if (shutting_down_.Acquire_Load()) {
|
509
|
+
// DB is being deleted; no more background compactions
|
510
|
+
} else if (imm_ == NULL && !versions_->NeedsCompaction()) {
|
511
|
+
// No work to be done
|
512
|
+
} else {
|
513
|
+
bg_compaction_scheduled_ = true;
|
514
|
+
env_->Schedule(&DBImpl::BGWork, this);
|
515
|
+
}
|
516
|
+
}
|
517
|
+
|
518
|
+
void DBImpl::BGWork(void* db) {
|
519
|
+
reinterpret_cast<DBImpl*>(db)->BackgroundCall();
|
520
|
+
}
|
521
|
+
|
522
|
+
void DBImpl::BackgroundCall() {
|
523
|
+
MutexLock l(&mutex_);
|
524
|
+
assert(bg_compaction_scheduled_);
|
525
|
+
if (!shutting_down_.Acquire_Load() &&
|
526
|
+
!compacting_) {
|
527
|
+
BackgroundCompaction();
|
528
|
+
}
|
529
|
+
bg_compaction_scheduled_ = false;
|
530
|
+
bg_cv_.SignalAll();
|
531
|
+
|
532
|
+
// Previous compaction may have produced too many files in a level,
|
533
|
+
// so reschedule another compaction if needed.
|
534
|
+
MaybeScheduleCompaction();
|
535
|
+
}
|
536
|
+
|
537
|
+
void DBImpl::BackgroundCompaction() {
|
538
|
+
mutex_.AssertHeld();
|
539
|
+
assert(!compacting_);
|
540
|
+
|
541
|
+
if (imm_ != NULL) {
|
542
|
+
compacting_ = true;
|
543
|
+
CompactMemTable();
|
544
|
+
compacting_ = false;
|
545
|
+
compacting_cv_.SignalAll();
|
546
|
+
return;
|
547
|
+
}
|
548
|
+
|
549
|
+
Compaction* c = versions_->PickCompaction();
|
550
|
+
if (c == NULL) {
|
551
|
+
// Nothing to do
|
552
|
+
return;
|
553
|
+
}
|
554
|
+
|
555
|
+
Status status;
|
556
|
+
if (c->IsTrivialMove()) {
|
557
|
+
// Move file to next level
|
558
|
+
assert(c->num_input_files(0) == 1);
|
559
|
+
FileMetaData* f = c->input(0, 0);
|
560
|
+
c->edit()->DeleteFile(c->level(), f->number);
|
561
|
+
c->edit()->AddFile(c->level() + 1, f->number, f->file_size,
|
562
|
+
f->smallest, f->largest);
|
563
|
+
status = versions_->LogAndApply(c->edit());
|
564
|
+
Log(env_, options_.info_log, "Moved #%lld to level-%d %lld bytes %s\n",
|
565
|
+
static_cast<unsigned long long>(f->number),
|
566
|
+
c->level() + 1,
|
567
|
+
static_cast<unsigned long long>(f->file_size),
|
568
|
+
status.ToString().c_str());
|
569
|
+
} else {
|
570
|
+
CompactionState* compact = new CompactionState(c);
|
571
|
+
status = DoCompactionWork(compact);
|
572
|
+
CleanupCompaction(compact);
|
573
|
+
}
|
574
|
+
delete c;
|
575
|
+
|
576
|
+
if (status.ok()) {
|
577
|
+
// Done
|
578
|
+
} else if (shutting_down_.Acquire_Load()) {
|
579
|
+
// Ignore compaction errors found during shutting down
|
580
|
+
} else {
|
581
|
+
Log(env_, options_.info_log,
|
582
|
+
"Compaction error: %s", status.ToString().c_str());
|
583
|
+
if (options_.paranoid_checks && bg_error_.ok()) {
|
584
|
+
bg_error_ = status;
|
585
|
+
}
|
586
|
+
}
|
587
|
+
}
|
588
|
+
|
589
|
+
void DBImpl::CleanupCompaction(CompactionState* compact) {
|
590
|
+
mutex_.AssertHeld();
|
591
|
+
if (compact->builder != NULL) {
|
592
|
+
// May happen if we get a shutdown call in the middle of compaction
|
593
|
+
compact->builder->Abandon();
|
594
|
+
delete compact->builder;
|
595
|
+
} else {
|
596
|
+
assert(compact->outfile == NULL);
|
597
|
+
}
|
598
|
+
delete compact->outfile;
|
599
|
+
for (size_t i = 0; i < compact->outputs.size(); i++) {
|
600
|
+
const CompactionState::Output& out = compact->outputs[i];
|
601
|
+
pending_outputs_.erase(out.number);
|
602
|
+
}
|
603
|
+
delete compact;
|
604
|
+
}
|
605
|
+
|
606
|
+
Status DBImpl::OpenCompactionOutputFile(CompactionState* compact) {
|
607
|
+
assert(compact != NULL);
|
608
|
+
assert(compact->builder == NULL);
|
609
|
+
uint64_t file_number;
|
610
|
+
{
|
611
|
+
mutex_.Lock();
|
612
|
+
file_number = versions_->NewFileNumber();
|
613
|
+
pending_outputs_.insert(file_number);
|
614
|
+
CompactionState::Output out;
|
615
|
+
out.number = file_number;
|
616
|
+
out.smallest.Clear();
|
617
|
+
out.largest.Clear();
|
618
|
+
compact->outputs.push_back(out);
|
619
|
+
mutex_.Unlock();
|
620
|
+
}
|
621
|
+
|
622
|
+
// Make the output file
|
623
|
+
std::string fname = TableFileName(dbname_, file_number);
|
624
|
+
Status s = env_->NewWritableFile(fname, &compact->outfile);
|
625
|
+
if (s.ok()) {
|
626
|
+
compact->builder = new TableBuilder(options_, compact->outfile);
|
627
|
+
}
|
628
|
+
return s;
|
629
|
+
}
|
630
|
+
|
631
|
+
Status DBImpl::FinishCompactionOutputFile(CompactionState* compact,
|
632
|
+
Iterator* input) {
|
633
|
+
assert(compact != NULL);
|
634
|
+
assert(compact->outfile != NULL);
|
635
|
+
assert(compact->builder != NULL);
|
636
|
+
|
637
|
+
const uint64_t output_number = compact->current_output()->number;
|
638
|
+
assert(output_number != 0);
|
639
|
+
|
640
|
+
// Check for iterator errors
|
641
|
+
Status s = input->status();
|
642
|
+
const uint64_t current_entries = compact->builder->NumEntries();
|
643
|
+
if (s.ok()) {
|
644
|
+
s = compact->builder->Finish();
|
645
|
+
} else {
|
646
|
+
compact->builder->Abandon();
|
647
|
+
}
|
648
|
+
const uint64_t current_bytes = compact->builder->FileSize();
|
649
|
+
compact->current_output()->file_size = current_bytes;
|
650
|
+
compact->total_bytes += current_bytes;
|
651
|
+
delete compact->builder;
|
652
|
+
compact->builder = NULL;
|
653
|
+
|
654
|
+
// Finish and check for file errors
|
655
|
+
if (s.ok()) {
|
656
|
+
s = compact->outfile->Sync();
|
657
|
+
}
|
658
|
+
if (s.ok()) {
|
659
|
+
s = compact->outfile->Close();
|
660
|
+
}
|
661
|
+
delete compact->outfile;
|
662
|
+
compact->outfile = NULL;
|
663
|
+
|
664
|
+
if (s.ok() && current_entries > 0) {
|
665
|
+
// Verify that the table is usable
|
666
|
+
Iterator* iter = table_cache_->NewIterator(ReadOptions(),
|
667
|
+
output_number,
|
668
|
+
current_bytes);
|
669
|
+
s = iter->status();
|
670
|
+
delete iter;
|
671
|
+
if (s.ok()) {
|
672
|
+
Log(env_, options_.info_log,
|
673
|
+
"Generated table #%llu: %lld keys, %lld bytes",
|
674
|
+
(unsigned long long) output_number,
|
675
|
+
(unsigned long long) current_entries,
|
676
|
+
(unsigned long long) current_bytes);
|
677
|
+
}
|
678
|
+
}
|
679
|
+
return s;
|
680
|
+
}
|
681
|
+
|
682
|
+
|
683
|
+
Status DBImpl::InstallCompactionResults(CompactionState* compact) {
|
684
|
+
mutex_.AssertHeld();
|
685
|
+
Log(env_, options_.info_log, "Compacted %d@%d + %d@%d files => %lld bytes",
|
686
|
+
compact->compaction->num_input_files(0),
|
687
|
+
compact->compaction->level(),
|
688
|
+
compact->compaction->num_input_files(1),
|
689
|
+
compact->compaction->level() + 1,
|
690
|
+
static_cast<long long>(compact->total_bytes));
|
691
|
+
|
692
|
+
// Add compaction outputs
|
693
|
+
compact->compaction->AddInputDeletions(compact->compaction->edit());
|
694
|
+
const int level = compact->compaction->level();
|
695
|
+
for (size_t i = 0; i < compact->outputs.size(); i++) {
|
696
|
+
const CompactionState::Output& out = compact->outputs[i];
|
697
|
+
compact->compaction->edit()->AddFile(
|
698
|
+
level + 1,
|
699
|
+
out.number, out.file_size, out.smallest, out.largest);
|
700
|
+
pending_outputs_.erase(out.number);
|
701
|
+
}
|
702
|
+
compact->outputs.clear();
|
703
|
+
|
704
|
+
Status s = versions_->LogAndApply(compact->compaction->edit());
|
705
|
+
if (s.ok()) {
|
706
|
+
compact->compaction->ReleaseInputs();
|
707
|
+
DeleteObsoleteFiles();
|
708
|
+
} else {
|
709
|
+
// Discard any files we may have created during this failed compaction
|
710
|
+
for (size_t i = 0; i < compact->outputs.size(); i++) {
|
711
|
+
env_->DeleteFile(TableFileName(dbname_, compact->outputs[i].number));
|
712
|
+
}
|
713
|
+
}
|
714
|
+
return s;
|
715
|
+
}
|
716
|
+
|
717
|
+
Status DBImpl::DoCompactionWork(CompactionState* compact) {
|
718
|
+
const uint64_t start_micros = env_->NowMicros();
|
719
|
+
int64_t imm_micros = 0; // Micros spent doing imm_ compactions
|
720
|
+
|
721
|
+
Log(env_, options_.info_log, "Compacting %d@%d + %d@%d files",
|
722
|
+
compact->compaction->num_input_files(0),
|
723
|
+
compact->compaction->level(),
|
724
|
+
compact->compaction->num_input_files(1),
|
725
|
+
compact->compaction->level() + 1);
|
726
|
+
|
727
|
+
assert(versions_->NumLevelFiles(compact->compaction->level()) > 0);
|
728
|
+
assert(compact->builder == NULL);
|
729
|
+
assert(compact->outfile == NULL);
|
730
|
+
if (snapshots_.empty()) {
|
731
|
+
compact->smallest_snapshot = versions_->LastSequence();
|
732
|
+
} else {
|
733
|
+
compact->smallest_snapshot = snapshots_.oldest()->number_;
|
734
|
+
}
|
735
|
+
|
736
|
+
// Release mutex while we're actually doing the compaction work
|
737
|
+
compacting_ = true;
|
738
|
+
mutex_.Unlock();
|
739
|
+
|
740
|
+
Iterator* input = versions_->MakeInputIterator(compact->compaction);
|
741
|
+
input->SeekToFirst();
|
742
|
+
Status status;
|
743
|
+
ParsedInternalKey ikey;
|
744
|
+
std::string current_user_key;
|
745
|
+
bool has_current_user_key = false;
|
746
|
+
SequenceNumber last_sequence_for_key = kMaxSequenceNumber;
|
747
|
+
for (; input->Valid() && !shutting_down_.Acquire_Load(); ) {
|
748
|
+
// Prioritize immutable compaction work
|
749
|
+
if (has_imm_.NoBarrier_Load() != NULL) {
|
750
|
+
const uint64_t imm_start = env_->NowMicros();
|
751
|
+
mutex_.Lock();
|
752
|
+
if (imm_ != NULL) {
|
753
|
+
CompactMemTable();
|
754
|
+
compacting_cv_.SignalAll(); // Wakeup MakeRoomForWrite() if necessary
|
755
|
+
}
|
756
|
+
mutex_.Unlock();
|
757
|
+
imm_micros += (env_->NowMicros() - imm_start);
|
758
|
+
}
|
759
|
+
|
760
|
+
Slice key = input->key();
|
761
|
+
if (compact->compaction->ShouldStopBefore(key) &&
|
762
|
+
compact->builder != NULL) {
|
763
|
+
status = FinishCompactionOutputFile(compact, input);
|
764
|
+
if (!status.ok()) {
|
765
|
+
break;
|
766
|
+
}
|
767
|
+
}
|
768
|
+
|
769
|
+
// Handle key/value, add to state, etc.
|
770
|
+
bool drop = false;
|
771
|
+
if (!ParseInternalKey(key, &ikey)) {
|
772
|
+
// Do not hide error keys
|
773
|
+
current_user_key.clear();
|
774
|
+
has_current_user_key = false;
|
775
|
+
last_sequence_for_key = kMaxSequenceNumber;
|
776
|
+
} else {
|
777
|
+
if (!has_current_user_key ||
|
778
|
+
user_comparator()->Compare(ikey.user_key,
|
779
|
+
Slice(current_user_key)) != 0) {
|
780
|
+
// First occurrence of this user key
|
781
|
+
current_user_key.assign(ikey.user_key.data(), ikey.user_key.size());
|
782
|
+
has_current_user_key = true;
|
783
|
+
last_sequence_for_key = kMaxSequenceNumber;
|
784
|
+
}
|
785
|
+
|
786
|
+
if (last_sequence_for_key <= compact->smallest_snapshot) {
|
787
|
+
// Hidden by an newer entry for same user key
|
788
|
+
drop = true; // (A)
|
789
|
+
} else if (ikey.type == kTypeDeletion &&
|
790
|
+
ikey.sequence <= compact->smallest_snapshot &&
|
791
|
+
compact->compaction->IsBaseLevelForKey(ikey.user_key)) {
|
792
|
+
// For this user key:
|
793
|
+
// (1) there is no data in higher levels
|
794
|
+
// (2) data in lower levels will have larger sequence numbers
|
795
|
+
// (3) data in layers that are being compacted here and have
|
796
|
+
// smaller sequence numbers will be dropped in the next
|
797
|
+
// few iterations of this loop (by rule (A) above).
|
798
|
+
// Therefore this deletion marker is obsolete and can be dropped.
|
799
|
+
drop = true;
|
800
|
+
}
|
801
|
+
|
802
|
+
last_sequence_for_key = ikey.sequence;
|
803
|
+
}
|
804
|
+
#if 0
|
805
|
+
Log(env_, options_.info_log,
|
806
|
+
" Compact: %s, seq %d, type: %d %d, drop: %d, is_base: %d, "
|
807
|
+
"%d smallest_snapshot: %d",
|
808
|
+
ikey.user_key.ToString().c_str(),
|
809
|
+
(int)ikey.sequence, ikey.type, kTypeValue, drop,
|
810
|
+
compact->compaction->IsBaseLevelForKey(ikey.user_key),
|
811
|
+
(int)last_sequence_for_key, (int)compact->smallest_snapshot);
|
812
|
+
#endif
|
813
|
+
|
814
|
+
if (!drop) {
|
815
|
+
// Open output file if necessary
|
816
|
+
if (compact->builder == NULL) {
|
817
|
+
status = OpenCompactionOutputFile(compact);
|
818
|
+
if (!status.ok()) {
|
819
|
+
break;
|
820
|
+
}
|
821
|
+
}
|
822
|
+
if (compact->builder->NumEntries() == 0) {
|
823
|
+
compact->current_output()->smallest.DecodeFrom(key);
|
824
|
+
}
|
825
|
+
compact->current_output()->largest.DecodeFrom(key);
|
826
|
+
compact->builder->Add(key, input->value());
|
827
|
+
|
828
|
+
// Close output file if it is big enough
|
829
|
+
if (compact->builder->FileSize() >=
|
830
|
+
compact->compaction->MaxOutputFileSize()) {
|
831
|
+
status = FinishCompactionOutputFile(compact, input);
|
832
|
+
if (!status.ok()) {
|
833
|
+
break;
|
834
|
+
}
|
835
|
+
}
|
836
|
+
}
|
837
|
+
|
838
|
+
input->Next();
|
839
|
+
}
|
840
|
+
|
841
|
+
if (status.ok() && shutting_down_.Acquire_Load()) {
|
842
|
+
status = Status::IOError("Deleting DB during compaction");
|
843
|
+
}
|
844
|
+
if (status.ok() && compact->builder != NULL) {
|
845
|
+
status = FinishCompactionOutputFile(compact, input);
|
846
|
+
}
|
847
|
+
if (status.ok()) {
|
848
|
+
status = input->status();
|
849
|
+
}
|
850
|
+
delete input;
|
851
|
+
input = NULL;
|
852
|
+
|
853
|
+
CompactionStats stats;
|
854
|
+
stats.micros = env_->NowMicros() - start_micros - imm_micros;
|
855
|
+
for (int which = 0; which < 2; which++) {
|
856
|
+
for (int i = 0; i < compact->compaction->num_input_files(which); i++) {
|
857
|
+
stats.bytes_read += compact->compaction->input(which, i)->file_size;
|
858
|
+
}
|
859
|
+
}
|
860
|
+
for (size_t i = 0; i < compact->outputs.size(); i++) {
|
861
|
+
stats.bytes_written += compact->outputs[i].file_size;
|
862
|
+
}
|
863
|
+
|
864
|
+
mutex_.Lock();
|
865
|
+
stats_[compact->compaction->level() + 1].Add(stats);
|
866
|
+
|
867
|
+
if (status.ok()) {
|
868
|
+
status = InstallCompactionResults(compact);
|
869
|
+
}
|
870
|
+
compacting_ = false;
|
871
|
+
compacting_cv_.SignalAll();
|
872
|
+
VersionSet::LevelSummaryStorage tmp;
|
873
|
+
Log(env_, options_.info_log,
|
874
|
+
"compacted to: %s", versions_->LevelSummary(&tmp));
|
875
|
+
return status;
|
876
|
+
}
|
877
|
+
|
878
|
+
namespace {
|
879
|
+
struct IterState {
|
880
|
+
port::Mutex* mu;
|
881
|
+
Version* version;
|
882
|
+
MemTable* mem;
|
883
|
+
MemTable* imm;
|
884
|
+
};
|
885
|
+
|
886
|
+
static void CleanupIteratorState(void* arg1, void* arg2) {
|
887
|
+
IterState* state = reinterpret_cast<IterState*>(arg1);
|
888
|
+
state->mu->Lock();
|
889
|
+
state->mem->Unref();
|
890
|
+
if (state->imm != NULL) state->imm->Unref();
|
891
|
+
state->version->Unref();
|
892
|
+
state->mu->Unlock();
|
893
|
+
delete state;
|
894
|
+
}
|
895
|
+
}
|
896
|
+
|
897
|
+
Iterator* DBImpl::NewInternalIterator(const ReadOptions& options,
|
898
|
+
SequenceNumber* latest_snapshot) {
|
899
|
+
IterState* cleanup = new IterState;
|
900
|
+
mutex_.Lock();
|
901
|
+
*latest_snapshot = versions_->LastSequence();
|
902
|
+
|
903
|
+
// Collect together all needed child iterators
|
904
|
+
std::vector<Iterator*> list;
|
905
|
+
list.push_back(mem_->NewIterator());
|
906
|
+
mem_->Ref();
|
907
|
+
if (imm_ != NULL) {
|
908
|
+
list.push_back(imm_->NewIterator());
|
909
|
+
imm_->Ref();
|
910
|
+
}
|
911
|
+
versions_->current()->AddIterators(options, &list);
|
912
|
+
Iterator* internal_iter =
|
913
|
+
NewMergingIterator(&internal_comparator_, &list[0], list.size());
|
914
|
+
versions_->current()->Ref();
|
915
|
+
|
916
|
+
cleanup->mu = &mutex_;
|
917
|
+
cleanup->mem = mem_;
|
918
|
+
cleanup->imm = imm_;
|
919
|
+
cleanup->version = versions_->current();
|
920
|
+
internal_iter->RegisterCleanup(CleanupIteratorState, cleanup, NULL);
|
921
|
+
|
922
|
+
mutex_.Unlock();
|
923
|
+
return internal_iter;
|
924
|
+
}
|
925
|
+
|
926
|
+
Iterator* DBImpl::TEST_NewInternalIterator() {
|
927
|
+
SequenceNumber ignored;
|
928
|
+
return NewInternalIterator(ReadOptions(), &ignored);
|
929
|
+
}
|
930
|
+
|
931
|
+
int64_t DBImpl::TEST_MaxNextLevelOverlappingBytes() {
|
932
|
+
MutexLock l(&mutex_);
|
933
|
+
return versions_->MaxNextLevelOverlappingBytes();
|
934
|
+
}
|
935
|
+
|
936
|
+
Status DBImpl::Get(const ReadOptions& options,
|
937
|
+
const Slice& key,
|
938
|
+
std::string* value) {
|
939
|
+
// TODO(opt): faster implementation
|
940
|
+
Iterator* iter = NewIterator(options);
|
941
|
+
iter->Seek(key);
|
942
|
+
bool found = false;
|
943
|
+
if (iter->Valid() && user_comparator()->Compare(key, iter->key()) == 0) {
|
944
|
+
Slice v = iter->value();
|
945
|
+
value->assign(v.data(), v.size());
|
946
|
+
found = true;
|
947
|
+
}
|
948
|
+
// Non-OK iterator status trumps everything else
|
949
|
+
Status result = iter->status();
|
950
|
+
if (result.ok() && !found) {
|
951
|
+
result = Status::NotFound(Slice()); // Use an empty error message for speed
|
952
|
+
}
|
953
|
+
delete iter;
|
954
|
+
return result;
|
955
|
+
}
|
956
|
+
|
957
|
+
Iterator* DBImpl::NewIterator(const ReadOptions& options) {
|
958
|
+
SequenceNumber latest_snapshot;
|
959
|
+
Iterator* internal_iter = NewInternalIterator(options, &latest_snapshot);
|
960
|
+
return NewDBIterator(
|
961
|
+
&dbname_, env_, user_comparator(), internal_iter,
|
962
|
+
(options.snapshot != NULL
|
963
|
+
? reinterpret_cast<const SnapshotImpl*>(options.snapshot)->number_
|
964
|
+
: latest_snapshot));
|
965
|
+
}
|
966
|
+
|
967
|
+
const Snapshot* DBImpl::GetSnapshot() {
|
968
|
+
MutexLock l(&mutex_);
|
969
|
+
return snapshots_.New(versions_->LastSequence());
|
970
|
+
}
|
971
|
+
|
972
|
+
void DBImpl::ReleaseSnapshot(const Snapshot* s) {
|
973
|
+
MutexLock l(&mutex_);
|
974
|
+
snapshots_.Delete(reinterpret_cast<const SnapshotImpl*>(s));
|
975
|
+
}
|
976
|
+
|
977
|
+
// Convenience methods
|
978
|
+
Status DBImpl::Put(const WriteOptions& o, const Slice& key, const Slice& val) {
|
979
|
+
return DB::Put(o, key, val);
|
980
|
+
}
|
981
|
+
|
982
|
+
Status DBImpl::Delete(const WriteOptions& options, const Slice& key) {
|
983
|
+
return DB::Delete(options, key);
|
984
|
+
}
|
985
|
+
|
986
|
+
Status DBImpl::Write(const WriteOptions& options, WriteBatch* updates) {
|
987
|
+
Status status;
|
988
|
+
MutexLock l(&mutex_);
|
989
|
+
status = MakeRoomForWrite(false); // May temporarily release lock and wait
|
990
|
+
uint64_t last_sequence = versions_->LastSequence();
|
991
|
+
if (status.ok()) {
|
992
|
+
WriteBatchInternal::SetSequence(updates, last_sequence + 1);
|
993
|
+
last_sequence += WriteBatchInternal::Count(updates);
|
994
|
+
versions_->SetLastSequence(last_sequence);
|
995
|
+
|
996
|
+
// Add to log and apply to memtable
|
997
|
+
status = log_->AddRecord(WriteBatchInternal::Contents(updates));
|
998
|
+
if (status.ok() && options.sync) {
|
999
|
+
status = logfile_->Sync();
|
1000
|
+
}
|
1001
|
+
if (status.ok()) {
|
1002
|
+
status = WriteBatchInternal::InsertInto(updates, mem_);
|
1003
|
+
}
|
1004
|
+
}
|
1005
|
+
if (options.post_write_snapshot != NULL) {
|
1006
|
+
*options.post_write_snapshot =
|
1007
|
+
status.ok() ? snapshots_.New(last_sequence) : NULL;
|
1008
|
+
}
|
1009
|
+
return status;
|
1010
|
+
}
|
1011
|
+
|
1012
|
+
Status DBImpl::MakeRoomForWrite(bool force) {
|
1013
|
+
mutex_.AssertHeld();
|
1014
|
+
bool allow_delay = !force;
|
1015
|
+
Status s;
|
1016
|
+
while (true) {
|
1017
|
+
if (!bg_error_.ok()) {
|
1018
|
+
// Yield previous error
|
1019
|
+
s = bg_error_;
|
1020
|
+
break;
|
1021
|
+
} else if (
|
1022
|
+
allow_delay &&
|
1023
|
+
versions_->NumLevelFiles(0) >= config::kL0_SlowdownWritesTrigger) {
|
1024
|
+
// We are getting close to hitting a hard limit on the number of
|
1025
|
+
// L0 files. Rather than delaying a single write by several
|
1026
|
+
// seconds when we hit the hard limit, start delaying each
|
1027
|
+
// individual write by 1ms to reduce latency variance. Also,
|
1028
|
+
// this delay hands over some CPU to the compaction thread in
|
1029
|
+
// case it is sharing the same core as the writer.
|
1030
|
+
mutex_.Unlock();
|
1031
|
+
env_->SleepForMicroseconds(1000);
|
1032
|
+
allow_delay = false; // Do not delay a single write more than once
|
1033
|
+
mutex_.Lock();
|
1034
|
+
} else if (!force &&
|
1035
|
+
(mem_->ApproximateMemoryUsage() <= options_.write_buffer_size)) {
|
1036
|
+
// There is room in current memtable
|
1037
|
+
break;
|
1038
|
+
} else if (imm_ != NULL) {
|
1039
|
+
// We have filled up the current memtable, but the previous
|
1040
|
+
// one is still being compacted, so we wait.
|
1041
|
+
compacting_cv_.Wait();
|
1042
|
+
} else if (versions_->NumLevelFiles(0) >= config::kL0_StopWritesTrigger) {
|
1043
|
+
// There are too many level-0 files.
|
1044
|
+
compacting_cv_.Wait();
|
1045
|
+
} else {
|
1046
|
+
// Attempt to switch to a new memtable and trigger compaction of old
|
1047
|
+
assert(versions_->PrevLogNumber() == 0);
|
1048
|
+
uint64_t new_log_number = versions_->NewFileNumber();
|
1049
|
+
WritableFile* lfile = NULL;
|
1050
|
+
s = env_->NewWritableFile(LogFileName(dbname_, new_log_number), &lfile);
|
1051
|
+
if (!s.ok()) {
|
1052
|
+
break;
|
1053
|
+
}
|
1054
|
+
VersionEdit edit;
|
1055
|
+
edit.SetPrevLogNumber(versions_->LogNumber());
|
1056
|
+
edit.SetLogNumber(new_log_number);
|
1057
|
+
s = versions_->LogAndApply(&edit);
|
1058
|
+
if (!s.ok()) {
|
1059
|
+
delete lfile;
|
1060
|
+
env_->DeleteFile(LogFileName(dbname_, new_log_number));
|
1061
|
+
break;
|
1062
|
+
}
|
1063
|
+
delete log_;
|
1064
|
+
delete logfile_;
|
1065
|
+
logfile_ = lfile;
|
1066
|
+
log_ = new log::Writer(lfile);
|
1067
|
+
imm_ = mem_;
|
1068
|
+
has_imm_.Release_Store(imm_);
|
1069
|
+
mem_ = new MemTable(internal_comparator_);
|
1070
|
+
mem_->Ref();
|
1071
|
+
force = false; // Do not force another compaction if have room
|
1072
|
+
MaybeScheduleCompaction();
|
1073
|
+
}
|
1074
|
+
}
|
1075
|
+
return s;
|
1076
|
+
}
|
1077
|
+
|
1078
|
+
bool DBImpl::GetProperty(const Slice& property, std::string* value) {
|
1079
|
+
value->clear();
|
1080
|
+
|
1081
|
+
MutexLock l(&mutex_);
|
1082
|
+
Slice in = property;
|
1083
|
+
Slice prefix("leveldb.");
|
1084
|
+
if (!in.starts_with(prefix)) return false;
|
1085
|
+
in.remove_prefix(prefix.size());
|
1086
|
+
|
1087
|
+
if (in.starts_with("num-files-at-level")) {
|
1088
|
+
in.remove_prefix(strlen("num-files-at-level"));
|
1089
|
+
uint64_t level;
|
1090
|
+
bool ok = ConsumeDecimalNumber(&in, &level) && in.empty();
|
1091
|
+
if (!ok || level < 0 || level >= config::kNumLevels) {
|
1092
|
+
return false;
|
1093
|
+
} else {
|
1094
|
+
char buf[100];
|
1095
|
+
snprintf(buf, sizeof(buf), "%d",
|
1096
|
+
versions_->NumLevelFiles(static_cast<int>(level)));
|
1097
|
+
*value = buf;
|
1098
|
+
return true;
|
1099
|
+
}
|
1100
|
+
} else if (in == "stats") {
|
1101
|
+
char buf[200];
|
1102
|
+
snprintf(buf, sizeof(buf),
|
1103
|
+
" Compactions\n"
|
1104
|
+
"Level Files Size(MB) Time(sec) Read(MB) Write(MB)\n"
|
1105
|
+
"--------------------------------------------------\n"
|
1106
|
+
);
|
1107
|
+
value->append(buf);
|
1108
|
+
for (int level = 0; level < config::kNumLevels; level++) {
|
1109
|
+
int files = versions_->NumLevelFiles(level);
|
1110
|
+
if (stats_[level].micros > 0 || files > 0) {
|
1111
|
+
snprintf(
|
1112
|
+
buf, sizeof(buf),
|
1113
|
+
"%3d %8d %8.0f %9.0f %8.0f %9.0f\n",
|
1114
|
+
level,
|
1115
|
+
files,
|
1116
|
+
versions_->NumLevelBytes(level) / 1048576.0,
|
1117
|
+
stats_[level].micros / 1e6,
|
1118
|
+
stats_[level].bytes_read / 1048576.0,
|
1119
|
+
stats_[level].bytes_written / 1048576.0);
|
1120
|
+
value->append(buf);
|
1121
|
+
}
|
1122
|
+
}
|
1123
|
+
return true;
|
1124
|
+
}
|
1125
|
+
|
1126
|
+
return false;
|
1127
|
+
}
|
1128
|
+
|
1129
|
+
void DBImpl::GetApproximateSizes(
|
1130
|
+
const Range* range, int n,
|
1131
|
+
uint64_t* sizes) {
|
1132
|
+
// TODO(opt): better implementation
|
1133
|
+
Version* v;
|
1134
|
+
{
|
1135
|
+
MutexLock l(&mutex_);
|
1136
|
+
versions_->current()->Ref();
|
1137
|
+
v = versions_->current();
|
1138
|
+
}
|
1139
|
+
|
1140
|
+
for (int i = 0; i < n; i++) {
|
1141
|
+
// Convert user_key into a corresponding internal key.
|
1142
|
+
InternalKey k1(range[i].start, kMaxSequenceNumber, kValueTypeForSeek);
|
1143
|
+
InternalKey k2(range[i].limit, kMaxSequenceNumber, kValueTypeForSeek);
|
1144
|
+
uint64_t start = versions_->ApproximateOffsetOf(v, k1);
|
1145
|
+
uint64_t limit = versions_->ApproximateOffsetOf(v, k2);
|
1146
|
+
sizes[i] = (limit >= start ? limit - start : 0);
|
1147
|
+
}
|
1148
|
+
|
1149
|
+
{
|
1150
|
+
MutexLock l(&mutex_);
|
1151
|
+
v->Unref();
|
1152
|
+
}
|
1153
|
+
}
|
1154
|
+
|
1155
|
+
// Default implementations of convenience methods that subclasses of DB
|
1156
|
+
// can call if they wish
|
1157
|
+
Status DB::Put(const WriteOptions& opt, const Slice& key, const Slice& value) {
|
1158
|
+
WriteBatch batch;
|
1159
|
+
batch.Put(key, value);
|
1160
|
+
return Write(opt, &batch);
|
1161
|
+
}
|
1162
|
+
|
1163
|
+
Status DB::Delete(const WriteOptions& opt, const Slice& key) {
|
1164
|
+
WriteBatch batch;
|
1165
|
+
batch.Delete(key);
|
1166
|
+
return Write(opt, &batch);
|
1167
|
+
}
|
1168
|
+
|
1169
|
+
DB::~DB() { }
|
1170
|
+
|
1171
|
+
Status DB::Open(const Options& options, const std::string& dbname,
|
1172
|
+
DB** dbptr) {
|
1173
|
+
*dbptr = NULL;
|
1174
|
+
|
1175
|
+
DBImpl* impl = new DBImpl(options, dbname);
|
1176
|
+
impl->mutex_.Lock();
|
1177
|
+
VersionEdit edit;
|
1178
|
+
Status s = impl->Recover(&edit); // Handles create_if_missing, error_if_exists
|
1179
|
+
if (s.ok()) {
|
1180
|
+
uint64_t new_log_number = impl->versions_->NewFileNumber();
|
1181
|
+
WritableFile* lfile;
|
1182
|
+
s = options.env->NewWritableFile(LogFileName(dbname, new_log_number),
|
1183
|
+
&lfile);
|
1184
|
+
if (s.ok()) {
|
1185
|
+
edit.SetLogNumber(new_log_number);
|
1186
|
+
impl->logfile_ = lfile;
|
1187
|
+
impl->log_ = new log::Writer(lfile);
|
1188
|
+
s = impl->versions_->LogAndApply(&edit);
|
1189
|
+
}
|
1190
|
+
if (s.ok()) {
|
1191
|
+
impl->DeleteObsoleteFiles();
|
1192
|
+
impl->MaybeScheduleCompaction();
|
1193
|
+
}
|
1194
|
+
}
|
1195
|
+
impl->mutex_.Unlock();
|
1196
|
+
if (s.ok()) {
|
1197
|
+
*dbptr = impl;
|
1198
|
+
} else {
|
1199
|
+
delete impl;
|
1200
|
+
}
|
1201
|
+
return s;
|
1202
|
+
}
|
1203
|
+
|
1204
|
+
Snapshot::~Snapshot() {
|
1205
|
+
}
|
1206
|
+
|
1207
|
+
Status DestroyDB(const std::string& dbname, const Options& options) {
|
1208
|
+
Env* env = options.env;
|
1209
|
+
std::vector<std::string> filenames;
|
1210
|
+
// Ignore error in case directory does not exist
|
1211
|
+
env->GetChildren(dbname, &filenames);
|
1212
|
+
if (filenames.empty()) {
|
1213
|
+
return Status::OK();
|
1214
|
+
}
|
1215
|
+
|
1216
|
+
FileLock* lock;
|
1217
|
+
Status result = env->LockFile(LockFileName(dbname), &lock);
|
1218
|
+
if (result.ok()) {
|
1219
|
+
uint64_t number;
|
1220
|
+
FileType type;
|
1221
|
+
for (size_t i = 0; i < filenames.size(); i++) {
|
1222
|
+
if (ParseFileName(filenames[i], &number, &type)) {
|
1223
|
+
Status del = env->DeleteFile(dbname + "/" + filenames[i]);
|
1224
|
+
if (result.ok() && !del.ok()) {
|
1225
|
+
result = del;
|
1226
|
+
}
|
1227
|
+
}
|
1228
|
+
}
|
1229
|
+
env->UnlockFile(lock); // Ignore error since state is already gone
|
1230
|
+
env->DeleteFile(LockFileName(dbname));
|
1231
|
+
env->DeleteDir(dbname); // Ignore error in case dir contains other files
|
1232
|
+
}
|
1233
|
+
return result;
|
1234
|
+
}
|
1235
|
+
|
1236
|
+
}
|