1 // Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
2 // This source code is licensed under both the GPLv2 (found in the
3 // COPYING file in the root directory) and Apache 2.0 License
4 // (found in the LICENSE.Apache file in the root directory).
5 //
6 // Copyright (c) 2011 The LevelDB Authors. All rights reserved.
7 // Use of this source code is governed by a BSD-style license that can be
8 // found in the LICENSE file. See the AUTHORS file for names of contributors.
9
10 #include "db/builder.h"
11
12 #include <algorithm>
13 #include <deque>
14 #include <vector>
15
16 #include "db/compaction/compaction_iterator.h"
17 #include "db/dbformat.h"
18 #include "db/event_helpers.h"
19 #include "db/internal_stats.h"
20 #include "db/merge_helper.h"
21 #include "db/range_del_aggregator.h"
22 #include "db/table_cache.h"
23 #include "db/version_edit.h"
24 #include "file/filename.h"
25 #include "file/read_write_util.h"
26 #include "file/writable_file_writer.h"
27 #include "monitoring/iostats_context_imp.h"
28 #include "monitoring/thread_status_util.h"
29 #include "rocksdb/db.h"
30 #include "rocksdb/env.h"
31 #include "rocksdb/iterator.h"
32 #include "rocksdb/options.h"
33 #include "rocksdb/table.h"
34 #include "table/block_based/block_based_table_builder.h"
35 #include "table/format.h"
36 #include "table/internal_iterator.h"
37 #include "test_util/sync_point.h"
38 #include "util/stop_watch.h"
39
40 namespace ROCKSDB_NAMESPACE {
41
42 class TableFactory;
43
NewTableBuilder(const ImmutableCFOptions & ioptions,const MutableCFOptions & moptions,const InternalKeyComparator & internal_comparator,const std::vector<std::unique_ptr<IntTblPropCollectorFactory>> * int_tbl_prop_collector_factories,uint32_t column_family_id,const std::string & column_family_name,WritableFileWriter * file,const CompressionType compression_type,uint64_t sample_for_compression,const CompressionOptions & compression_opts,int level,const bool skip_filters,const uint64_t creation_time,const uint64_t oldest_key_time,const uint64_t target_file_size,const uint64_t file_creation_time)44 TableBuilder* NewTableBuilder(
45 const ImmutableCFOptions& ioptions, const MutableCFOptions& moptions,
46 const InternalKeyComparator& internal_comparator,
47 const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
48 int_tbl_prop_collector_factories,
49 uint32_t column_family_id, const std::string& column_family_name,
50 WritableFileWriter* file, const CompressionType compression_type,
51 uint64_t sample_for_compression, const CompressionOptions& compression_opts,
52 int level, const bool skip_filters, const uint64_t creation_time,
53 const uint64_t oldest_key_time, const uint64_t target_file_size,
54 const uint64_t file_creation_time) {
55 assert((column_family_id ==
56 TablePropertiesCollectorFactory::Context::kUnknownColumnFamily) ==
57 column_family_name.empty());
58 return ioptions.table_factory->NewTableBuilder(
59 TableBuilderOptions(ioptions, moptions, internal_comparator,
60 int_tbl_prop_collector_factories, compression_type,
61 sample_for_compression, compression_opts,
62 skip_filters, column_family_name, level,
63 creation_time, oldest_key_time, target_file_size,
64 file_creation_time),
65 column_family_id, file);
66 }
67
BuildTable(const std::string & dbname,Env * env,FileSystem * fs,const ImmutableCFOptions & ioptions,const MutableCFOptions & mutable_cf_options,const FileOptions & file_options,TableCache * table_cache,InternalIterator * iter,std::vector<std::unique_ptr<FragmentedRangeTombstoneIterator>> range_del_iters,FileMetaData * meta,const InternalKeyComparator & internal_comparator,const std::vector<std::unique_ptr<IntTblPropCollectorFactory>> * int_tbl_prop_collector_factories,uint32_t column_family_id,const std::string & column_family_name,std::vector<SequenceNumber> snapshots,SequenceNumber earliest_write_conflict_snapshot,SnapshotChecker * snapshot_checker,const CompressionType compression,uint64_t sample_for_compression,const CompressionOptions & compression_opts,bool paranoid_file_checks,InternalStats * internal_stats,TableFileCreationReason reason,IOStatus * io_status,EventLogger * event_logger,int job_id,const Env::IOPriority io_priority,TableProperties * table_properties,int level,const uint64_t creation_time,const uint64_t oldest_key_time,Env::WriteLifeTimeHint write_hint,const uint64_t file_creation_time)68 Status BuildTable(
69 const std::string& dbname, Env* env, FileSystem* fs,
70 const ImmutableCFOptions& ioptions,
71 const MutableCFOptions& mutable_cf_options, const FileOptions& file_options,
72 TableCache* table_cache, InternalIterator* iter,
73 std::vector<std::unique_ptr<FragmentedRangeTombstoneIterator>>
74 range_del_iters,
75 FileMetaData* meta, const InternalKeyComparator& internal_comparator,
76 const std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
77 int_tbl_prop_collector_factories,
78 uint32_t column_family_id, const std::string& column_family_name,
79 std::vector<SequenceNumber> snapshots,
80 SequenceNumber earliest_write_conflict_snapshot,
81 SnapshotChecker* snapshot_checker, const CompressionType compression,
82 uint64_t sample_for_compression, const CompressionOptions& compression_opts,
83 bool paranoid_file_checks, InternalStats* internal_stats,
84 TableFileCreationReason reason, IOStatus* io_status,
85 EventLogger* event_logger, int job_id, const Env::IOPriority io_priority,
86 TableProperties* table_properties, int level, const uint64_t creation_time,
87 const uint64_t oldest_key_time, Env::WriteLifeTimeHint write_hint,
88 const uint64_t file_creation_time) {
89 assert((column_family_id ==
90 TablePropertiesCollectorFactory::Context::kUnknownColumnFamily) ==
91 column_family_name.empty());
92 // Reports the IOStats for flush for every following bytes.
93 const size_t kReportFlushIOStatsEvery = 1048576;
94 Status s;
95 meta->fd.file_size = 0;
96 iter->SeekToFirst();
97 std::unique_ptr<CompactionRangeDelAggregator> range_del_agg(
98 new CompactionRangeDelAggregator(&internal_comparator, snapshots));
99 for (auto& range_del_iter : range_del_iters) {
100 range_del_agg->AddTombstones(std::move(range_del_iter));
101 }
102
103 std::string fname = TableFileName(ioptions.cf_paths, meta->fd.GetNumber(),
104 meta->fd.GetPathId());
105 #ifndef ROCKSDB_LITE
106 EventHelpers::NotifyTableFileCreationStarted(
107 ioptions.listeners, dbname, column_family_name, fname, job_id, reason);
108 #endif // !ROCKSDB_LITE
109 TableProperties tp;
110
111 if (iter->Valid() || !range_del_agg->IsEmpty()) {
112 TableBuilder* builder;
113 std::unique_ptr<WritableFileWriter> file_writer;
114 // Currently we only enable dictionary compression during compaction to the
115 // bottommost level.
116 CompressionOptions compression_opts_for_flush(compression_opts);
117 compression_opts_for_flush.max_dict_bytes = 0;
118 compression_opts_for_flush.zstd_max_train_bytes = 0;
119 {
120 std::unique_ptr<FSWritableFile> file;
121 #ifndef NDEBUG
122 bool use_direct_writes = file_options.use_direct_writes;
123 TEST_SYNC_POINT_CALLBACK("BuildTable:create_file", &use_direct_writes);
124 #endif // !NDEBUG
125 s = NewWritableFile(fs, fname, &file, file_options);
126 if (!s.ok()) {
127 EventHelpers::LogAndNotifyTableFileCreationFinished(
128 event_logger, ioptions.listeners, dbname, column_family_name, fname,
129 job_id, meta->fd, kInvalidBlobFileNumber, tp, reason, s);
130 return s;
131 }
132 file->SetIOPriority(io_priority);
133 file->SetWriteLifeTimeHint(write_hint);
134
135 file_writer.reset(new WritableFileWriter(
136 std::move(file), fname, file_options, env, ioptions.statistics,
137 ioptions.listeners, ioptions.file_checksum_gen_factory));
138
139 builder = NewTableBuilder(
140 ioptions, mutable_cf_options, internal_comparator,
141 int_tbl_prop_collector_factories, column_family_id,
142 column_family_name, file_writer.get(), compression,
143 sample_for_compression, compression_opts_for_flush, level,
144 false /* skip_filters */, creation_time, oldest_key_time,
145 0 /*target_file_size*/, file_creation_time);
146 }
147
148 MergeHelper merge(env, internal_comparator.user_comparator(),
149 ioptions.merge_operator, nullptr, ioptions.info_log,
150 true /* internal key corruption is not ok */,
151 snapshots.empty() ? 0 : snapshots.back(),
152 snapshot_checker);
153
154 CompactionIterator c_iter(
155 iter, internal_comparator.user_comparator(), &merge, kMaxSequenceNumber,
156 &snapshots, earliest_write_conflict_snapshot, snapshot_checker, env,
157 ShouldReportDetailedTime(env, ioptions.statistics),
158 true /* internal key corruption is not ok */, range_del_agg.get());
159 c_iter.SeekToFirst();
160 for (; c_iter.Valid(); c_iter.Next()) {
161 const Slice& key = c_iter.key();
162 const Slice& value = c_iter.value();
163 const ParsedInternalKey& ikey = c_iter.ikey();
164 builder->Add(key, value);
165 meta->UpdateBoundaries(key, value, ikey.sequence, ikey.type);
166
167 // TODO(noetzli): Update stats after flush, too.
168 if (io_priority == Env::IO_HIGH &&
169 IOSTATS(bytes_written) >= kReportFlushIOStatsEvery) {
170 ThreadStatusUtil::SetThreadOperationProperty(
171 ThreadStatus::FLUSH_BYTES_WRITTEN, IOSTATS(bytes_written));
172 }
173 }
174
175 auto range_del_it = range_del_agg->NewIterator();
176 for (range_del_it->SeekToFirst(); range_del_it->Valid();
177 range_del_it->Next()) {
178 auto tombstone = range_del_it->Tombstone();
179 auto kv = tombstone.Serialize();
180 builder->Add(kv.first.Encode(), kv.second);
181 meta->UpdateBoundariesForRange(kv.first, tombstone.SerializeEndKey(),
182 tombstone.seq_, internal_comparator);
183 }
184
185 // Finish and check for builder errors
186 tp = builder->GetTableProperties();
187 bool empty = builder->NumEntries() == 0 && tp.num_range_deletions == 0;
188 s = c_iter.status();
189 TEST_SYNC_POINT("BuildTable:BeforeFinishBuildTable");
190 if (!s.ok() || empty) {
191 builder->Abandon();
192 } else {
193 s = builder->Finish();
194 }
195 *io_status = builder->io_status();
196
197 if (s.ok() && !empty) {
198 uint64_t file_size = builder->FileSize();
199 meta->fd.file_size = file_size;
200 meta->marked_for_compaction = builder->NeedCompact();
201 assert(meta->fd.GetFileSize() > 0);
202 tp = builder->GetTableProperties(); // refresh now that builder is finished
203 if (table_properties) {
204 *table_properties = tp;
205 }
206 }
207 delete builder;
208
209 // Finish and check for file errors
210 if (s.ok() && !empty) {
211 StopWatch sw(env, ioptions.statistics, TABLE_SYNC_MICROS);
212 *io_status = file_writer->Sync(ioptions.use_fsync);
213 }
214 if (io_status->ok() && !empty) {
215 *io_status = file_writer->Close();
216 }
217 if (io_status->ok() && !empty) {
218 // Add the checksum information to file metadata.
219 meta->file_checksum = file_writer->GetFileChecksum();
220 meta->file_checksum_func_name = file_writer->GetFileChecksumFuncName();
221 }
222
223 if (!io_status->ok()) {
224 s = *io_status;
225 }
226
227 // TODO Also check the IO status when create the Iterator.
228
229 if (s.ok() && !empty) {
230 // Verify that the table is usable
231 // We set for_compaction to false and don't OptimizeForCompactionTableRead
232 // here because this is a special case after we finish the table building
233 // No matter whether use_direct_io_for_flush_and_compaction is true,
234 // we will regrad this verification as user reads since the goal is
235 // to cache it here for further user reads
236 std::unique_ptr<InternalIterator> it(table_cache->NewIterator(
237 ReadOptions(), file_options, internal_comparator, *meta,
238 nullptr /* range_del_agg */,
239 mutable_cf_options.prefix_extractor.get(), nullptr,
240 (internal_stats == nullptr) ? nullptr
241 : internal_stats->GetFileReadHist(0),
242 TableReaderCaller::kFlush, /*arena=*/nullptr,
243 /*skip_filter=*/false, level, /*smallest_compaction_key=*/nullptr,
244 /*largest_compaction_key*/ nullptr));
245 s = it->status();
246 if (s.ok() && paranoid_file_checks) {
247 for (it->SeekToFirst(); it->Valid(); it->Next()) {
248 }
249 s = it->status();
250 }
251 }
252 }
253
254 // Check for input iterator errors
255 if (!iter->status().ok()) {
256 s = iter->status();
257 }
258
259 if (!s.ok() || meta->fd.GetFileSize() == 0) {
260 fs->DeleteFile(fname, IOOptions(), nullptr);
261 }
262
263 if (meta->fd.GetFileSize() == 0) {
264 fname = "(nil)";
265 }
266 // Output to event logger and fire events.
267 EventHelpers::LogAndNotifyTableFileCreationFinished(
268 event_logger, ioptions.listeners, dbname, column_family_name, fname,
269 job_id, meta->fd, meta->oldest_blob_file_number, tp, reason, s);
270
271 return s;
272 }
273
274 } // namespace ROCKSDB_NAMESPACE
275