]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | // Copyright (c) 2011-present, Facebook, Inc. All rights reserved. |
2 | // This source code is licensed under the BSD-style license found in the | |
3 | // LICENSE file in the root directory of this source tree. An additional grant | |
4 | // of patent rights can be found in the PATENTS file in the same 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/compaction_picker.h" | |
11 | ||
12 | #ifndef __STDC_FORMAT_MACROS | |
13 | #define __STDC_FORMAT_MACROS | |
14 | #endif | |
15 | ||
16 | #include <inttypes.h> | |
17 | #include <limits> | |
18 | #include <queue> | |
19 | #include <string> | |
20 | #include <utility> | |
21 | #include "db/column_family.h" | |
22 | #include "monitoring/statistics.h" | |
23 | #include "util/filename.h" | |
24 | #include "util/log_buffer.h" | |
25 | #include "util/random.h" | |
26 | #include "util/string_util.h" | |
27 | #include "util/sync_point.h" | |
28 | ||
29 | namespace rocksdb { | |
30 | ||
31 | namespace { | |
32 | uint64_t TotalCompensatedFileSize(const std::vector<FileMetaData*>& files) { | |
33 | uint64_t sum = 0; | |
34 | for (size_t i = 0; i < files.size() && files[i]; i++) { | |
35 | sum += files[i]->compensated_file_size; | |
36 | } | |
37 | return sum; | |
38 | } | |
39 | } // anonymous namespace | |
40 | ||
41 | // Determine compression type, based on user options, level of the output | |
42 | // file and whether compression is disabled. | |
43 | // If enable_compression is false, then compression is always disabled no | |
44 | // matter what the values of the other two parameters are. | |
45 | // Otherwise, the compression type is determined based on options and level. | |
46 | CompressionType GetCompressionType(const ImmutableCFOptions& ioptions, | |
47 | const VersionStorageInfo* vstorage, | |
48 | const MutableCFOptions& mutable_cf_options, | |
49 | int level, int base_level, | |
50 | const bool enable_compression) { | |
51 | if (!enable_compression) { | |
52 | // disable compression | |
53 | return kNoCompression; | |
54 | } | |
55 | ||
56 | // If bottommost_compression is set and we are compacting to the | |
57 | // bottommost level then we should use it. | |
58 | if (ioptions.bottommost_compression != kDisableCompressionOption && | |
59 | level > base_level && level >= (vstorage->num_non_empty_levels() - 1)) { | |
60 | return ioptions.bottommost_compression; | |
61 | } | |
62 | // If the user has specified a different compression level for each level, | |
63 | // then pick the compression for that level. | |
64 | if (!ioptions.compression_per_level.empty()) { | |
65 | assert(level == 0 || level >= base_level); | |
66 | int idx = (level == 0) ? 0 : level - base_level + 1; | |
67 | ||
68 | const int n = static_cast<int>(ioptions.compression_per_level.size()) - 1; | |
69 | // It is possible for level_ to be -1; in that case, we use level | |
70 | // 0's compression. This occurs mostly in backwards compatibility | |
71 | // situations when the builder doesn't know what level the file | |
72 | // belongs to. Likewise, if level is beyond the end of the | |
73 | // specified compression levels, use the last value. | |
74 | return ioptions.compression_per_level[std::max(0, std::min(idx, n))]; | |
75 | } else { | |
76 | return mutable_cf_options.compression; | |
77 | } | |
78 | } | |
79 | ||
80 | CompactionPicker::CompactionPicker(const ImmutableCFOptions& ioptions, | |
81 | const InternalKeyComparator* icmp) | |
82 | : ioptions_(ioptions), icmp_(icmp) {} | |
83 | ||
84 | CompactionPicker::~CompactionPicker() {} | |
85 | ||
86 | // Delete this compaction from the list of running compactions. | |
87 | void CompactionPicker::ReleaseCompactionFiles(Compaction* c, Status status) { | |
88 | UnregisterCompaction(c); | |
89 | if (!status.ok()) { | |
90 | c->ResetNextCompactionIndex(); | |
91 | } | |
92 | } | |
93 | ||
94 | void CompactionPicker::GetRange(const CompactionInputFiles& inputs, | |
95 | InternalKey* smallest, | |
96 | InternalKey* largest) const { | |
97 | const int level = inputs.level; | |
98 | assert(!inputs.empty()); | |
99 | smallest->Clear(); | |
100 | largest->Clear(); | |
101 | ||
102 | if (level == 0) { | |
103 | for (size_t i = 0; i < inputs.size(); i++) { | |
104 | FileMetaData* f = inputs[i]; | |
105 | if (i == 0) { | |
106 | *smallest = f->smallest; | |
107 | *largest = f->largest; | |
108 | } else { | |
109 | if (icmp_->Compare(f->smallest, *smallest) < 0) { | |
110 | *smallest = f->smallest; | |
111 | } | |
112 | if (icmp_->Compare(f->largest, *largest) > 0) { | |
113 | *largest = f->largest; | |
114 | } | |
115 | } | |
116 | } | |
117 | } else { | |
118 | *smallest = inputs[0]->smallest; | |
119 | *largest = inputs[inputs.size() - 1]->largest; | |
120 | } | |
121 | } | |
122 | ||
123 | void CompactionPicker::GetRange(const CompactionInputFiles& inputs1, | |
124 | const CompactionInputFiles& inputs2, | |
125 | InternalKey* smallest, | |
126 | InternalKey* largest) const { | |
127 | assert(!inputs1.empty() || !inputs2.empty()); | |
128 | if (inputs1.empty()) { | |
129 | GetRange(inputs2, smallest, largest); | |
130 | } else if (inputs2.empty()) { | |
131 | GetRange(inputs1, smallest, largest); | |
132 | } else { | |
133 | InternalKey smallest1, smallest2, largest1, largest2; | |
134 | GetRange(inputs1, &smallest1, &largest1); | |
135 | GetRange(inputs2, &smallest2, &largest2); | |
136 | *smallest = | |
137 | icmp_->Compare(smallest1, smallest2) < 0 ? smallest1 : smallest2; | |
138 | *largest = icmp_->Compare(largest1, largest2) < 0 ? largest2 : largest1; | |
139 | } | |
140 | } | |
141 | ||
142 | void CompactionPicker::GetRange(const std::vector<CompactionInputFiles>& inputs, | |
143 | InternalKey* smallest, | |
144 | InternalKey* largest) const { | |
145 | InternalKey current_smallest; | |
146 | InternalKey current_largest; | |
147 | bool initialized = false; | |
148 | for (const auto& in : inputs) { | |
149 | if (in.empty()) { | |
150 | continue; | |
151 | } | |
152 | GetRange(in, ¤t_smallest, ¤t_largest); | |
153 | if (!initialized) { | |
154 | *smallest = current_smallest; | |
155 | *largest = current_largest; | |
156 | initialized = true; | |
157 | } else { | |
158 | if (icmp_->Compare(current_smallest, *smallest) < 0) { | |
159 | *smallest = current_smallest; | |
160 | } | |
161 | if (icmp_->Compare(current_largest, *largest) > 0) { | |
162 | *largest = current_largest; | |
163 | } | |
164 | } | |
165 | } | |
166 | assert(initialized); | |
167 | } | |
168 | ||
169 | bool CompactionPicker::ExpandInputsToCleanCut(const std::string& cf_name, | |
170 | VersionStorageInfo* vstorage, | |
171 | CompactionInputFiles* inputs) { | |
172 | // This isn't good compaction | |
173 | assert(!inputs->empty()); | |
174 | ||
175 | const int level = inputs->level; | |
176 | // GetOverlappingInputs will always do the right thing for level-0. | |
177 | // So we don't need to do any expansion if level == 0. | |
178 | if (level == 0) { | |
179 | return true; | |
180 | } | |
181 | ||
182 | InternalKey smallest, largest; | |
183 | ||
184 | // Keep expanding inputs until we are sure that there is a "clean cut" | |
185 | // boundary between the files in input and the surrounding files. | |
186 | // This will ensure that no parts of a key are lost during compaction. | |
187 | int hint_index = -1; | |
188 | size_t old_size; | |
189 | do { | |
190 | old_size = inputs->size(); | |
191 | GetRange(*inputs, &smallest, &largest); | |
192 | inputs->clear(); | |
193 | vstorage->GetOverlappingInputs(level, &smallest, &largest, &inputs->files, | |
194 | hint_index, &hint_index); | |
195 | } while (inputs->size() > old_size); | |
196 | ||
197 | // we started off with inputs non-empty and the previous loop only grew | |
198 | // inputs. thus, inputs should be non-empty here | |
199 | assert(!inputs->empty()); | |
200 | ||
201 | // If, after the expansion, there are files that are already under | |
202 | // compaction, then we must drop/cancel this compaction. | |
203 | if (AreFilesInCompaction(inputs->files)) { | |
204 | ROCKS_LOG_WARN( | |
205 | ioptions_.info_log, | |
206 | "[%s] ExpandWhileOverlapping() failure because some of the necessary" | |
207 | " compaction input files are currently being compacted.", | |
208 | cf_name.c_str()); | |
209 | return false; | |
210 | } | |
211 | return true; | |
212 | } | |
213 | ||
214 | bool CompactionPicker::RangeOverlapWithCompaction( | |
215 | const Slice& smallest_user_key, const Slice& largest_user_key, | |
216 | int level) const { | |
217 | const Comparator* ucmp = icmp_->user_comparator(); | |
218 | for (Compaction* c : compactions_in_progress_) { | |
219 | if (c->output_level() == level && | |
220 | ucmp->Compare(smallest_user_key, c->GetLargestUserKey()) <= 0 && | |
221 | ucmp->Compare(largest_user_key, c->GetSmallestUserKey()) >= 0) { | |
222 | // Overlap | |
223 | return true; | |
224 | } | |
225 | } | |
226 | // Did not overlap with any running compaction in level `level` | |
227 | return false; | |
228 | } | |
229 | ||
230 | bool CompactionPicker::FilesRangeOverlapWithCompaction( | |
231 | const std::vector<CompactionInputFiles>& inputs, int level) const { | |
232 | bool is_empty = true; | |
233 | for (auto& in : inputs) { | |
234 | if (!in.empty()) { | |
235 | is_empty = false; | |
236 | break; | |
237 | } | |
238 | } | |
239 | if (is_empty) { | |
240 | // No files in inputs | |
241 | return false; | |
242 | } | |
243 | ||
244 | InternalKey smallest, largest; | |
245 | GetRange(inputs, &smallest, &largest); | |
246 | return RangeOverlapWithCompaction(smallest.user_key(), largest.user_key(), | |
247 | level); | |
248 | } | |
249 | ||
250 | // Returns true if any one of specified files are being compacted | |
251 | bool CompactionPicker::AreFilesInCompaction( | |
252 | const std::vector<FileMetaData*>& files) { | |
253 | for (size_t i = 0; i < files.size(); i++) { | |
254 | if (files[i]->being_compacted) { | |
255 | return true; | |
256 | } | |
257 | } | |
258 | return false; | |
259 | } | |
260 | ||
261 | Compaction* CompactionPicker::CompactFiles( | |
262 | const CompactionOptions& compact_options, | |
263 | const std::vector<CompactionInputFiles>& input_files, int output_level, | |
264 | VersionStorageInfo* vstorage, const MutableCFOptions& mutable_cf_options, | |
265 | uint32_t output_path_id) { | |
266 | assert(input_files.size()); | |
267 | ||
268 | // TODO(rven ): we might be able to run concurrent level 0 compaction | |
269 | // if the key ranges of the two compactions do not overlap, but for now | |
270 | // we do not allow it. | |
271 | if ((input_files[0].level == 0) && !level0_compactions_in_progress_.empty()) { | |
272 | return nullptr; | |
273 | } | |
274 | // This compaction output could overlap with a running compaction | |
275 | if (FilesRangeOverlapWithCompaction(input_files, output_level)) { | |
276 | return nullptr; | |
277 | } | |
278 | auto c = | |
279 | new Compaction(vstorage, ioptions_, mutable_cf_options, input_files, | |
280 | output_level, compact_options.output_file_size_limit, | |
281 | mutable_cf_options.max_compaction_bytes, output_path_id, | |
282 | compact_options.compression, /* grandparents */ {}, true); | |
283 | ||
284 | // If it's level 0 compaction, make sure we don't execute any other level 0 | |
285 | // compactions in parallel | |
286 | RegisterCompaction(c); | |
287 | return c; | |
288 | } | |
289 | ||
290 | Status CompactionPicker::GetCompactionInputsFromFileNumbers( | |
291 | std::vector<CompactionInputFiles>* input_files, | |
292 | std::unordered_set<uint64_t>* input_set, const VersionStorageInfo* vstorage, | |
293 | const CompactionOptions& compact_options) const { | |
294 | if (input_set->size() == 0U) { | |
295 | return Status::InvalidArgument( | |
296 | "Compaction must include at least one file."); | |
297 | } | |
298 | assert(input_files); | |
299 | ||
300 | std::vector<CompactionInputFiles> matched_input_files; | |
301 | matched_input_files.resize(vstorage->num_levels()); | |
302 | int first_non_empty_level = -1; | |
303 | int last_non_empty_level = -1; | |
304 | // TODO(yhchiang): use a lazy-initialized mapping from | |
305 | // file_number to FileMetaData in Version. | |
306 | for (int level = 0; level < vstorage->num_levels(); ++level) { | |
307 | for (auto file : vstorage->LevelFiles(level)) { | |
308 | auto iter = input_set->find(file->fd.GetNumber()); | |
309 | if (iter != input_set->end()) { | |
310 | matched_input_files[level].files.push_back(file); | |
311 | input_set->erase(iter); | |
312 | last_non_empty_level = level; | |
313 | if (first_non_empty_level == -1) { | |
314 | first_non_empty_level = level; | |
315 | } | |
316 | } | |
317 | } | |
318 | } | |
319 | ||
320 | if (!input_set->empty()) { | |
321 | std::string message( | |
322 | "Cannot find matched SST files for the following file numbers:"); | |
323 | for (auto fn : *input_set) { | |
324 | message += " "; | |
325 | message += ToString(fn); | |
326 | } | |
327 | return Status::InvalidArgument(message); | |
328 | } | |
329 | ||
330 | for (int level = first_non_empty_level; level <= last_non_empty_level; | |
331 | ++level) { | |
332 | matched_input_files[level].level = level; | |
333 | input_files->emplace_back(std::move(matched_input_files[level])); | |
334 | } | |
335 | ||
336 | return Status::OK(); | |
337 | } | |
338 | ||
339 | // Returns true if any one of the parent files are being compacted | |
340 | bool CompactionPicker::IsRangeInCompaction(VersionStorageInfo* vstorage, | |
341 | const InternalKey* smallest, | |
342 | const InternalKey* largest, | |
343 | int level, int* level_index) { | |
344 | std::vector<FileMetaData*> inputs; | |
345 | assert(level < NumberLevels()); | |
346 | ||
347 | vstorage->GetOverlappingInputs(level, smallest, largest, &inputs, | |
348 | *level_index, level_index); | |
349 | return AreFilesInCompaction(inputs); | |
350 | } | |
351 | ||
352 | // Populates the set of inputs of all other levels that overlap with the | |
353 | // start level. | |
354 | // Now we assume all levels except start level and output level are empty. | |
355 | // Will also attempt to expand "start level" if that doesn't expand | |
356 | // "output level" or cause "level" to include a file for compaction that has an | |
357 | // overlapping user-key with another file. | |
358 | // REQUIRES: input_level and output_level are different | |
359 | // REQUIRES: inputs->empty() == false | |
360 | // Returns false if files on parent level are currently in compaction, which | |
361 | // means that we can't compact them | |
362 | bool CompactionPicker::SetupOtherInputs( | |
363 | const std::string& cf_name, const MutableCFOptions& mutable_cf_options, | |
364 | VersionStorageInfo* vstorage, CompactionInputFiles* inputs, | |
365 | CompactionInputFiles* output_level_inputs, int* parent_index, | |
366 | int base_index) { | |
367 | assert(!inputs->empty()); | |
368 | assert(output_level_inputs->empty()); | |
369 | const int input_level = inputs->level; | |
370 | const int output_level = output_level_inputs->level; | |
371 | assert(input_level != output_level); | |
372 | ||
373 | // For now, we only support merging two levels, start level and output level. | |
374 | // We need to assert other levels are empty. | |
375 | for (int l = input_level + 1; l < output_level; l++) { | |
376 | assert(vstorage->NumLevelFiles(l) == 0); | |
377 | } | |
378 | ||
379 | InternalKey smallest, largest; | |
380 | ||
381 | // Get the range one last time. | |
382 | GetRange(*inputs, &smallest, &largest); | |
383 | ||
384 | // Populate the set of next-level files (inputs_GetOutputLevelInputs()) to | |
385 | // include in compaction | |
386 | vstorage->GetOverlappingInputs(output_level, &smallest, &largest, | |
387 | &output_level_inputs->files, *parent_index, | |
388 | parent_index); | |
389 | if (AreFilesInCompaction(output_level_inputs->files)) { | |
390 | return false; | |
391 | } | |
392 | if (!output_level_inputs->empty()) { | |
393 | if (!ExpandInputsToCleanCut(cf_name, vstorage, output_level_inputs)) { | |
394 | return false; | |
395 | } | |
396 | } | |
397 | ||
398 | // See if we can further grow the number of inputs in "level" without | |
399 | // changing the number of "level+1" files we pick up. We also choose NOT | |
400 | // to expand if this would cause "level" to include some entries for some | |
401 | // user key, while excluding other entries for the same user key. This | |
402 | // can happen when one user key spans multiple files. | |
403 | if (!output_level_inputs->empty()) { | |
404 | const uint64_t limit = mutable_cf_options.max_compaction_bytes; | |
405 | const uint64_t output_level_inputs_size = | |
406 | TotalCompensatedFileSize(output_level_inputs->files); | |
407 | const uint64_t inputs_size = TotalCompensatedFileSize(inputs->files); | |
408 | bool expand_inputs = false; | |
409 | ||
410 | CompactionInputFiles expanded_inputs; | |
411 | expanded_inputs.level = input_level; | |
412 | // Get closed interval of output level | |
413 | InternalKey all_start, all_limit; | |
414 | GetRange(*inputs, *output_level_inputs, &all_start, &all_limit); | |
415 | bool try_overlapping_inputs = true; | |
416 | vstorage->GetOverlappingInputs(input_level, &all_start, &all_limit, | |
417 | &expanded_inputs.files, base_index, nullptr); | |
418 | uint64_t expanded_inputs_size = | |
419 | TotalCompensatedFileSize(expanded_inputs.files); | |
420 | if (!ExpandInputsToCleanCut(cf_name, vstorage, &expanded_inputs)) { | |
421 | try_overlapping_inputs = false; | |
422 | } | |
423 | if (try_overlapping_inputs && expanded_inputs.size() > inputs->size() && | |
424 | output_level_inputs_size + expanded_inputs_size < limit && | |
425 | !AreFilesInCompaction(expanded_inputs.files)) { | |
426 | InternalKey new_start, new_limit; | |
427 | GetRange(expanded_inputs, &new_start, &new_limit); | |
428 | CompactionInputFiles expanded_output_level_inputs; | |
429 | expanded_output_level_inputs.level = output_level; | |
430 | vstorage->GetOverlappingInputs(output_level, &new_start, &new_limit, | |
431 | &expanded_output_level_inputs.files, | |
432 | *parent_index, parent_index); | |
433 | assert(!expanded_output_level_inputs.empty()); | |
434 | if (!AreFilesInCompaction(expanded_output_level_inputs.files) && | |
435 | ExpandInputsToCleanCut(cf_name, vstorage, | |
436 | &expanded_output_level_inputs) && | |
437 | expanded_output_level_inputs.size() == output_level_inputs->size()) { | |
438 | expand_inputs = true; | |
439 | } | |
440 | } | |
441 | if (!expand_inputs) { | |
442 | vstorage->GetCleanInputsWithinInterval(input_level, &all_start, | |
443 | &all_limit, &expanded_inputs.files, | |
444 | base_index, nullptr); | |
445 | expanded_inputs_size = TotalCompensatedFileSize(expanded_inputs.files); | |
446 | if (expanded_inputs.size() > inputs->size() && | |
447 | output_level_inputs_size + expanded_inputs_size < limit && | |
448 | !AreFilesInCompaction(expanded_inputs.files)) { | |
449 | expand_inputs = true; | |
450 | } | |
451 | } | |
452 | if (expand_inputs) { | |
453 | ROCKS_LOG_INFO(ioptions_.info_log, | |
454 | "[%s] Expanding@%d %" ROCKSDB_PRIszt "+%" ROCKSDB_PRIszt | |
455 | "(%" PRIu64 "+%" PRIu64 " bytes) to %" ROCKSDB_PRIszt | |
456 | "+%" ROCKSDB_PRIszt " (%" PRIu64 "+%" PRIu64 "bytes)\n", | |
457 | cf_name.c_str(), input_level, inputs->size(), | |
458 | output_level_inputs->size(), inputs_size, | |
459 | output_level_inputs_size, expanded_inputs.size(), | |
460 | output_level_inputs->size(), expanded_inputs_size, | |
461 | output_level_inputs_size); | |
462 | inputs->files = expanded_inputs.files; | |
463 | } | |
464 | } | |
465 | return true; | |
466 | } | |
467 | ||
468 | void CompactionPicker::GetGrandparents( | |
469 | VersionStorageInfo* vstorage, const CompactionInputFiles& inputs, | |
470 | const CompactionInputFiles& output_level_inputs, | |
471 | std::vector<FileMetaData*>* grandparents) { | |
472 | InternalKey start, limit; | |
473 | GetRange(inputs, output_level_inputs, &start, &limit); | |
474 | // Compute the set of grandparent files that overlap this compaction | |
475 | // (parent == level+1; grandparent == level+2) | |
476 | if (output_level_inputs.level + 1 < NumberLevels()) { | |
477 | vstorage->GetOverlappingInputs(output_level_inputs.level + 1, &start, | |
478 | &limit, grandparents); | |
479 | } | |
480 | } | |
481 | ||
482 | Compaction* CompactionPicker::CompactRange( | |
483 | const std::string& cf_name, const MutableCFOptions& mutable_cf_options, | |
484 | VersionStorageInfo* vstorage, int input_level, int output_level, | |
485 | uint32_t output_path_id, const InternalKey* begin, const InternalKey* end, | |
486 | InternalKey** compaction_end, bool* manual_conflict) { | |
487 | // CompactionPickerFIFO has its own implementation of compact range | |
488 | assert(ioptions_.compaction_style != kCompactionStyleFIFO); | |
489 | ||
490 | if (input_level == ColumnFamilyData::kCompactAllLevels) { | |
491 | assert(ioptions_.compaction_style == kCompactionStyleUniversal); | |
492 | ||
493 | // Universal compaction with more than one level always compacts all the | |
494 | // files together to the last level. | |
495 | assert(vstorage->num_levels() > 1); | |
496 | // DBImpl::CompactRange() set output level to be the last level | |
497 | assert(output_level == vstorage->num_levels() - 1); | |
498 | // DBImpl::RunManualCompaction will make full range for universal compaction | |
499 | assert(begin == nullptr); | |
500 | assert(end == nullptr); | |
501 | *compaction_end = nullptr; | |
502 | ||
503 | int start_level = 0; | |
504 | for (; start_level < vstorage->num_levels() && | |
505 | vstorage->NumLevelFiles(start_level) == 0; | |
506 | start_level++) { | |
507 | } | |
508 | if (start_level == vstorage->num_levels()) { | |
509 | return nullptr; | |
510 | } | |
511 | ||
512 | if ((start_level == 0) && (!level0_compactions_in_progress_.empty())) { | |
513 | *manual_conflict = true; | |
514 | // Only one level 0 compaction allowed | |
515 | return nullptr; | |
516 | } | |
517 | ||
518 | std::vector<CompactionInputFiles> inputs(vstorage->num_levels() - | |
519 | start_level); | |
520 | for (int level = start_level; level < vstorage->num_levels(); level++) { | |
521 | inputs[level - start_level].level = level; | |
522 | auto& files = inputs[level - start_level].files; | |
523 | for (FileMetaData* f : vstorage->LevelFiles(level)) { | |
524 | files.push_back(f); | |
525 | } | |
526 | if (AreFilesInCompaction(files)) { | |
527 | *manual_conflict = true; | |
528 | return nullptr; | |
529 | } | |
530 | } | |
531 | ||
532 | // 2 non-exclusive manual compactions could run at the same time producing | |
533 | // overlaping outputs in the same level. | |
534 | if (FilesRangeOverlapWithCompaction(inputs, output_level)) { | |
535 | // This compaction output could potentially conflict with the output | |
536 | // of a currently running compaction, we cannot run it. | |
537 | *manual_conflict = true; | |
538 | return nullptr; | |
539 | } | |
540 | ||
541 | Compaction* c = new Compaction( | |
542 | vstorage, ioptions_, mutable_cf_options, std::move(inputs), | |
543 | output_level, mutable_cf_options.MaxFileSizeForLevel(output_level), | |
544 | /* max_compaction_bytes */ LLONG_MAX, output_path_id, | |
545 | GetCompressionType(ioptions_, vstorage, mutable_cf_options, | |
546 | output_level, 1), | |
547 | /* grandparents */ {}, /* is manual */ true); | |
548 | RegisterCompaction(c); | |
549 | return c; | |
550 | } | |
551 | ||
552 | CompactionInputFiles inputs; | |
553 | inputs.level = input_level; | |
554 | bool covering_the_whole_range = true; | |
555 | ||
556 | // All files are 'overlapping' in universal style compaction. | |
557 | // We have to compact the entire range in one shot. | |
558 | if (ioptions_.compaction_style == kCompactionStyleUniversal) { | |
559 | begin = nullptr; | |
560 | end = nullptr; | |
561 | } | |
562 | ||
563 | vstorage->GetOverlappingInputs(input_level, begin, end, &inputs.files); | |
564 | if (inputs.empty()) { | |
565 | return nullptr; | |
566 | } | |
567 | ||
568 | if ((input_level == 0) && (!level0_compactions_in_progress_.empty())) { | |
569 | // Only one level 0 compaction allowed | |
570 | TEST_SYNC_POINT("CompactionPicker::CompactRange:Conflict"); | |
571 | *manual_conflict = true; | |
572 | return nullptr; | |
573 | } | |
574 | ||
575 | // Avoid compacting too much in one shot in case the range is large. | |
576 | // But we cannot do this for level-0 since level-0 files can overlap | |
577 | // and we must not pick one file and drop another older file if the | |
578 | // two files overlap. | |
579 | if (input_level > 0) { | |
580 | const uint64_t limit = mutable_cf_options.max_compaction_bytes; | |
581 | uint64_t total = 0; | |
582 | for (size_t i = 0; i + 1 < inputs.size(); ++i) { | |
583 | uint64_t s = inputs[i]->compensated_file_size; | |
584 | total += s; | |
585 | if (total >= limit) { | |
586 | **compaction_end = inputs[i + 1]->smallest; | |
587 | covering_the_whole_range = false; | |
588 | inputs.files.resize(i + 1); | |
589 | break; | |
590 | } | |
591 | } | |
592 | } | |
593 | assert(output_path_id < static_cast<uint32_t>(ioptions_.db_paths.size())); | |
594 | ||
595 | if (ExpandInputsToCleanCut(cf_name, vstorage, &inputs) == false) { | |
596 | // manual compaction is now multi-threaded, so it can | |
597 | // happen that ExpandWhileOverlapping fails | |
598 | // we handle it higher in RunManualCompaction | |
599 | *manual_conflict = true; | |
600 | return nullptr; | |
601 | } | |
602 | ||
603 | if (covering_the_whole_range) { | |
604 | *compaction_end = nullptr; | |
605 | } | |
606 | ||
607 | CompactionInputFiles output_level_inputs; | |
608 | if (output_level == ColumnFamilyData::kCompactToBaseLevel) { | |
609 | assert(input_level == 0); | |
610 | output_level = vstorage->base_level(); | |
611 | assert(output_level > 0); | |
612 | } | |
613 | output_level_inputs.level = output_level; | |
614 | if (input_level != output_level) { | |
615 | int parent_index = -1; | |
616 | if (!SetupOtherInputs(cf_name, mutable_cf_options, vstorage, &inputs, | |
617 | &output_level_inputs, &parent_index, -1)) { | |
618 | // manual compaction is now multi-threaded, so it can | |
619 | // happen that SetupOtherInputs fails | |
620 | // we handle it higher in RunManualCompaction | |
621 | *manual_conflict = true; | |
622 | return nullptr; | |
623 | } | |
624 | } | |
625 | ||
626 | std::vector<CompactionInputFiles> compaction_inputs({inputs}); | |
627 | if (!output_level_inputs.empty()) { | |
628 | compaction_inputs.push_back(output_level_inputs); | |
629 | } | |
630 | for (size_t i = 0; i < compaction_inputs.size(); i++) { | |
631 | if (AreFilesInCompaction(compaction_inputs[i].files)) { | |
632 | *manual_conflict = true; | |
633 | return nullptr; | |
634 | } | |
635 | } | |
636 | ||
637 | // 2 non-exclusive manual compactions could run at the same time producing | |
638 | // overlaping outputs in the same level. | |
639 | if (FilesRangeOverlapWithCompaction(compaction_inputs, output_level)) { | |
640 | // This compaction output could potentially conflict with the output | |
641 | // of a currently running compaction, we cannot run it. | |
642 | *manual_conflict = true; | |
643 | return nullptr; | |
644 | } | |
645 | ||
646 | std::vector<FileMetaData*> grandparents; | |
647 | GetGrandparents(vstorage, inputs, output_level_inputs, &grandparents); | |
648 | Compaction* compaction = new Compaction( | |
649 | vstorage, ioptions_, mutable_cf_options, std::move(compaction_inputs), | |
650 | output_level, mutable_cf_options.MaxFileSizeForLevel(output_level), | |
651 | mutable_cf_options.max_compaction_bytes, output_path_id, | |
652 | GetCompressionType(ioptions_, vstorage, mutable_cf_options, output_level, | |
653 | vstorage->base_level()), | |
654 | std::move(grandparents), /* is manual compaction */ true); | |
655 | ||
656 | TEST_SYNC_POINT_CALLBACK("CompactionPicker::CompactRange:Return", compaction); | |
657 | RegisterCompaction(compaction); | |
658 | ||
659 | // Creating a compaction influences the compaction score because the score | |
660 | // takes running compactions into account (by skipping files that are already | |
661 | // being compacted). Since we just changed compaction score, we recalculate it | |
662 | // here | |
663 | vstorage->ComputeCompactionScore(ioptions_, mutable_cf_options); | |
664 | ||
665 | return compaction; | |
666 | } | |
667 | ||
668 | #ifndef ROCKSDB_LITE | |
669 | namespace { | |
670 | // Test whether two files have overlapping key-ranges. | |
671 | bool HaveOverlappingKeyRanges(const Comparator* c, const SstFileMetaData& a, | |
672 | const SstFileMetaData& b) { | |
673 | if (c->Compare(a.smallestkey, b.smallestkey) >= 0) { | |
674 | if (c->Compare(a.smallestkey, b.largestkey) <= 0) { | |
675 | // b.smallestkey <= a.smallestkey <= b.largestkey | |
676 | return true; | |
677 | } | |
678 | } else if (c->Compare(a.largestkey, b.smallestkey) >= 0) { | |
679 | // a.smallestkey < b.smallestkey <= a.largestkey | |
680 | return true; | |
681 | } | |
682 | if (c->Compare(a.largestkey, b.largestkey) <= 0) { | |
683 | if (c->Compare(a.largestkey, b.smallestkey) >= 0) { | |
684 | // b.smallestkey <= a.largestkey <= b.largestkey | |
685 | return true; | |
686 | } | |
687 | } else if (c->Compare(a.smallestkey, b.largestkey) <= 0) { | |
688 | // a.smallestkey <= b.largestkey < a.largestkey | |
689 | return true; | |
690 | } | |
691 | return false; | |
692 | } | |
693 | } // namespace | |
694 | ||
695 | Status CompactionPicker::SanitizeCompactionInputFilesForAllLevels( | |
696 | std::unordered_set<uint64_t>* input_files, | |
697 | const ColumnFamilyMetaData& cf_meta, const int output_level) const { | |
698 | auto& levels = cf_meta.levels; | |
699 | auto comparator = icmp_->user_comparator(); | |
700 | ||
701 | // TODO(yhchiang): If there is any input files of L1 or up and there | |
702 | // is at least one L0 files. All L0 files older than the L0 file needs | |
703 | // to be included. Otherwise, it is a false conditoin | |
704 | ||
705 | // TODO(yhchiang): add is_adjustable to CompactionOptions | |
706 | ||
707 | // the smallest and largest key of the current compaction input | |
708 | std::string smallestkey; | |
709 | std::string largestkey; | |
710 | // a flag for initializing smallest and largest key | |
711 | bool is_first = false; | |
712 | const int kNotFound = -1; | |
713 | ||
714 | // For each level, it does the following things: | |
715 | // 1. Find the first and the last compaction input files | |
716 | // in the current level. | |
717 | // 2. Include all files between the first and the last | |
718 | // compaction input files. | |
719 | // 3. Update the compaction key-range. | |
720 | // 4. For all remaining levels, include files that have | |
721 | // overlapping key-range with the compaction key-range. | |
722 | for (int l = 0; l <= output_level; ++l) { | |
723 | auto& current_files = levels[l].files; | |
724 | int first_included = static_cast<int>(current_files.size()); | |
725 | int last_included = kNotFound; | |
726 | ||
727 | // identify the first and the last compaction input files | |
728 | // in the current level. | |
729 | for (size_t f = 0; f < current_files.size(); ++f) { | |
730 | if (input_files->find(TableFileNameToNumber(current_files[f].name)) != | |
731 | input_files->end()) { | |
732 | first_included = std::min(first_included, static_cast<int>(f)); | |
733 | last_included = std::max(last_included, static_cast<int>(f)); | |
734 | if (is_first == false) { | |
735 | smallestkey = current_files[f].smallestkey; | |
736 | largestkey = current_files[f].largestkey; | |
737 | is_first = true; | |
738 | } | |
739 | } | |
740 | } | |
741 | if (last_included == kNotFound) { | |
742 | continue; | |
743 | } | |
744 | ||
745 | if (l != 0) { | |
746 | // expend the compaction input of the current level if it | |
747 | // has overlapping key-range with other non-compaction input | |
748 | // files in the same level. | |
749 | while (first_included > 0) { | |
750 | if (comparator->Compare(current_files[first_included - 1].largestkey, | |
751 | current_files[first_included].smallestkey) < | |
752 | 0) { | |
753 | break; | |
754 | } | |
755 | first_included--; | |
756 | } | |
757 | ||
758 | while (last_included < static_cast<int>(current_files.size()) - 1) { | |
759 | if (comparator->Compare(current_files[last_included + 1].smallestkey, | |
760 | current_files[last_included].largestkey) > 0) { | |
761 | break; | |
762 | } | |
763 | last_included++; | |
764 | } | |
765 | } | |
766 | ||
767 | // include all files between the first and the last compaction input files. | |
768 | for (int f = first_included; f <= last_included; ++f) { | |
769 | if (current_files[f].being_compacted) { | |
770 | return Status::Aborted("Necessary compaction input file " + | |
771 | current_files[f].name + | |
772 | " is currently being compacted."); | |
773 | } | |
774 | input_files->insert(TableFileNameToNumber(current_files[f].name)); | |
775 | } | |
776 | ||
777 | // update smallest and largest key | |
778 | if (l == 0) { | |
779 | for (int f = first_included; f <= last_included; ++f) { | |
780 | if (comparator->Compare(smallestkey, current_files[f].smallestkey) > | |
781 | 0) { | |
782 | smallestkey = current_files[f].smallestkey; | |
783 | } | |
784 | if (comparator->Compare(largestkey, current_files[f].largestkey) < 0) { | |
785 | largestkey = current_files[f].largestkey; | |
786 | } | |
787 | } | |
788 | } else { | |
789 | if (comparator->Compare(smallestkey, | |
790 | current_files[first_included].smallestkey) > 0) { | |
791 | smallestkey = current_files[first_included].smallestkey; | |
792 | } | |
793 | if (comparator->Compare(largestkey, | |
794 | current_files[last_included].largestkey) < 0) { | |
795 | largestkey = current_files[last_included].largestkey; | |
796 | } | |
797 | } | |
798 | ||
799 | SstFileMetaData aggregated_file_meta; | |
800 | aggregated_file_meta.smallestkey = smallestkey; | |
801 | aggregated_file_meta.largestkey = largestkey; | |
802 | ||
803 | // For all lower levels, include all overlapping files. | |
804 | // We need to add overlapping files from the current level too because even | |
805 | // if there no input_files in level l, we would still need to add files | |
806 | // which overlap with the range containing the input_files in levels 0 to l | |
807 | // Level 0 doesn't need to be handled this way because files are sorted by | |
808 | // time and not by key | |
809 | for (int m = std::max(l, 1); m <= output_level; ++m) { | |
810 | for (auto& next_lv_file : levels[m].files) { | |
811 | if (HaveOverlappingKeyRanges(comparator, aggregated_file_meta, | |
812 | next_lv_file)) { | |
813 | if (next_lv_file.being_compacted) { | |
814 | return Status::Aborted( | |
815 | "File " + next_lv_file.name + | |
816 | " that has overlapping key range with one of the compaction " | |
817 | " input file is currently being compacted."); | |
818 | } | |
819 | input_files->insert(TableFileNameToNumber(next_lv_file.name)); | |
820 | } | |
821 | } | |
822 | } | |
823 | } | |
824 | return Status::OK(); | |
825 | } | |
826 | ||
827 | Status CompactionPicker::SanitizeCompactionInputFiles( | |
828 | std::unordered_set<uint64_t>* input_files, | |
829 | const ColumnFamilyMetaData& cf_meta, const int output_level) const { | |
830 | assert(static_cast<int>(cf_meta.levels.size()) - 1 == | |
831 | cf_meta.levels[cf_meta.levels.size() - 1].level); | |
832 | if (output_level >= static_cast<int>(cf_meta.levels.size())) { | |
833 | return Status::InvalidArgument( | |
834 | "Output level for column family " + cf_meta.name + | |
835 | " must between [0, " + | |
836 | ToString(cf_meta.levels[cf_meta.levels.size() - 1].level) + "]."); | |
837 | } | |
838 | ||
839 | if (output_level > MaxOutputLevel()) { | |
840 | return Status::InvalidArgument( | |
841 | "Exceed the maximum output level defined by " | |
842 | "the current compaction algorithm --- " + | |
843 | ToString(MaxOutputLevel())); | |
844 | } | |
845 | ||
846 | if (output_level < 0) { | |
847 | return Status::InvalidArgument("Output level cannot be negative."); | |
848 | } | |
849 | ||
850 | if (input_files->size() == 0) { | |
851 | return Status::InvalidArgument( | |
852 | "A compaction must contain at least one file."); | |
853 | } | |
854 | ||
855 | Status s = SanitizeCompactionInputFilesForAllLevels(input_files, cf_meta, | |
856 | output_level); | |
857 | ||
858 | if (!s.ok()) { | |
859 | return s; | |
860 | } | |
861 | ||
862 | // for all input files, check whether the file number matches | |
863 | // any currently-existing files. | |
864 | for (auto file_num : *input_files) { | |
865 | bool found = false; | |
866 | for (auto level_meta : cf_meta.levels) { | |
867 | for (auto file_meta : level_meta.files) { | |
868 | if (file_num == TableFileNameToNumber(file_meta.name)) { | |
869 | if (file_meta.being_compacted) { | |
870 | return Status::Aborted("Specified compaction input file " + | |
871 | MakeTableFileName("", file_num) + | |
872 | " is already being compacted."); | |
873 | } | |
874 | found = true; | |
875 | break; | |
876 | } | |
877 | } | |
878 | if (found) { | |
879 | break; | |
880 | } | |
881 | } | |
882 | if (!found) { | |
883 | return Status::InvalidArgument( | |
884 | "Specified compaction input file " + MakeTableFileName("", file_num) + | |
885 | " does not exist in column family " + cf_meta.name + "."); | |
886 | } | |
887 | } | |
888 | ||
889 | return Status::OK(); | |
890 | } | |
891 | #endif // !ROCKSDB_LITE | |
892 | ||
893 | void CompactionPicker::RegisterCompaction(Compaction* c) { | |
894 | if (c == nullptr) { | |
895 | return; | |
896 | } | |
897 | assert(ioptions_.compaction_style != kCompactionStyleLevel || | |
898 | c->output_level() == 0 || | |
899 | !FilesRangeOverlapWithCompaction(*c->inputs(), c->output_level())); | |
900 | if (c->start_level() == 0 || | |
901 | ioptions_.compaction_style == kCompactionStyleUniversal) { | |
902 | level0_compactions_in_progress_.insert(c); | |
903 | } | |
904 | compactions_in_progress_.insert(c); | |
905 | } | |
906 | ||
907 | void CompactionPicker::UnregisterCompaction(Compaction* c) { | |
908 | if (c == nullptr) { | |
909 | return; | |
910 | } | |
911 | if (c->start_level() == 0 || | |
912 | ioptions_.compaction_style == kCompactionStyleUniversal) { | |
913 | level0_compactions_in_progress_.erase(c); | |
914 | } | |
915 | compactions_in_progress_.erase(c); | |
916 | } | |
917 | ||
918 | bool LevelCompactionPicker::NeedsCompaction( | |
919 | const VersionStorageInfo* vstorage) const { | |
920 | if (!vstorage->FilesMarkedForCompaction().empty()) { | |
921 | return true; | |
922 | } | |
923 | for (int i = 0; i <= vstorage->MaxInputLevel(); i++) { | |
924 | if (vstorage->CompactionScore(i) >= 1) { | |
925 | return true; | |
926 | } | |
927 | } | |
928 | return false; | |
929 | } | |
930 | ||
931 | namespace { | |
932 | // A class to build a leveled compaction step-by-step. | |
933 | class LevelCompactionBuilder { | |
934 | public: | |
935 | LevelCompactionBuilder(const std::string& cf_name, | |
936 | VersionStorageInfo* vstorage, | |
937 | CompactionPicker* compaction_picker, | |
938 | LogBuffer* log_buffer, | |
939 | const MutableCFOptions& mutable_cf_options, | |
940 | const ImmutableCFOptions& ioptions) | |
941 | : cf_name_(cf_name), | |
942 | vstorage_(vstorage), | |
943 | compaction_picker_(compaction_picker), | |
944 | log_buffer_(log_buffer), | |
945 | mutable_cf_options_(mutable_cf_options), | |
946 | ioptions_(ioptions) {} | |
947 | ||
948 | // Pick and return a compaction. | |
949 | Compaction* PickCompaction(); | |
950 | ||
951 | // Pick the initial files to compact to the next level. (or together | |
952 | // in Intra-L0 compactions) | |
953 | void SetupInitialFiles(); | |
954 | ||
955 | // If the initial files are from L0 level, pick other L0 | |
956 | // files if needed. | |
957 | bool SetupOtherL0FilesIfNeeded(); | |
958 | ||
959 | // Based on initial files, setup other files need to be compacted | |
960 | // in this compaction, accordingly. | |
961 | bool SetupOtherInputsIfNeeded(); | |
962 | ||
963 | Compaction* GetCompaction(); | |
964 | ||
965 | // For the specfied level, pick a file that we want to compact. | |
966 | // Returns false if there is no file to compact. | |
967 | // If it returns true, inputs->files.size() will be exactly one. | |
968 | // If level is 0 and there is already a compaction on that level, this | |
969 | // function will return false. | |
970 | bool PickFileToCompact(); | |
971 | ||
972 | // For L0->L0, picks the longest span of files that aren't currently | |
973 | // undergoing compaction for which work-per-deleted-file decreases. The span | |
974 | // always starts from the newest L0 file. | |
975 | // | |
976 | // Intra-L0 compaction is independent of all other files, so it can be | |
977 | // performed even when L0->base_level compactions are blocked. | |
978 | // | |
979 | // Returns true if `inputs` is populated with a span of files to be compacted; | |
980 | // otherwise, returns false. | |
981 | bool PickIntraL0Compaction(); | |
982 | ||
983 | // If there is any file marked for compaction, put put it into inputs. | |
984 | void PickFilesMarkedForCompaction(); | |
985 | ||
986 | const std::string& cf_name_; | |
987 | VersionStorageInfo* vstorage_; | |
988 | CompactionPicker* compaction_picker_; | |
989 | LogBuffer* log_buffer_; | |
990 | int start_level_ = -1; | |
991 | int output_level_ = -1; | |
992 | int parent_index_ = -1; | |
993 | int base_index_ = -1; | |
994 | double start_level_score_ = 0; | |
995 | bool is_manual_ = false; | |
996 | CompactionInputFiles start_level_inputs_; | |
997 | std::vector<CompactionInputFiles> compaction_inputs_; | |
998 | CompactionInputFiles output_level_inputs_; | |
999 | std::vector<FileMetaData*> grandparents_; | |
1000 | CompactionReason compaction_reason_ = CompactionReason::kUnknown; | |
1001 | ||
1002 | const MutableCFOptions& mutable_cf_options_; | |
1003 | const ImmutableCFOptions& ioptions_; | |
1004 | // Pick a path ID to place a newly generated file, with its level | |
1005 | static uint32_t GetPathId(const ImmutableCFOptions& ioptions, | |
1006 | const MutableCFOptions& mutable_cf_options, | |
1007 | int level); | |
1008 | ||
1009 | static const int kMinFilesForIntraL0Compaction = 4; | |
1010 | }; | |
1011 | ||
1012 | void LevelCompactionBuilder::PickFilesMarkedForCompaction() { | |
1013 | if (vstorage_->FilesMarkedForCompaction().empty()) { | |
1014 | return; | |
1015 | } | |
1016 | ||
1017 | auto continuation = [&](std::pair<int, FileMetaData*> level_file) { | |
1018 | // If it's being compacted it has nothing to do here. | |
1019 | // If this assert() fails that means that some function marked some | |
1020 | // files as being_compacted, but didn't call ComputeCompactionScore() | |
1021 | assert(!level_file.second->being_compacted); | |
1022 | start_level_ = level_file.first; | |
1023 | output_level_ = | |
1024 | (start_level_ == 0) ? vstorage_->base_level() : start_level_ + 1; | |
1025 | ||
1026 | if (start_level_ == 0 && | |
1027 | !compaction_picker_->level0_compactions_in_progress()->empty()) { | |
1028 | return false; | |
1029 | } | |
1030 | ||
1031 | start_level_inputs_.files = {level_file.second}; | |
1032 | start_level_inputs_.level = start_level_; | |
1033 | return compaction_picker_->ExpandInputsToCleanCut(cf_name_, vstorage_, | |
1034 | &start_level_inputs_); | |
1035 | }; | |
1036 | ||
1037 | // take a chance on a random file first | |
1038 | Random64 rnd(/* seed */ reinterpret_cast<uint64_t>(vstorage_)); | |
1039 | size_t random_file_index = static_cast<size_t>(rnd.Uniform( | |
1040 | static_cast<uint64_t>(vstorage_->FilesMarkedForCompaction().size()))); | |
1041 | ||
1042 | if (continuation(vstorage_->FilesMarkedForCompaction()[random_file_index])) { | |
1043 | // found the compaction! | |
1044 | return; | |
1045 | } | |
1046 | ||
1047 | for (auto& level_file : vstorage_->FilesMarkedForCompaction()) { | |
1048 | if (continuation(level_file)) { | |
1049 | // found the compaction! | |
1050 | return; | |
1051 | } | |
1052 | } | |
1053 | start_level_inputs_.files.clear(); | |
1054 | } | |
1055 | ||
1056 | void LevelCompactionBuilder::SetupInitialFiles() { | |
1057 | // Find the compactions by size on all levels. | |
1058 | bool skipped_l0_to_base = false; | |
1059 | for (int i = 0; i < compaction_picker_->NumberLevels() - 1; i++) { | |
1060 | start_level_score_ = vstorage_->CompactionScore(i); | |
1061 | start_level_ = vstorage_->CompactionScoreLevel(i); | |
1062 | assert(i == 0 || start_level_score_ <= vstorage_->CompactionScore(i - 1)); | |
1063 | if (start_level_score_ >= 1) { | |
1064 | if (skipped_l0_to_base && start_level_ == vstorage_->base_level()) { | |
1065 | // If L0->base_level compaction is pending, don't schedule further | |
1066 | // compaction from base level. Otherwise L0->base_level compaction | |
1067 | // may starve. | |
1068 | continue; | |
1069 | } | |
1070 | output_level_ = | |
1071 | (start_level_ == 0) ? vstorage_->base_level() : start_level_ + 1; | |
1072 | if (PickFileToCompact() && | |
1073 | compaction_picker_->ExpandInputsToCleanCut(cf_name_, vstorage_, | |
1074 | &start_level_inputs_) && | |
1075 | !compaction_picker_->FilesRangeOverlapWithCompaction( | |
1076 | {start_level_inputs_}, output_level_)) { | |
1077 | // found the compaction! | |
1078 | if (start_level_ == 0) { | |
1079 | // L0 score = `num L0 files` / `level0_file_num_compaction_trigger` | |
1080 | compaction_reason_ = CompactionReason::kLevelL0FilesNum; | |
1081 | } else { | |
1082 | // L1+ score = `Level files size` / `MaxBytesForLevel` | |
1083 | compaction_reason_ = CompactionReason::kLevelMaxLevelSize; | |
1084 | } | |
1085 | break; | |
1086 | } else { | |
1087 | // didn't find the compaction, clear the inputs | |
1088 | start_level_inputs_.clear(); | |
1089 | if (start_level_ == 0) { | |
1090 | skipped_l0_to_base = true; | |
1091 | // L0->base_level may be blocked due to ongoing L0->base_level | |
1092 | // compactions. It may also be blocked by an ongoing compaction from | |
1093 | // base_level downwards. | |
1094 | // | |
1095 | // In these cases, to reduce L0 file count and thus reduce likelihood | |
1096 | // of write stalls, we can attempt compacting a span of files within | |
1097 | // L0. | |
1098 | if (PickIntraL0Compaction()) { | |
1099 | output_level_ = 0; | |
1100 | compaction_reason_ = CompactionReason::kLevelL0FilesNum; | |
1101 | break; | |
1102 | } | |
1103 | } | |
1104 | } | |
1105 | } | |
1106 | } | |
1107 | ||
1108 | // if we didn't find a compaction, check if there are any files marked for | |
1109 | // compaction | |
1110 | if (start_level_inputs_.empty()) { | |
1111 | is_manual_ = true; | |
1112 | parent_index_ = base_index_ = -1; | |
1113 | PickFilesMarkedForCompaction(); | |
1114 | if (!start_level_inputs_.empty()) { | |
1115 | compaction_reason_ = CompactionReason::kFilesMarkedForCompaction; | |
1116 | } | |
1117 | } | |
1118 | } | |
1119 | ||
1120 | bool LevelCompactionBuilder::SetupOtherL0FilesIfNeeded() { | |
1121 | if (start_level_ == 0 && output_level_ != 0) { | |
1122 | // Two level 0 compaction won't run at the same time, so don't need to worry | |
1123 | // about files on level 0 being compacted. | |
1124 | assert(compaction_picker_->level0_compactions_in_progress()->empty()); | |
1125 | InternalKey smallest, largest; | |
1126 | compaction_picker_->GetRange(start_level_inputs_, &smallest, &largest); | |
1127 | // Note that the next call will discard the file we placed in | |
1128 | // c->inputs_[0] earlier and replace it with an overlapping set | |
1129 | // which will include the picked file. | |
1130 | start_level_inputs_.files.clear(); | |
1131 | vstorage_->GetOverlappingInputs(0, &smallest, &largest, | |
1132 | &start_level_inputs_.files); | |
1133 | ||
1134 | // If we include more L0 files in the same compaction run it can | |
1135 | // cause the 'smallest' and 'largest' key to get extended to a | |
1136 | // larger range. So, re-invoke GetRange to get the new key range | |
1137 | compaction_picker_->GetRange(start_level_inputs_, &smallest, &largest); | |
1138 | if (compaction_picker_->IsRangeInCompaction( | |
1139 | vstorage_, &smallest, &largest, output_level_, &parent_index_)) { | |
1140 | return false; | |
1141 | } | |
1142 | } | |
1143 | assert(!start_level_inputs_.files.empty()); | |
1144 | ||
1145 | return true; | |
1146 | } | |
1147 | ||
1148 | bool LevelCompactionBuilder::SetupOtherInputsIfNeeded() { | |
1149 | // Setup input files from output level. For output to L0, we only compact | |
1150 | // spans of files that do not interact with any pending compactions, so don't | |
1151 | // need to consider other levels. | |
1152 | if (output_level_ != 0) { | |
1153 | output_level_inputs_.level = output_level_; | |
1154 | if (!compaction_picker_->SetupOtherInputs( | |
1155 | cf_name_, mutable_cf_options_, vstorage_, &start_level_inputs_, | |
1156 | &output_level_inputs_, &parent_index_, base_index_)) { | |
1157 | return false; | |
1158 | } | |
1159 | ||
1160 | compaction_inputs_.push_back(start_level_inputs_); | |
1161 | if (!output_level_inputs_.empty()) { | |
1162 | compaction_inputs_.push_back(output_level_inputs_); | |
1163 | } | |
1164 | ||
1165 | // In some edge cases we could pick a compaction that will be compacting | |
1166 | // a key range that overlap with another running compaction, and both | |
1167 | // of them have the same output level. This could happen if | |
1168 | // (1) we are running a non-exclusive manual compaction | |
1169 | // (2) AddFile ingest a new file into the LSM tree | |
1170 | // We need to disallow this from happening. | |
1171 | if (compaction_picker_->FilesRangeOverlapWithCompaction(compaction_inputs_, | |
1172 | output_level_)) { | |
1173 | // This compaction output could potentially conflict with the output | |
1174 | // of a currently running compaction, we cannot run it. | |
1175 | return false; | |
1176 | } | |
1177 | compaction_picker_->GetGrandparents(vstorage_, start_level_inputs_, | |
1178 | output_level_inputs_, &grandparents_); | |
1179 | } else { | |
1180 | compaction_inputs_.push_back(start_level_inputs_); | |
1181 | } | |
1182 | return true; | |
1183 | } | |
1184 | ||
1185 | Compaction* LevelCompactionBuilder::PickCompaction() { | |
1186 | // Pick up the first file to start compaction. It may have been extended | |
1187 | // to a clean cut. | |
1188 | SetupInitialFiles(); | |
1189 | if (start_level_inputs_.empty()) { | |
1190 | return nullptr; | |
1191 | } | |
1192 | assert(start_level_ >= 0 && output_level_ >= 0); | |
1193 | ||
1194 | // If it is a L0 -> base level compaction, we need to set up other L0 | |
1195 | // files if needed. | |
1196 | if (!SetupOtherL0FilesIfNeeded()) { | |
1197 | return nullptr; | |
1198 | } | |
1199 | ||
1200 | // Pick files in the output level and expand more files in the start level | |
1201 | // if needed. | |
1202 | if (!SetupOtherInputsIfNeeded()) { | |
1203 | return nullptr; | |
1204 | } | |
1205 | ||
1206 | // Form a compaction object containing the files we picked. | |
1207 | Compaction* c = GetCompaction(); | |
1208 | ||
1209 | TEST_SYNC_POINT_CALLBACK("LevelCompactionPicker::PickCompaction:Return", c); | |
1210 | ||
1211 | return c; | |
1212 | } | |
1213 | ||
1214 | Compaction* LevelCompactionBuilder::GetCompaction() { | |
1215 | auto c = new Compaction( | |
1216 | vstorage_, ioptions_, mutable_cf_options_, std::move(compaction_inputs_), | |
1217 | output_level_, mutable_cf_options_.MaxFileSizeForLevel(output_level_), | |
1218 | mutable_cf_options_.max_compaction_bytes, | |
1219 | GetPathId(ioptions_, mutable_cf_options_, output_level_), | |
1220 | GetCompressionType(ioptions_, vstorage_, mutable_cf_options_, | |
1221 | output_level_, vstorage_->base_level()), | |
1222 | std::move(grandparents_), is_manual_, start_level_score_, | |
1223 | false /* deletion_compaction */, compaction_reason_); | |
1224 | ||
1225 | // If it's level 0 compaction, make sure we don't execute any other level 0 | |
1226 | // compactions in parallel | |
1227 | compaction_picker_->RegisterCompaction(c); | |
1228 | ||
1229 | // Creating a compaction influences the compaction score because the score | |
1230 | // takes running compactions into account (by skipping files that are already | |
1231 | // being compacted). Since we just changed compaction score, we recalculate it | |
1232 | // here | |
1233 | vstorage_->ComputeCompactionScore(ioptions_, mutable_cf_options_); | |
1234 | return c; | |
1235 | } | |
1236 | ||
1237 | /* | |
1238 | * Find the optimal path to place a file | |
1239 | * Given a level, finds the path where levels up to it will fit in levels | |
1240 | * up to and including this path | |
1241 | */ | |
1242 | uint32_t LevelCompactionBuilder::GetPathId( | |
1243 | const ImmutableCFOptions& ioptions, | |
1244 | const MutableCFOptions& mutable_cf_options, int level) { | |
1245 | uint32_t p = 0; | |
1246 | assert(!ioptions.db_paths.empty()); | |
1247 | ||
1248 | // size remaining in the most recent path | |
1249 | uint64_t current_path_size = ioptions.db_paths[0].target_size; | |
1250 | ||
1251 | uint64_t level_size; | |
1252 | int cur_level = 0; | |
1253 | ||
1254 | level_size = mutable_cf_options.max_bytes_for_level_base; | |
1255 | ||
1256 | // Last path is the fallback | |
1257 | while (p < ioptions.db_paths.size() - 1) { | |
1258 | if (level_size <= current_path_size) { | |
1259 | if (cur_level == level) { | |
1260 | // Does desired level fit in this path? | |
1261 | return p; | |
1262 | } else { | |
1263 | current_path_size -= level_size; | |
1264 | level_size = static_cast<uint64_t>( | |
1265 | level_size * mutable_cf_options.max_bytes_for_level_multiplier); | |
1266 | cur_level++; | |
1267 | continue; | |
1268 | } | |
1269 | } | |
1270 | p++; | |
1271 | current_path_size = ioptions.db_paths[p].target_size; | |
1272 | } | |
1273 | return p; | |
1274 | } | |
1275 | ||
1276 | bool LevelCompactionBuilder::PickFileToCompact() { | |
1277 | // level 0 files are overlapping. So we cannot pick more | |
1278 | // than one concurrent compactions at this level. This | |
1279 | // could be made better by looking at key-ranges that are | |
1280 | // being compacted at level 0. | |
1281 | if (start_level_ == 0 && | |
1282 | !compaction_picker_->level0_compactions_in_progress()->empty()) { | |
1283 | TEST_SYNC_POINT("LevelCompactionPicker::PickCompactionBySize:0"); | |
1284 | return false; | |
1285 | } | |
1286 | ||
1287 | start_level_inputs_.clear(); | |
1288 | ||
1289 | assert(start_level_ >= 0); | |
1290 | ||
1291 | // Pick the largest file in this level that is not already | |
1292 | // being compacted | |
1293 | const std::vector<int>& file_size = | |
1294 | vstorage_->FilesByCompactionPri(start_level_); | |
1295 | const std::vector<FileMetaData*>& level_files = | |
1296 | vstorage_->LevelFiles(start_level_); | |
1297 | ||
1298 | // record the first file that is not yet compacted | |
1299 | int nextIndex = -1; | |
1300 | ||
1301 | for (unsigned int i = vstorage_->NextCompactionIndex(start_level_); | |
1302 | i < file_size.size(); i++) { | |
1303 | int index = file_size[i]; | |
1304 | auto* f = level_files[index]; | |
1305 | ||
1306 | // do not pick a file to compact if it is being compacted | |
1307 | // from n-1 level. | |
1308 | if (f->being_compacted) { | |
1309 | continue; | |
1310 | } | |
1311 | ||
1312 | // remember the startIndex for the next call to PickCompaction | |
1313 | if (nextIndex == -1) { | |
1314 | nextIndex = i; | |
1315 | } | |
1316 | ||
1317 | // Do not pick this file if its parents at level+1 are being compacted. | |
1318 | // Maybe we can avoid redoing this work in SetupOtherInputs | |
1319 | parent_index_ = -1; | |
1320 | if (compaction_picker_->IsRangeInCompaction(vstorage_, &f->smallest, | |
1321 | &f->largest, output_level_, | |
1322 | &parent_index_)) { | |
1323 | continue; | |
1324 | } | |
1325 | start_level_inputs_.files.push_back(f); | |
1326 | start_level_inputs_.level = start_level_; | |
1327 | base_index_ = index; | |
1328 | break; | |
1329 | } | |
1330 | ||
1331 | // store where to start the iteration in the next call to PickCompaction | |
1332 | vstorage_->SetNextCompactionIndex(start_level_, nextIndex); | |
1333 | ||
1334 | return start_level_inputs_.size() > 0; | |
1335 | } | |
1336 | ||
1337 | bool LevelCompactionBuilder::PickIntraL0Compaction() { | |
1338 | start_level_inputs_.clear(); | |
1339 | const std::vector<FileMetaData*>& level_files = | |
1340 | vstorage_->LevelFiles(0 /* level */); | |
1341 | if (level_files.size() < | |
1342 | static_cast<size_t>( | |
1343 | mutable_cf_options_.level0_file_num_compaction_trigger + 2) || | |
1344 | level_files[0]->being_compacted) { | |
1345 | // If L0 isn't accumulating much files beyond the regular trigger, don't | |
1346 | // resort to L0->L0 compaction yet. | |
1347 | return false; | |
1348 | } | |
1349 | ||
1350 | size_t compact_bytes = level_files[0]->fd.file_size; | |
1351 | size_t compact_bytes_per_del_file = port::kMaxSizet; | |
1352 | // compaction range will be [0, span_len). | |
1353 | size_t span_len; | |
1354 | // pull in files until the amount of compaction work per deleted file begins | |
1355 | // increasing. | |
1356 | for (span_len = 1; span_len < level_files.size(); ++span_len) { | |
1357 | compact_bytes += level_files[span_len]->fd.file_size; | |
1358 | size_t new_compact_bytes_per_del_file = compact_bytes / span_len; | |
1359 | if (level_files[span_len]->being_compacted || | |
1360 | new_compact_bytes_per_del_file > compact_bytes_per_del_file) { | |
1361 | break; | |
1362 | } | |
1363 | compact_bytes_per_del_file = new_compact_bytes_per_del_file; | |
1364 | } | |
1365 | ||
1366 | if (span_len >= kMinFilesForIntraL0Compaction) { | |
1367 | start_level_inputs_.level = 0; | |
1368 | for (size_t i = 0; i < span_len; ++i) { | |
1369 | start_level_inputs_.files.push_back(level_files[i]); | |
1370 | } | |
1371 | return true; | |
1372 | } | |
1373 | return false; | |
1374 | } | |
1375 | } // namespace | |
1376 | ||
1377 | Compaction* LevelCompactionPicker::PickCompaction( | |
1378 | const std::string& cf_name, const MutableCFOptions& mutable_cf_options, | |
1379 | VersionStorageInfo* vstorage, LogBuffer* log_buffer) { | |
1380 | LevelCompactionBuilder builder(cf_name, vstorage, this, log_buffer, | |
1381 | mutable_cf_options, ioptions_); | |
1382 | return builder.PickCompaction(); | |
1383 | } | |
1384 | ||
1385 | #ifndef ROCKSDB_LITE | |
1386 | bool FIFOCompactionPicker::NeedsCompaction( | |
1387 | const VersionStorageInfo* vstorage) const { | |
1388 | const int kLevel0 = 0; | |
1389 | return vstorage->CompactionScore(kLevel0) >= 1; | |
1390 | } | |
1391 | ||
1392 | Compaction* FIFOCompactionPicker::PickCompaction( | |
1393 | const std::string& cf_name, const MutableCFOptions& mutable_cf_options, | |
1394 | VersionStorageInfo* vstorage, LogBuffer* log_buffer) { | |
1395 | assert(vstorage->num_levels() == 1); | |
1396 | const int kLevel0 = 0; | |
1397 | const std::vector<FileMetaData*>& level_files = vstorage->LevelFiles(kLevel0); | |
1398 | uint64_t total_size = 0; | |
1399 | for (const auto& file : level_files) { | |
1400 | total_size += file->fd.file_size; | |
1401 | } | |
1402 | ||
1403 | if (total_size <= ioptions_.compaction_options_fifo.max_table_files_size || | |
1404 | level_files.size() == 0) { | |
1405 | // total size not exceeded | |
1406 | ROCKS_LOG_BUFFER(log_buffer, | |
1407 | "[%s] FIFO compaction: nothing to do. Total size %" PRIu64 | |
1408 | ", max size %" PRIu64 "\n", | |
1409 | cf_name.c_str(), total_size, | |
1410 | ioptions_.compaction_options_fifo.max_table_files_size); | |
1411 | return nullptr; | |
1412 | } | |
1413 | ||
1414 | if (!level0_compactions_in_progress_.empty()) { | |
1415 | ROCKS_LOG_BUFFER( | |
1416 | log_buffer, | |
1417 | "[%s] FIFO compaction: Already executing compaction. No need " | |
1418 | "to run parallel compactions since compactions are very fast", | |
1419 | cf_name.c_str()); | |
1420 | return nullptr; | |
1421 | } | |
1422 | ||
1423 | std::vector<CompactionInputFiles> inputs; | |
1424 | inputs.emplace_back(); | |
1425 | inputs[0].level = 0; | |
1426 | // delete old files (FIFO) | |
1427 | for (auto ritr = level_files.rbegin(); ritr != level_files.rend(); ++ritr) { | |
1428 | auto f = *ritr; | |
1429 | total_size -= f->compensated_file_size; | |
1430 | inputs[0].files.push_back(f); | |
1431 | char tmp_fsize[16]; | |
1432 | AppendHumanBytes(f->fd.GetFileSize(), tmp_fsize, sizeof(tmp_fsize)); | |
1433 | ROCKS_LOG_BUFFER(log_buffer, "[%s] FIFO compaction: picking file %" PRIu64 | |
1434 | " with size %s for deletion", | |
1435 | cf_name.c_str(), f->fd.GetNumber(), tmp_fsize); | |
1436 | if (total_size <= ioptions_.compaction_options_fifo.max_table_files_size) { | |
1437 | break; | |
1438 | } | |
1439 | } | |
1440 | Compaction* c = new Compaction( | |
1441 | vstorage, ioptions_, mutable_cf_options, std::move(inputs), 0, 0, 0, 0, | |
1442 | kNoCompression, {}, /* is manual */ false, vstorage->CompactionScore(0), | |
1443 | /* is deletion compaction */ true, CompactionReason::kFIFOMaxSize); | |
1444 | RegisterCompaction(c); | |
1445 | return c; | |
1446 | } | |
1447 | ||
1448 | Compaction* FIFOCompactionPicker::CompactRange( | |
1449 | const std::string& cf_name, const MutableCFOptions& mutable_cf_options, | |
1450 | VersionStorageInfo* vstorage, int input_level, int output_level, | |
1451 | uint32_t output_path_id, const InternalKey* begin, const InternalKey* end, | |
1452 | InternalKey** compaction_end, bool* manual_conflict) { | |
1453 | assert(input_level == 0); | |
1454 | assert(output_level == 0); | |
1455 | *compaction_end = nullptr; | |
1456 | LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL, ioptions_.info_log); | |
1457 | Compaction* c = | |
1458 | PickCompaction(cf_name, mutable_cf_options, vstorage, &log_buffer); | |
1459 | log_buffer.FlushBufferToLog(); | |
1460 | return c; | |
1461 | } | |
1462 | ||
1463 | #endif // !ROCKSDB_LITE | |
1464 | ||
1465 | } // namespace rocksdb |