/root/doris/be/src/olap/tablet_meta.h
Line | Count | Source (jump to first uncovered line) |
1 | | // Licensed to the Apache Software Foundation (ASF) under one |
2 | | // or more contributor license agreements. See the NOTICE file |
3 | | // distributed with this work for additional information |
4 | | // regarding copyright ownership. The ASF licenses this file |
5 | | // to you under the Apache License, Version 2.0 (the |
6 | | // "License"); you may not use this file except in compliance |
7 | | // with the License. You may obtain a copy of the License at |
8 | | // |
9 | | // http://www.apache.org/licenses/LICENSE-2.0 |
10 | | // |
11 | | // Unless required by applicable law or agreed to in writing, |
12 | | // software distributed under the License is distributed on an |
13 | | // "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY |
14 | | // KIND, either express or implied. See the License for the |
15 | | // specific language governing permissions and limitations |
16 | | // under the License. |
17 | | |
18 | | #pragma once |
19 | | |
20 | | #include <gen_cpp/AgentService_types.h> |
21 | | #include <gen_cpp/olap_file.pb.h> |
22 | | #include <stdint.h> |
23 | | |
24 | | #include <atomic> |
25 | | #include <cstddef> |
26 | | #include <limits> |
27 | | #include <map> |
28 | | #include <memory> |
29 | | #include <mutex> |
30 | | #include <optional> |
31 | | #include <ostream> |
32 | | #include <roaring/roaring.hh> |
33 | | #include <shared_mutex> |
34 | | #include <string> |
35 | | #include <tuple> |
36 | | #include <unordered_map> |
37 | | #include <utility> |
38 | | #include <vector> |
39 | | |
40 | | #include "common/logging.h" |
41 | | #include "common/status.h" |
42 | | #include "gutil/stringprintf.h" |
43 | | #include "io/fs/file_system.h" |
44 | | #include "olap/binlog_config.h" |
45 | | #include "olap/lru_cache.h" |
46 | | #include "olap/metadata_adder.h" |
47 | | #include "olap/olap_common.h" |
48 | | #include "olap/rowset/rowset_meta.h" |
49 | | #include "olap/tablet_schema.h" |
50 | | #include "runtime/memory/lru_cache_policy.h" |
51 | | #include "util/uid_util.h" |
52 | | |
53 | | namespace json2pb { |
54 | | struct Pb2JsonOptions; |
55 | | } // namespace json2pb |
56 | | |
57 | | namespace doris { |
58 | | class TColumn; |
59 | | |
60 | | // Lifecycle states that a Tablet can be in. Legal state transitions for a |
61 | | // Tablet object: |
62 | | // |
63 | | // NOTREADY -> RUNNING -> TOMBSTONED -> STOPPED -> SHUTDOWN |
64 | | // | | | ^^^ |
65 | | // | | +----------++| |
66 | | // | +------------------------+| |
67 | | // +-------------------------------------+ |
68 | | |
69 | | enum TabletState { |
70 | | // Tablet is under alter table, rollup, clone |
71 | | TABLET_NOTREADY, |
72 | | |
73 | | TABLET_RUNNING, |
74 | | |
75 | | // Tablet integrity has been violated, such as missing versions. |
76 | | // In this state, tablet will not accept any incoming request. |
77 | | // Report this state to FE, scheduling BE to drop tablet. |
78 | | TABLET_TOMBSTONED, |
79 | | |
80 | | // Tablet is shutting down, files in disk still remained. |
81 | | TABLET_STOPPED, |
82 | | |
83 | | // Files have been removed, tablet has been shutdown completely. |
84 | | TABLET_SHUTDOWN |
85 | | }; |
86 | | |
87 | | class DataDir; |
88 | | class TabletMeta; |
89 | | class DeleteBitmap; |
90 | | class TBinlogConfig; |
91 | | |
92 | | // Class encapsulates meta of tablet. |
93 | | // The concurrency control is handled in Tablet Class, not in this class. |
94 | | class TabletMeta : public MetadataAdder<TabletMeta> { |
95 | | public: |
96 | | static TabletMetaSharedPtr create( |
97 | | const TCreateTabletReq& request, const TabletUid& tablet_uid, uint64_t shard_id, |
98 | | uint32_t next_unique_id, |
99 | | const std::unordered_map<uint32_t, uint32_t>& col_ordinal_to_unique_id); |
100 | | |
101 | | TabletMeta(); |
102 | | TabletMeta(int64_t table_id, int64_t partition_id, int64_t tablet_id, int64_t replica_id, |
103 | | int32_t schema_hash, uint64_t shard_id, const TTabletSchema& tablet_schema, |
104 | | uint32_t next_unique_id, |
105 | | const std::unordered_map<uint32_t, uint32_t>& col_ordinal_to_unique_id, |
106 | | TabletUid tablet_uid, TTabletType::type tabletType, |
107 | | TCompressionType::type compression_type, int64_t storage_policy_id = 0, |
108 | | bool enable_unique_key_merge_on_write = false, |
109 | | std::optional<TBinlogConfig> binlog_config = {}, |
110 | | std::string compaction_policy = "size_based", |
111 | | int64_t time_series_compaction_goal_size_mbytes = 1024, |
112 | | int64_t time_series_compaction_file_count_threshold = 2000, |
113 | | int64_t time_series_compaction_time_threshold_seconds = 3600, |
114 | | int64_t time_series_compaction_empty_rowsets_threshold = 5, |
115 | | int64_t time_series_compaction_level_threshold = 1, |
116 | | TInvertedIndexFileStorageFormat::type inverted_index_file_storage_format = |
117 | | TInvertedIndexFileStorageFormat::V2); |
118 | | // If need add a filed in TableMeta, filed init copy in copy construct function |
119 | | TabletMeta(const TabletMeta& tablet_meta); |
120 | | TabletMeta(TabletMeta&& tablet_meta) = delete; |
121 | | |
122 | | // UT |
123 | | #ifdef BE_TEST |
124 | 3 | TabletMeta(TabletSchemaSPtr tablet_schema) : _schema(tablet_schema) {} |
125 | | #endif |
126 | | |
127 | | // Function create_from_file is used to be compatible with previous tablet_meta. |
128 | | // Previous tablet_meta is a physical file in tablet dir, which is not stored in rocksdb. |
129 | | Status create_from_file(const std::string& file_path); |
130 | | Status save(const std::string& file_path); |
131 | | Status save_as_json(const string& file_path); |
132 | | static Status save(const std::string& file_path, const TabletMetaPB& tablet_meta_pb); |
133 | | static std::string construct_header_file_path(const std::string& schema_hash_path, |
134 | | int64_t tablet_id); |
135 | | Status save_meta(DataDir* data_dir); |
136 | | |
137 | | void serialize(std::string* meta_binary); |
138 | | Status deserialize(std::string_view meta_binary); |
139 | | void init_from_pb(const TabletMetaPB& tablet_meta_pb); |
140 | | |
141 | | void to_meta_pb(TabletMetaPB* tablet_meta_pb); |
142 | | void to_json(std::string* json_string, json2pb::Pb2JsonOptions& options); |
143 | | // Don't use. |
144 | | // TODO: memory size of TabletSchema cannot be accurately tracked. |
145 | | // In some places, temporarily use num_columns() as TabletSchema size. |
146 | | int64_t mem_size() const; |
147 | 139 | size_t tablet_columns_num() const { return _schema->num_columns(); } |
148 | | |
149 | 0 | TabletTypePB tablet_type() const { return _tablet_type; } |
150 | | TabletUid tablet_uid() const; |
151 | 28 | void set_tablet_uid(TabletUid uid) { _tablet_uid = uid; } |
152 | | int64_t table_id() const; |
153 | | int64_t index_id() const; |
154 | | int64_t partition_id() const; |
155 | | int64_t tablet_id() const; |
156 | | int64_t replica_id() const; |
157 | 0 | void set_replica_id(int64_t replica_id) { _replica_id = replica_id; } |
158 | | int32_t schema_hash() const; |
159 | | int16_t shard_id() const; |
160 | | void set_shard_id(int32_t shard_id); |
161 | | int64_t creation_time() const; |
162 | | void set_creation_time(int64_t creation_time); |
163 | | int64_t cumulative_layer_point() const; |
164 | | void set_cumulative_layer_point(int64_t new_point); |
165 | | |
166 | | size_t num_rows() const; |
167 | | // Disk space occupied by tablet, contain local and remote. |
168 | | size_t tablet_footprint() const; |
169 | | // Local disk space occupied by tablet. |
170 | | size_t tablet_local_size() const; |
171 | | // Remote disk space occupied by tablet. |
172 | | size_t tablet_remote_size() const; |
173 | | size_t version_count() const; |
174 | | size_t stale_version_count() const; |
175 | | size_t version_count_cross_with_range(const Version& range) const; |
176 | | Version max_version() const; |
177 | | |
178 | | TabletState tablet_state() const; |
179 | | void set_tablet_state(TabletState state); |
180 | | |
181 | | bool in_restore_mode() const; |
182 | | void set_in_restore_mode(bool in_restore_mode); |
183 | | |
184 | | const TabletSchemaSPtr& tablet_schema() const; |
185 | | |
186 | | TabletSchema* mutable_tablet_schema(); |
187 | | |
188 | | const std::vector<RowsetMetaSharedPtr>& all_rs_metas() const; |
189 | | std::vector<RowsetMetaSharedPtr>& all_mutable_rs_metas(); |
190 | | Status add_rs_meta(const RowsetMetaSharedPtr& rs_meta); |
191 | | void delete_rs_meta_by_version(const Version& version, |
192 | | std::vector<RowsetMetaSharedPtr>* deleted_rs_metas); |
193 | | // If same_version is true, the rowset in "to_delete" will not be added |
194 | | // to _stale_rs_meta, but to be deleted from rs_meta directly. |
195 | | void modify_rs_metas(const std::vector<RowsetMetaSharedPtr>& to_add, |
196 | | const std::vector<RowsetMetaSharedPtr>& to_delete, |
197 | | bool same_version = false); |
198 | | void revise_rs_metas(std::vector<RowsetMetaSharedPtr>&& rs_metas); |
199 | | void revise_delete_bitmap_unlocked(const DeleteBitmap& delete_bitmap); |
200 | | |
201 | | const std::vector<RowsetMetaSharedPtr>& all_stale_rs_metas() const; |
202 | | RowsetMetaSharedPtr acquire_rs_meta_by_version(const Version& version) const; |
203 | | void delete_stale_rs_meta_by_version(const Version& version); |
204 | | RowsetMetaSharedPtr acquire_stale_rs_meta_by_version(const Version& version) const; |
205 | | |
206 | | Status set_partition_id(int64_t partition_id); |
207 | | |
208 | 121 | RowsetTypePB preferred_rowset_type() const { return _preferred_rowset_type; } |
209 | | |
210 | 52 | void set_preferred_rowset_type(RowsetTypePB preferred_rowset_type) { |
211 | 52 | _preferred_rowset_type = preferred_rowset_type; |
212 | 52 | } |
213 | | |
214 | | // used for after tablet cloned to clear stale rowset |
215 | 0 | void clear_stale_rowset() { _stale_rs_metas.clear(); } |
216 | | |
217 | 0 | void clear_rowsets() { _rs_metas.clear(); } |
218 | | |
219 | | // MUST hold EXCLUSIVE `_meta_lock` in belonged Tablet |
220 | | // `to_add` MUST NOT have overlapped version with `_rs_metas` in tablet meta. |
221 | | void add_rowsets_unchecked(const std::vector<RowsetSharedPtr>& to_add); |
222 | | |
223 | | bool all_beta() const; |
224 | | |
225 | 22 | int64_t storage_policy_id() const { return _storage_policy_id; } |
226 | | |
227 | 4 | void set_storage_policy_id(int64_t id) { |
228 | 4 | VLOG_NOTICE << "set tablet_id : " << _table_id << " storage policy from " |
229 | 0 | << _storage_policy_id << " to " << id; |
230 | 4 | _storage_policy_id = id; |
231 | 4 | } |
232 | | |
233 | 14 | UniqueId cooldown_meta_id() const { return _cooldown_meta_id; } |
234 | 5 | void set_cooldown_meta_id(UniqueId uid) { _cooldown_meta_id = uid; } |
235 | | |
236 | | static void init_column_from_tcolumn(uint32_t unique_id, const TColumn& tcolumn, |
237 | | ColumnPB* column); |
238 | | |
239 | 18 | DeleteBitmap& delete_bitmap() { return *_delete_bitmap; } |
240 | | |
241 | 607 | bool enable_unique_key_merge_on_write() const { return _enable_unique_key_merge_on_write; } |
242 | | |
243 | | // TODO(Drogon): thread safety |
244 | 0 | const BinlogConfig& binlog_config() const { return _binlog_config; } |
245 | 0 | void set_binlog_config(BinlogConfig binlog_config) { |
246 | 0 | _binlog_config = std::move(binlog_config); |
247 | 0 | } |
248 | | |
249 | 12 | void set_compaction_policy(std::string compaction_policy) { |
250 | 12 | _compaction_policy = compaction_policy; |
251 | 12 | } |
252 | 479 | std::string compaction_policy() const { return _compaction_policy; } |
253 | 12 | void set_time_series_compaction_goal_size_mbytes(int64_t goal_size_mbytes) { |
254 | 12 | _time_series_compaction_goal_size_mbytes = goal_size_mbytes; |
255 | 12 | } |
256 | 186 | int64_t time_series_compaction_goal_size_mbytes() const { |
257 | 186 | return _time_series_compaction_goal_size_mbytes; |
258 | 186 | } |
259 | 12 | void set_time_series_compaction_file_count_threshold(int64_t file_count_threshold) { |
260 | 12 | _time_series_compaction_file_count_threshold = file_count_threshold; |
261 | 12 | } |
262 | 184 | int64_t time_series_compaction_file_count_threshold() const { |
263 | 184 | return _time_series_compaction_file_count_threshold; |
264 | 184 | } |
265 | 12 | void set_time_series_compaction_time_threshold_seconds(int64_t time_threshold) { |
266 | 12 | _time_series_compaction_time_threshold_seconds = time_threshold; |
267 | 12 | } |
268 | 182 | int64_t time_series_compaction_time_threshold_seconds() const { |
269 | 182 | return _time_series_compaction_time_threshold_seconds; |
270 | 182 | } |
271 | 0 | void set_time_series_compaction_empty_rowsets_threshold(int64_t empty_rowsets_threshold) { |
272 | 0 | _time_series_compaction_empty_rowsets_threshold = empty_rowsets_threshold; |
273 | 0 | } |
274 | 183 | int64_t time_series_compaction_empty_rowsets_threshold() const { |
275 | 183 | return _time_series_compaction_empty_rowsets_threshold; |
276 | 183 | } |
277 | 0 | void set_time_series_compaction_level_threshold(int64_t level_threshold) { |
278 | 0 | _time_series_compaction_level_threshold = level_threshold; |
279 | 0 | } |
280 | 205 | int64_t time_series_compaction_level_threshold() const { |
281 | 205 | return _time_series_compaction_level_threshold; |
282 | 205 | } |
283 | | |
284 | 141 | int64_t ttl_seconds() const { |
285 | 141 | std::shared_lock rlock(_meta_lock); |
286 | 141 | return _ttl_seconds; |
287 | 141 | } |
288 | | |
289 | 0 | void set_ttl_seconds(int64_t ttl_seconds) { |
290 | 0 | std::lock_guard wlock(_meta_lock); |
291 | 0 | _ttl_seconds = ttl_seconds; |
292 | 0 | } |
293 | | |
294 | 25 | int64_t avg_rs_meta_serialize_size() const { return _avg_rs_meta_serialize_size; } |
295 | | |
296 | | private: |
297 | | Status _save_meta(DataDir* data_dir); |
298 | | |
299 | | // _del_predicates is ignored to compare. |
300 | | friend bool operator==(const TabletMeta& a, const TabletMeta& b); |
301 | | friend bool operator!=(const TabletMeta& a, const TabletMeta& b); |
302 | | |
303 | | private: |
304 | | int64_t _table_id = 0; |
305 | | int64_t _index_id = 0; |
306 | | int64_t _partition_id = 0; |
307 | | int64_t _tablet_id = 0; |
308 | | int64_t _replica_id = 0; |
309 | | int32_t _schema_hash = 0; |
310 | | int32_t _shard_id = 0; |
311 | | int64_t _creation_time = 0; |
312 | | int64_t _cumulative_layer_point = 0; |
313 | | TabletUid _tablet_uid; |
314 | | TabletTypePB _tablet_type = TabletTypePB::TABLET_TYPE_DISK; |
315 | | |
316 | | TabletState _tablet_state = TABLET_NOTREADY; |
317 | | // the reference of _schema may use in tablet, so here need keep |
318 | | // the lifetime of tablemeta and _schema is same with tablet |
319 | | TabletSchemaSPtr _schema; |
320 | | |
321 | | std::vector<RowsetMetaSharedPtr> _rs_metas; |
322 | | // This variable _stale_rs_metas is used to record these rowsets‘ meta which are be compacted. |
323 | | // These stale rowsets meta are been removed when rowsets' pathVersion is expired, |
324 | | // this policy is judged and computed by TimestampedVersionTracker. |
325 | | std::vector<RowsetMetaSharedPtr> _stale_rs_metas; |
326 | | bool _in_restore_mode = false; |
327 | | RowsetTypePB _preferred_rowset_type = BETA_ROWSET; |
328 | | |
329 | | // meta for cooldown |
330 | | int64_t _storage_policy_id = 0; // <= 0 means no storage policy |
331 | | UniqueId _cooldown_meta_id; |
332 | | |
333 | | // For unique key data model, the feature Merge-on-Write will leverage a primary |
334 | | // key index and a delete-bitmap to mark duplicate keys as deleted in load stage, |
335 | | // which can avoid the merging cost in read stage, and accelerate the aggregation |
336 | | // query performance significantly. |
337 | | bool _enable_unique_key_merge_on_write = false; |
338 | | std::shared_ptr<DeleteBitmap> _delete_bitmap; |
339 | | |
340 | | // binlog config |
341 | | BinlogConfig _binlog_config {}; |
342 | | |
343 | | // meta for compaction |
344 | | std::string _compaction_policy; |
345 | | int64_t _time_series_compaction_goal_size_mbytes = 0; |
346 | | int64_t _time_series_compaction_file_count_threshold = 0; |
347 | | int64_t _time_series_compaction_time_threshold_seconds = 0; |
348 | | int64_t _time_series_compaction_empty_rowsets_threshold = 0; |
349 | | int64_t _time_series_compaction_level_threshold = 0; |
350 | | |
351 | | int64_t _avg_rs_meta_serialize_size = 0; |
352 | | |
353 | | // cloud |
354 | | int64_t _ttl_seconds = 0; |
355 | | |
356 | | mutable std::shared_mutex _meta_lock; |
357 | | }; |
358 | | |
359 | | /** |
360 | | * Wraps multiple bitmaps for recording rows (row id) that are deleted or |
361 | | * overwritten. For now, it's only used when unique key merge-on-write property |
362 | | * enabled. |
363 | | * |
364 | | * RowsetId and SegmentId are for locating segment, Version here is a single |
365 | | * uint32_t means that at which "version" of the load causes the delete or |
366 | | * overwrite. |
367 | | * |
368 | | * The start and end version of a load is the same, it's ok and straightforward |
369 | | * to use a single uint32_t. |
370 | | * |
371 | | * e.g. |
372 | | * There is a key "key1" in rowset id 1, version [1,1], segment id 1, row id 1. |
373 | | * A new load also contains "key1", the rowset id 2, version [2,2], segment id 1 |
374 | | * the delete bitmap will be `{1,1,2} -> 1`, which means the "row id 1" in |
375 | | * "rowset id 1, segment id 1" is deleted/overitten by some loads at "version 2" |
376 | | */ |
377 | | class DeleteBitmap { |
378 | | public: |
379 | | mutable std::shared_mutex lock; |
380 | | mutable std::shared_mutex stale_delete_bitmap_lock; |
381 | | using SegmentId = uint32_t; |
382 | | using Version = uint64_t; |
383 | | using BitmapKey = std::tuple<RowsetId, SegmentId, Version>; |
384 | | std::map<BitmapKey, roaring::Roaring> delete_bitmap; // Ordered map |
385 | | constexpr static inline uint32_t INVALID_SEGMENT_ID = std::numeric_limits<uint32_t>::max() - 1; |
386 | | constexpr static inline uint32_t ROWSET_SENTINEL_MARK = |
387 | | std::numeric_limits<uint32_t>::max() - 1; |
388 | | |
389 | | // When a delete bitmap is merged into tablet's delete bitmap, the version of entries in the delete bitmap |
390 | | // will be replaced to the correspoding correct version. So before we finally merge a delete bitmap into |
391 | | // tablet's delete bitmap we can use arbitary version number in BitmapKey. Here we define some version numbers |
392 | | // for specific usage during this periods to avoid conflicts |
393 | | constexpr static inline uint64_t TEMP_VERSION_COMMON = 0; |
394 | | |
395 | | /** |
396 | | * |
397 | | * @param tablet_id the tablet which this delete bitmap associates with |
398 | | */ |
399 | | DeleteBitmap(int64_t tablet_id); |
400 | | |
401 | | /** |
402 | | * Copy c-tor for making delete bitmap snapshot on read path |
403 | | */ |
404 | | DeleteBitmap(const DeleteBitmap& r); |
405 | | DeleteBitmap& operator=(const DeleteBitmap& r); |
406 | | /** |
407 | | * Move c-tor for making delete bitmap snapshot on read path |
408 | | */ |
409 | | DeleteBitmap(DeleteBitmap&& r); |
410 | | DeleteBitmap& operator=(DeleteBitmap&& r); |
411 | | |
412 | | /** |
413 | | * Makes a snapshot of delete bitmap, read lock will be acquired in this |
414 | | * process |
415 | | */ |
416 | | DeleteBitmap snapshot() const; |
417 | | |
418 | | /** |
419 | | * Makes a snapshot of delete bitmap on given version, read lock will be |
420 | | * acquired temporary in this process |
421 | | */ |
422 | | DeleteBitmap snapshot(Version version) const; |
423 | | |
424 | | /** |
425 | | * Marks the specific row deleted |
426 | | */ |
427 | | void add(const BitmapKey& bmk, uint32_t row_id); |
428 | | |
429 | | /** |
430 | | * Clears the deletetion mark specific row |
431 | | * |
432 | | * @return non-zero if the associated delete bitmap does not exist |
433 | | */ |
434 | | int remove(const BitmapKey& bmk, uint32_t row_id); |
435 | | |
436 | | /** |
437 | | * Clears bitmaps in range [lower_key, upper_key) |
438 | | */ |
439 | | void remove(const BitmapKey& lower_key, const BitmapKey& upper_key); |
440 | | |
441 | | /** |
442 | | * Checks if the given row is marked deleted |
443 | | * |
444 | | * @return true if marked deleted |
445 | | */ |
446 | | bool contains(const BitmapKey& bmk, uint32_t row_id) const; |
447 | | |
448 | | /** |
449 | | * Checks if this delete bitmap is empty |
450 | | * |
451 | | * @return true if empty |
452 | | */ |
453 | | bool empty() const; |
454 | | |
455 | | /** |
456 | | * return the total cardinality of the Delete Bitmap |
457 | | */ |
458 | | uint64_t cardinality() const; |
459 | | |
460 | | /** |
461 | | * return the total size of the Delete Bitmap(after serialized) |
462 | | */ |
463 | | |
464 | | uint64_t get_size() const; |
465 | | |
466 | | /** |
467 | | * Sets the bitmap of specific segment, it's may be insertion or replacement |
468 | | * |
469 | | * @return 1 if the insertion took place, 0 if the assignment took place |
470 | | */ |
471 | | int set(const BitmapKey& bmk, const roaring::Roaring& segment_delete_bitmap); |
472 | | |
473 | | /** |
474 | | * Gets a copy of specific delete bmk |
475 | | * |
476 | | * @param segment_delete_bitmap output param |
477 | | * @return non-zero if the associated delete bitmap does not exist |
478 | | */ |
479 | | int get(const BitmapKey& bmk, roaring::Roaring* segment_delete_bitmap) const; |
480 | | |
481 | | /** |
482 | | * Gets reference to a specific delete map, DO NOT use this function on a |
483 | | * mutable DeleteBitmap object |
484 | | * @return nullptr if the given bitmap does not exist |
485 | | */ |
486 | | const roaring::Roaring* get(const BitmapKey& bmk) const; |
487 | | |
488 | | /** |
489 | | * Gets subset of delete_bitmap with given range [start, end) |
490 | | * |
491 | | * @parma start start |
492 | | * @parma end end |
493 | | * @parma subset_delete_map output param |
494 | | */ |
495 | | void subset(const BitmapKey& start, const BitmapKey& end, |
496 | | DeleteBitmap* subset_delete_map) const; |
497 | | |
498 | | /** |
499 | | * Merges the given segment delete bitmap into *this |
500 | | * |
501 | | * @param bmk |
502 | | * @param segment_delete_bitmap |
503 | | */ |
504 | | void merge(const BitmapKey& bmk, const roaring::Roaring& segment_delete_bitmap); |
505 | | |
506 | | /** |
507 | | * Merges the given delete bitmap into *this |
508 | | * |
509 | | * @param other |
510 | | */ |
511 | | void merge(const DeleteBitmap& other); |
512 | | |
513 | | /** |
514 | | * Checks if the given row is marked deleted in bitmap with the condition: |
515 | | * all the bitmaps that |
516 | | * RowsetId and SegmentId are the same as the given ones, |
517 | | * and Version <= the given Version |
518 | | * |
519 | | * Note: aggregation cache may be used. |
520 | | * |
521 | | * @return true if marked deleted |
522 | | */ |
523 | | bool contains_agg(const BitmapKey& bitmap, uint32_t row_id) const; |
524 | | |
525 | | bool contains_agg_without_cache(const BitmapKey& bmk, uint32_t row_id) const; |
526 | | /** |
527 | | * Gets aggregated delete_bitmap on rowset_id and version, the same effect: |
528 | | * `select sum(roaring::Roaring) where RowsetId=rowset_id and SegmentId=seg_id and Version <= version` |
529 | | * |
530 | | * @return shared_ptr to a bitmap, which may be empty |
531 | | */ |
532 | | std::shared_ptr<roaring::Roaring> get_agg(const BitmapKey& bmk) const; |
533 | | |
534 | | void remove_sentinel_marks(); |
535 | | |
536 | | void add_to_remove_queue(const std::string& version_str, |
537 | | const std::vector<std::tuple<int64_t, DeleteBitmap::BitmapKey, |
538 | | DeleteBitmap::BitmapKey>>& vector); |
539 | | void remove_stale_delete_bitmap_from_queue(const std::vector<std::string>& vector); |
540 | | |
541 | | uint64_t get_delete_bitmap_count(); |
542 | | |
543 | | class AggCachePolicy : public LRUCachePolicy { |
544 | | public: |
545 | | AggCachePolicy(size_t capacity) |
546 | | : LRUCachePolicy(CachePolicy::CacheType::DELETE_BITMAP_AGG_CACHE, capacity, |
547 | | LRUCacheType::SIZE, |
548 | 1 | config::delete_bitmap_agg_cache_stale_sweep_time_sec, 256) {} |
549 | | }; |
550 | | |
551 | | class AggCache { |
552 | | public: |
553 | | class Value : public LRUCacheValueBase { |
554 | | public: |
555 | | roaring::Roaring bitmap; |
556 | | }; |
557 | | |
558 | 371 | AggCache(size_t size_in_bytes) { |
559 | 371 | static std::once_flag once; |
560 | 371 | std::call_once(once, [size_in_bytes] { |
561 | 1 | auto* tmp = new AggCachePolicy(size_in_bytes); |
562 | 1 | AggCache::s_repr.store(tmp, std::memory_order_release); |
563 | 1 | }); |
564 | | |
565 | 371 | while (!s_repr.load(std::memory_order_acquire)) { |
566 | 0 | } |
567 | 371 | } |
568 | | |
569 | 132 | static LRUCachePolicy* repr() { return s_repr.load(std::memory_order_acquire); } |
570 | | static std::atomic<AggCachePolicy*> s_repr; |
571 | | }; |
572 | | |
573 | | private: |
574 | | mutable std::shared_ptr<AggCache> _agg_cache; |
575 | | int64_t _tablet_id; |
576 | | // <version, <tablet_id, BitmapKeyStart, BitmapKeyEnd>> |
577 | | std::map<std::string, |
578 | | std::vector<std::tuple<int64_t, DeleteBitmap::BitmapKey, DeleteBitmap::BitmapKey>>> |
579 | | _stale_delete_bitmap; |
580 | | }; |
581 | | |
582 | 1.18k | inline TabletUid TabletMeta::tablet_uid() const { |
583 | 1.18k | return _tablet_uid; |
584 | 1.18k | } |
585 | | |
586 | 182 | inline int64_t TabletMeta::table_id() const { |
587 | 182 | return _table_id; |
588 | 182 | } |
589 | | |
590 | 180 | inline int64_t TabletMeta::index_id() const { |
591 | 180 | return _index_id; |
592 | 180 | } |
593 | | |
594 | 562 | inline int64_t TabletMeta::partition_id() const { |
595 | 562 | return _partition_id; |
596 | 562 | } |
597 | | |
598 | 4.65k | inline int64_t TabletMeta::tablet_id() const { |
599 | 4.65k | return _tablet_id; |
600 | 4.65k | } |
601 | | |
602 | 243 | inline int64_t TabletMeta::replica_id() const { |
603 | 243 | return _replica_id; |
604 | 243 | } |
605 | | |
606 | 662 | inline int32_t TabletMeta::schema_hash() const { |
607 | 662 | return _schema_hash; |
608 | 662 | } |
609 | | |
610 | 410 | inline int16_t TabletMeta::shard_id() const { |
611 | 410 | return _shard_id; |
612 | 410 | } |
613 | | |
614 | 16 | inline void TabletMeta::set_shard_id(int32_t shard_id) { |
615 | 16 | _shard_id = shard_id; |
616 | 16 | } |
617 | | |
618 | 180 | inline int64_t TabletMeta::creation_time() const { |
619 | 180 | return _creation_time; |
620 | 180 | } |
621 | | |
622 | 0 | inline void TabletMeta::set_creation_time(int64_t creation_time) { |
623 | 0 | _creation_time = creation_time; |
624 | 0 | } |
625 | | |
626 | 180 | inline int64_t TabletMeta::cumulative_layer_point() const { |
627 | 180 | return _cumulative_layer_point; |
628 | 180 | } |
629 | | |
630 | 0 | inline void TabletMeta::set_cumulative_layer_point(int64_t new_point) { |
631 | 0 | _cumulative_layer_point = new_point; |
632 | 0 | } |
633 | | |
634 | 16 | inline size_t TabletMeta::num_rows() const { |
635 | 16 | size_t num_rows = 0; |
636 | 28 | for (auto& rs : _rs_metas) { |
637 | 28 | num_rows += rs->num_rows(); |
638 | 28 | } |
639 | 16 | return num_rows; |
640 | 16 | } |
641 | | |
642 | 8 | inline size_t TabletMeta::tablet_footprint() const { |
643 | 8 | size_t total_size = 0; |
644 | 26 | for (auto& rs : _rs_metas) { |
645 | 26 | total_size += rs->total_disk_size(); |
646 | 26 | } |
647 | 8 | return total_size; |
648 | 8 | } |
649 | | |
650 | 0 | inline size_t TabletMeta::tablet_local_size() const { |
651 | 0 | size_t total_size = 0; |
652 | 0 | for (auto& rs : _rs_metas) { |
653 | 0 | if (rs->is_local()) { |
654 | 0 | total_size += rs->total_disk_size(); |
655 | 0 | } |
656 | 0 | } |
657 | 0 | return total_size; |
658 | 0 | } |
659 | | |
660 | 0 | inline size_t TabletMeta::tablet_remote_size() const { |
661 | 0 | size_t total_size = 0; |
662 | 0 | for (auto& rs : _rs_metas) { |
663 | 0 | if (!rs->is_local()) { |
664 | 0 | total_size += rs->total_disk_size(); |
665 | 0 | } |
666 | 0 | } |
667 | 0 | return total_size; |
668 | 0 | } |
669 | | |
670 | 50 | inline size_t TabletMeta::version_count() const { |
671 | 50 | return _rs_metas.size(); |
672 | 50 | } |
673 | | |
674 | 25 | inline size_t TabletMeta::stale_version_count() const { |
675 | 25 | return _rs_metas.size(); |
676 | 25 | } |
677 | | |
678 | 8.82k | inline TabletState TabletMeta::tablet_state() const { |
679 | 8.82k | return _tablet_state; |
680 | 8.82k | } |
681 | | |
682 | 46 | inline void TabletMeta::set_tablet_state(TabletState state) { |
683 | 46 | _tablet_state = state; |
684 | 46 | } |
685 | | |
686 | 180 | inline bool TabletMeta::in_restore_mode() const { |
687 | 180 | return _in_restore_mode; |
688 | 180 | } |
689 | | |
690 | 0 | inline void TabletMeta::set_in_restore_mode(bool in_restore_mode) { |
691 | 0 | _in_restore_mode = in_restore_mode; |
692 | 0 | } |
693 | | |
694 | 1.56k | inline const TabletSchemaSPtr& TabletMeta::tablet_schema() const { |
695 | 1.56k | return _schema; |
696 | 1.56k | } |
697 | | |
698 | 0 | inline TabletSchema* TabletMeta::mutable_tablet_schema() { |
699 | 0 | return _schema.get(); |
700 | 0 | } |
701 | | |
702 | 943 | inline const std::vector<RowsetMetaSharedPtr>& TabletMeta::all_rs_metas() const { |
703 | 943 | return _rs_metas; |
704 | 943 | } |
705 | | |
706 | 0 | inline std::vector<RowsetMetaSharedPtr>& TabletMeta::all_mutable_rs_metas() { |
707 | 0 | return _rs_metas; |
708 | 0 | } |
709 | | |
710 | 515 | inline const std::vector<RowsetMetaSharedPtr>& TabletMeta::all_stale_rs_metas() const { |
711 | 515 | return _stale_rs_metas; |
712 | 515 | } |
713 | | |
714 | 0 | inline bool TabletMeta::all_beta() const { |
715 | 0 | for (auto& rs : _rs_metas) { |
716 | 0 | if (rs->rowset_type() != RowsetTypePB::BETA_ROWSET) { |
717 | 0 | return false; |
718 | 0 | } |
719 | 0 | } |
720 | 0 | for (auto& rs : _stale_rs_metas) { |
721 | 0 | if (rs->rowset_type() != RowsetTypePB::BETA_ROWSET) { |
722 | 0 | return false; |
723 | 0 | } |
724 | 0 | } |
725 | 0 | return true; |
726 | 0 | } |
727 | | |
728 | | std::string tablet_state_name(TabletState state); |
729 | | |
730 | | // Only for unit test now. |
731 | | bool operator==(const TabletMeta& a, const TabletMeta& b); |
732 | | bool operator!=(const TabletMeta& a, const TabletMeta& b); |
733 | | |
734 | | } // namespace doris |