Coverage Report

Created: 2025-05-20 17:42

/root/doris/be/src/olap/tablet.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 <butil/macros.h>
21
#include <glog/logging.h>
22
23
#include <atomic>
24
#include <cstddef>
25
#include <cstdint>
26
#include <functional>
27
#include <memory>
28
#include <mutex>
29
#include <ostream>
30
#include <set>
31
#include <shared_mutex>
32
#include <string>
33
#include <string_view>
34
#include <utility>
35
#include <vector>
36
37
#include "common/config.h"
38
#include "common/status.h"
39
#include "olap/base_tablet.h"
40
#include "olap/binlog_config.h"
41
#include "olap/data_dir.h"
42
#include "olap/olap_common.h"
43
#include "olap/partial_update_info.h"
44
#include "olap/rowset/rowset.h"
45
#include "olap/rowset/rowset_meta.h"
46
#include "olap/rowset/rowset_reader.h"
47
#include "olap/rowset/segment_v2/segment.h"
48
#include "olap/version_graph.h"
49
#include "segment_loader.h"
50
#include "util/metrics.h"
51
#include "util/once.h"
52
#include "util/slice.h"
53
54
namespace bvar {
55
template <typename T>
56
class Adder;
57
}
58
59
namespace doris {
60
61
class Tablet;
62
class CumulativeCompactionPolicy;
63
class CompactionMixin;
64
class SingleReplicaCompaction;
65
class RowsetWriter;
66
struct RowsetWriterContext;
67
class TTabletInfo;
68
class TabletMetaPB;
69
class TupleDescriptor;
70
class CalcDeleteBitmapToken;
71
enum CompressKind : int;
72
class RowsetBinlogMetasPB;
73
74
namespace io {
75
class RemoteFileSystem;
76
} // namespace io
77
namespace vectorized {
78
class Block;
79
} // namespace vectorized
80
struct RowLocation;
81
enum KeysType : int;
82
enum SortType : int;
83
84
enum TabletStorageType { STORAGE_TYPE_LOCAL, STORAGE_TYPE_REMOTE, STORAGE_TYPE_REMOTE_AND_LOCAL };
85
86
extern bvar::Adder<uint64_t> unused_remote_rowset_num;
87
88
static inline constexpr auto TRACE_TABLET_LOCK_THRESHOLD = std::chrono::seconds(1);
89
90
struct WriteCooldownMetaExecutors {
91
    WriteCooldownMetaExecutors(size_t executor_nums = 5);
92
93
    void stop();
94
95
    void submit(TabletSharedPtr tablet);
96
5
    size_t _get_executor_pos(int64_t tablet_id) const {
97
5
        return std::hash<int64_t>()(tablet_id) % _executor_nums;
98
5
    };
99
    // Each executor is a mpsc to ensure uploads of the same tablet meta are not concurrent
100
    // FIXME(AlexYue): Use mpsc instead of `ThreadPool` with 1 thread
101
    // We use PriorityThreadPool since it would call status inside it's `shutdown` function.
102
    // Consider one situation where the StackTraceCache's singleton is detructed before
103
    // this WriteCooldownMetaExecutors's singleton, then invoking the status would also call
104
    // StackTraceCache which would then result in heap use after free like #23834
105
    std::vector<std::unique_ptr<PriorityThreadPool>> _executors;
106
    std::unordered_set<int64_t> _pending_tablets;
107
    std::mutex _latch;
108
    size_t _executor_nums;
109
};
110
111
class Tablet final : public BaseTablet {
112
public:
113
    Tablet(StorageEngine& engine, TabletMetaSharedPtr tablet_meta, DataDir* data_dir,
114
           const std::string_view& cumulative_compaction_type = "");
115
116
3.79k
    DataDir* data_dir() const { return _data_dir; }
117
273
    int64_t replica_id() const { return _tablet_meta->replica_id(); }
118
119
11.8k
    const std::string& tablet_path() const { return _tablet_path; }
120
121
    bool set_tablet_schema_into_rowset_meta();
122
    Status init();
123
    bool init_succeeded();
124
125
    bool is_used();
126
127
    void register_tablet_into_dir();
128
    void deregister_tablet_from_dir();
129
130
    void save_meta();
131
    // Used in clone task, to update local meta when finishing a clone job
132
    Status revise_tablet_meta(const std::vector<RowsetSharedPtr>& to_add,
133
                              const std::vector<RowsetSharedPtr>& to_delete,
134
                              bool is_incremental_clone);
135
136
    int64_t cumulative_layer_point() const;
137
    void set_cumulative_layer_point(int64_t new_point);
138
    inline int64_t cumulative_promotion_size() const;
139
    inline void set_cumulative_promotion_size(int64_t new_size);
140
141
    // Disk space occupied by tablet, contain local and remote.
142
    size_t tablet_footprint() override;
143
    // Local disk space occupied by tablet.
144
    size_t tablet_local_size();
145
    // Remote disk space occupied by tablet.
146
    size_t tablet_remote_size();
147
148
    size_t num_rows();
149
    int version_count() const;
150
    int stale_version_count() const;
151
    bool exceed_version_limit(int32_t limit) override;
152
    uint64_t segment_count() const;
153
    Version max_version() const;
154
    CumulativeCompactionPolicy* cumulative_compaction_policy();
155
156
    // properties encapsulated in TabletSchema
157
    SortType sort_type() const;
158
    size_t sort_col_num() const;
159
    size_t num_columns() const;
160
    size_t num_null_columns() const;
161
    size_t num_short_key_columns() const;
162
    size_t num_rows_per_row_block() const;
163
    CompressKind compress_kind() const;
164
    double bloom_filter_fpp() const;
165
    size_t next_unique_id() const;
166
    size_t row_size() const;
167
    int64_t avg_rs_meta_serialize_size() const;
168
169
    // operation in rowsets
170
    Status add_rowset(RowsetSharedPtr rowset);
171
    Status create_initial_rowset(const int64_t version);
172
173
    // MUST hold EXCLUSIVE `_meta_lock`.
174
    Status modify_rowsets(std::vector<RowsetSharedPtr>& to_add,
175
                          std::vector<RowsetSharedPtr>& to_delete, bool check_delete = false);
176
    bool rowset_exists_unlocked(const RowsetSharedPtr& rowset);
177
178
    Status add_inc_rowset(const RowsetSharedPtr& rowset);
179
    /// Delete stale rowset by timing. This delete policy uses now() minutes
180
    /// config::tablet_rowset_expired_stale_sweep_time_sec to compute the deadline of expired rowset
181
    /// to delete.  When rowset is deleted, it will be added to StorageEngine unused map and record
182
    /// need to delete flag.
183
    void delete_expired_stale_rowset();
184
185
    // Given spec_version, find a continuous version path and store it in version_path.
186
    // If quiet is true, then only "does this path exist" is returned.
187
    // If skip_missing_version is true, return ok even there are missing versions.
188
    Status capture_consistent_versions_unlocked(const Version& spec_version, Versions* version_path,
189
                                                bool skip_missing_version, bool quiet) const;
190
191
    // if quiet is true, no error log will be printed if there are missing versions
192
    Status check_version_integrity(const Version& version, bool quiet = false);
193
    bool check_version_exist(const Version& version) const;
194
    void acquire_version_and_rowsets(
195
            std::vector<std::pair<Version, RowsetSharedPtr>>* version_rowsets) const;
196
197
    Status capture_consistent_rowsets_unlocked(
198
            const Version& spec_version, std::vector<RowsetSharedPtr>* rowsets) const override;
199
200
    // If skip_missing_version is true, skip versions if they are missing.
201
    Status capture_rs_readers(const Version& spec_version, std::vector<RowSetSplits>* rs_splits,
202
                              bool skip_missing_version) override;
203
204
    // Find the missed versions until the spec_version.
205
    //
206
    // for example:
207
    //     [0-4][5-5][8-8][9-9][14-14]
208
    // if spec_version = 12, it will return [6, 6], [7, 7], [10, 10], [11, 11], [12, 12]
209
    Versions calc_missed_versions(int64_t spec_version, Versions existing_versions) const override;
210
211
    // meta lock
212
766
    std::shared_mutex& get_header_lock() { return _meta_lock; }
213
2
    std::mutex& get_rowset_update_lock() { return _rowset_update_lock; }
214
32
    std::mutex& get_push_lock() { return _ingest_lock; }
215
21
    std::mutex& get_base_compaction_lock() { return _base_compaction_lock; }
216
298
    std::mutex& get_cumulative_compaction_lock() { return _cumulative_compaction_lock; }
217
1
    std::shared_mutex& get_meta_store_lock() { return _meta_store_lock; }
218
219
52
    std::shared_timed_mutex& get_migration_lock() { return _migration_lock; }
220
221
19
    std::mutex& get_build_inverted_index_lock() { return _build_inverted_index_lock; }
222
223
    // operation for compaction
224
    bool can_do_compaction(size_t path_hash, CompactionType compaction_type);
225
    bool suitable_for_compaction(
226
            CompactionType compaction_type,
227
            std::shared_ptr<CumulativeCompactionPolicy> cumulative_compaction_policy);
228
229
    uint32_t calc_compaction_score();
230
231
    // This function to find max continuous version from the beginning.
232
    // For example: If there are 1, 2, 3, 5, 6, 7 versions belongs tablet, then 3 is target.
233
    // 3 will be saved in "version", and 7 will be saved in "max_version", if max_version != nullptr
234
    void max_continuous_version_from_beginning(Version* version, Version* max_version = nullptr);
235
236
0
    void set_bad(bool is_bad) { _is_bad = is_bad; }
237
238
257
    int64_t last_cumu_compaction_failure_time() { return _last_cumu_compaction_failure_millis; }
239
0
    void set_last_cumu_compaction_failure_time(int64_t millis) {
240
0
        _last_cumu_compaction_failure_millis = millis;
241
0
    }
242
243
0
    int64_t last_base_compaction_failure_time() { return _last_base_compaction_failure_millis; }
244
0
    void set_last_base_compaction_failure_time(int64_t millis) {
245
0
        _last_base_compaction_failure_millis = millis;
246
0
    }
247
248
0
    int64_t last_full_compaction_failure_time() { return _last_full_compaction_failure_millis; }
249
0
    void set_last_full_compaction_failure_time(int64_t millis) {
250
0
        _last_full_compaction_failure_millis = millis;
251
0
    }
252
253
6
    int64_t last_cumu_compaction_success_time() { return _last_cumu_compaction_success_millis; }
254
1
    void set_last_cumu_compaction_success_time(int64_t millis) {
255
1
        _last_cumu_compaction_success_millis = millis;
256
1
    }
257
258
0
    int64_t last_base_compaction_success_time() { return _last_base_compaction_success_millis; }
259
0
    void set_last_base_compaction_success_time(int64_t millis) {
260
0
        _last_base_compaction_success_millis = millis;
261
0
    }
262
263
0
    int64_t last_full_compaction_success_time() { return _last_full_compaction_success_millis; }
264
0
    void set_last_full_compaction_success_time(int64_t millis) {
265
0
        _last_full_compaction_success_millis = millis;
266
0
    }
267
268
0
    int64_t last_cumu_compaction_schedule_time() { return _last_cumu_compaction_schedule_millis; }
269
0
    void set_last_cumu_compaction_schedule_time(int64_t millis) {
270
0
        _last_cumu_compaction_schedule_millis = millis;
271
0
    }
272
273
0
    int64_t last_base_compaction_schedule_time() { return _last_base_compaction_schedule_millis; }
274
0
    void set_last_base_compaction_schedule_time(int64_t millis) {
275
0
        _last_base_compaction_schedule_millis = millis;
276
0
    }
277
278
0
    int64_t last_full_compaction_schedule_time() { return _last_full_compaction_schedule_millis; }
279
0
    void set_last_full_compaction_schedule_time(int64_t millis) {
280
0
        _last_full_compaction_schedule_millis = millis;
281
0
    }
282
283
0
    void set_last_single_compaction_failure_status(std::string status) {
284
0
        _last_single_compaction_failure_status = std::move(status);
285
0
    }
286
287
0
    void set_last_fetched_version(Version version) { _last_fetched_version = std::move(version); }
288
289
    void delete_all_files();
290
291
    void check_tablet_path_exists();
292
293
    std::vector<RowsetSharedPtr> pick_candidate_rowsets_to_cumulative_compaction();
294
    std::vector<RowsetSharedPtr> pick_candidate_rowsets_to_base_compaction();
295
    std::vector<RowsetSharedPtr> pick_candidate_rowsets_to_full_compaction();
296
    std::vector<RowsetSharedPtr> pick_candidate_rowsets_to_build_inverted_index(
297
            const std::set<int64_t>& alter_index_uids, bool is_drop_op);
298
299
    // used for single compaction to get the local versions
300
    // Single compaction does not require remote rowsets and cannot violate the cooldown semantics
301
    std::vector<Version> get_all_local_versions();
302
303
    void calculate_cumulative_point();
304
    // TODO(ygl):
305
0
    bool is_primary_replica() { return false; }
306
307
    // return true if the checkpoint is actually done
308
    bool do_tablet_meta_checkpoint();
309
310
    // Check whether the rowset is useful or not, unuseful rowset can be swept up then.
311
    // Rowset which is under tablet's management is useful, i.e. rowset is in
312
    // _rs_version_map, or _stale_rs_version_map.
313
    // Rowset whose version range is not covered by this tablet is also useful.
314
    bool rowset_meta_is_useful(RowsetMetaSharedPtr rowset_meta);
315
316
    void build_tablet_report_info(TTabletInfo* tablet_info,
317
                                  bool enable_consecutive_missing_check = false,
318
                                  bool enable_path_check = false);
319
320
    // return a json string to show the compaction status of this tablet
321
    void get_compaction_status(std::string* json_result);
322
323
    static Status prepare_compaction_and_calculate_permits(
324
            CompactionType compaction_type, const TabletSharedPtr& tablet,
325
            std::shared_ptr<CompactionMixin>& compaction, int64_t& permits);
326
327
    void execute_compaction(CompactionMixin& compaction);
328
    void execute_single_replica_compaction(SingleReplicaCompaction& compaction);
329
330
    void set_cumulative_compaction_policy(
331
0
            std::shared_ptr<CumulativeCompactionPolicy> cumulative_compaction_policy) {
332
0
        _cumulative_compaction_policy = cumulative_compaction_policy;
333
0
    }
334
335
0
    std::shared_ptr<CumulativeCompactionPolicy> get_cumulative_compaction_policy() {
336
0
        return _cumulative_compaction_policy;
337
0
    }
338
339
0
    void set_last_cumu_compaction_status(std::string status) {
340
0
        _last_cumu_compaction_status = std::move(status);
341
0
    }
342
343
0
    std::string get_last_cumu_compaction_status() { return _last_cumu_compaction_status; }
344
345
0
    void set_last_base_compaction_status(std::string status) {
346
0
        _last_base_compaction_status = std::move(status);
347
0
    }
348
349
0
    std::string get_last_base_compaction_status() { return _last_base_compaction_status; }
350
351
0
    void set_last_full_compaction_status(std::string status) {
352
0
        _last_full_compaction_status = std::move(status);
353
0
    }
354
355
0
    std::string get_last_full_compaction_status() { return _last_full_compaction_status; }
356
357
    std::tuple<int64_t, int64_t> get_visible_version_and_time() const;
358
359
559
    void set_visible_version(const std::shared_ptr<const VersionWithTime>& visible_version) {
360
559
        std::atomic_store_explicit(&_visible_version, visible_version, std::memory_order_relaxed);
361
559
    }
362
363
    bool should_fetch_from_peer();
364
365
0
    inline bool all_beta() const {
366
0
        std::shared_lock rdlock(_meta_lock);
367
0
        return _tablet_meta->all_beta();
368
0
    }
369
370
0
    const TabletSchemaSPtr& tablet_schema_unlocked() const { return _max_version_schema; }
371
372
    Result<std::unique_ptr<RowsetWriter>> create_rowset_writer(RowsetWriterContext& context,
373
                                                               bool vertical) override;
374
375
    Result<std::unique_ptr<RowsetWriter>> create_transient_rowset_writer(
376
            const Rowset& rowset, std::shared_ptr<PartialUpdateInfo> partial_update_info,
377
            int64_t txn_expiration = 0) override;
378
    Result<std::unique_ptr<RowsetWriter>> create_transient_rowset_writer(
379
            RowsetWriterContext& context, const RowsetId& rowset_id);
380
381
    Status create_rowset(const RowsetMetaSharedPtr& rowset_meta, RowsetSharedPtr* rowset);
382
383
    // MUST hold EXCLUSIVE `_meta_lock`
384
    void add_rowsets(const std::vector<RowsetSharedPtr>& to_add);
385
    // MUST hold EXCLUSIVE `_meta_lock`
386
    Status delete_rowsets(const std::vector<RowsetSharedPtr>& to_delete, bool move_to_stale);
387
388
    // MUST hold SHARED `_meta_lock`
389
485
    const auto& rowset_map() const { return _rs_version_map; }
390
    // MUST hold SHARED `_meta_lock`
391
480
    const auto& stale_rowset_map() const { return _stale_rs_version_map; }
392
393
    ////////////////////////////////////////////////////////////////////////////
394
    // begin cooldown functions
395
    ////////////////////////////////////////////////////////////////////////////
396
22
    int64_t storage_policy_id() const { return _tablet_meta->storage_policy_id(); }
397
4
    void set_storage_policy_id(int64_t id) { _tablet_meta->set_storage_policy_id(id); }
398
399
0
    int64_t last_failed_follow_cooldown_time() const { return _last_failed_follow_cooldown_time; }
400
401
    // Cooldown to remote fs.
402
    Status cooldown(RowsetSharedPtr rowset = nullptr);
403
404
    RowsetSharedPtr pick_cooldown_rowset();
405
406
    RowsetSharedPtr need_cooldown(int64_t* cooldown_timestamp, size_t* file_size);
407
408
    struct CooldownConf {
409
        int64_t term = -1;
410
        int64_t cooldown_replica_id = -1;
411
    };
412
413
0
    CooldownConf cooldown_conf() const {
414
0
        std::shared_lock rlock(_cooldown_conf_lock);
415
0
        return _cooldown_conf;
416
0
    }
417
418
0
    CooldownConf cooldown_conf_unlocked() const { return _cooldown_conf; }
419
420
    // Return `true` if update success
421
    bool update_cooldown_conf(int64_t cooldown_term, int64_t cooldown_replica_id);
422
423
    Status remove_all_remote_rowsets();
424
425
    void record_unused_remote_rowset(const RowsetId& rowset_id, const std::string& resource,
426
                                     int64_t num_segments);
427
428
    uint32_t calc_cold_data_compaction_score() const;
429
430
19
    std::mutex& get_cold_compaction_lock() { return _cold_compaction_lock; }
431
432
0
    std::shared_mutex& get_cooldown_conf_lock() { return _cooldown_conf_lock; }
433
434
    static void async_write_cooldown_meta(TabletSharedPtr tablet);
435
    // Return `ABORTED` if should not to retry again
436
    Status write_cooldown_meta();
437
    ////////////////////////////////////////////////////////////////////////////
438
    // end cooldown functions
439
    ////////////////////////////////////////////////////////////////////////////
440
441
    CalcDeleteBitmapExecutor* calc_delete_bitmap_executor() override;
442
    Status save_delete_bitmap(const TabletTxnInfo* txn_info, int64_t txn_id,
443
                              DeleteBitmapPtr delete_bitmap, RowsetWriter* rowset_writer,
444
                              const RowsetIdUnorderedSet& cur_rowset_ids, int64_t lock_id = -1,
445
                              int64_t next_visible_version = -1) override;
446
447
    void merge_delete_bitmap(const DeleteBitmap& delete_bitmap);
448
    bool check_all_rowset_segment();
449
450
    void update_max_version_schema(const TabletSchemaSPtr& tablet_schema);
451
452
    void set_skip_compaction(bool skip,
453
                             CompactionType compaction_type = CompactionType::CUMULATIVE_COMPACTION,
454
                             int64_t start = -1);
455
    bool should_skip_compaction(CompactionType compaction_type, int64_t now);
456
457
    std::vector<std::string> get_binlog_filepath(std::string_view binlog_version) const;
458
    std::pair<std::string, int64_t> get_binlog_info(std::string_view binlog_version) const;
459
    std::string get_rowset_binlog_meta(std::string_view binlog_version,
460
                                       std::string_view rowset_id) const;
461
    Status get_rowset_binlog_metas(const std::vector<int64_t>& binlog_versions,
462
                                   RowsetBinlogMetasPB* metas_pb);
463
    Status get_rowset_binlog_metas(Version binlog_versions, RowsetBinlogMetasPB* metas_pb);
464
    std::string get_segment_filepath(std::string_view rowset_id,
465
                                     std::string_view segment_index) const;
466
    std::string get_segment_filepath(std::string_view rowset_id, int64_t segment_index) const;
467
    bool can_add_binlog(uint64_t total_binlog_size) const;
468
    void gc_binlogs(int64_t version);
469
    Status ingest_binlog_metas(RowsetBinlogMetasPB* metas_pb);
470
471
    void report_error(const Status& st);
472
473
0
    inline int64_t get_io_error_times() const { return _io_error_times; }
474
475
0
    inline bool is_io_error_too_times() const {
476
0
        return config::max_tablet_io_errors > 0 && _io_error_times >= config::max_tablet_io_errors;
477
0
    }
478
479
0
    int64_t get_table_id() { return _tablet_meta->table_id(); }
480
481
    // binlog related functions
482
    bool is_enable_binlog();
483
0
    bool is_binlog_enabled() { return _tablet_meta->binlog_config().is_enable(); }
484
0
    int64_t binlog_ttl_ms() const { return _tablet_meta->binlog_config().ttl_seconds(); }
485
0
    int64_t binlog_max_bytes() const { return _tablet_meta->binlog_config().max_bytes(); }
486
487
    void set_binlog_config(BinlogConfig binlog_config);
488
489
0
    void set_alter_failed(bool alter_failed) { _alter_failed = alter_failed; }
490
0
    bool is_alter_failed() { return _alter_failed; }
491
492
0
    void set_is_full_compaction_running(bool is_full_compaction_running) {
493
0
        _is_full_compaction_running = is_full_compaction_running;
494
0
    }
495
0
    inline bool is_full_compaction_running() const { return _is_full_compaction_running; }
496
    void clear_cache() override;
497
498
3
    int32_t get_compaction_score() const { return _compaction_score; }
499
500
0
    void set_compaction_score(int32_t compaction_score) { _compaction_score = compaction_score; }
501
502
10.9k
    void add_compaction_score(int32_t score) {
503
10.9k
        if (_compaction_score < 0) {
504
10.9k
            return;
505
10.9k
        }
506
21
        _compaction_score += score;
507
21
    }
508
509
0
    void minus_compaction_score(int32_t score) {
510
0
        if (_compaction_score < 0) {
511
0
            return;
512
0
        }
513
0
        _compaction_score -= score;
514
0
    }
515
516
private:
517
    Status _init_once_action();
518
    bool _contains_rowset(const RowsetId rowset_id);
519
    Status _contains_version(const Version& version);
520
521
    // Returns:
522
    // version: the max continuous version from beginning
523
    // max_version: the max version of this tablet
524
    void _max_continuous_version_from_beginning_unlocked(Version* version, Version* max_version,
525
                                                         bool* has_version_cross) const;
526
    RowsetSharedPtr _rowset_with_largest_size();
527
    /// Delete stale rowset by version. This method not only delete the version in expired rowset map,
528
    /// but also delete the version in rowset meta vector.
529
    void _delete_stale_rowset_by_version(const Version& version);
530
531
    uint32_t _calc_cumulative_compaction_score(
532
            std::shared_ptr<CumulativeCompactionPolicy> cumulative_compaction_policy);
533
    uint32_t _calc_base_compaction_score() const;
534
535
    std::vector<RowsetSharedPtr> _pick_visible_rowsets_to_compaction(int64_t min_start_version,
536
                                                                     int64_t max_start_version);
537
538
    void _init_context_common_fields(RowsetWriterContext& context);
539
540
    ////////////////////////////////////////////////////////////////////////////
541
    // begin cooldown functions
542
    ////////////////////////////////////////////////////////////////////////////
543
    Status _cooldown_data(RowsetSharedPtr rowset);
544
    Status _follow_cooldowned_data();
545
    Status _read_cooldown_meta(const StorageResource& storage_resource,
546
                               TabletMetaPB* tablet_meta_pb);
547
    bool _has_data_to_cooldown();
548
    int64_t _get_newest_cooldown_time(const RowsetSharedPtr& rowset);
549
    ////////////////////////////////////////////////////////////////////////////
550
    // end cooldown functions
551
    ////////////////////////////////////////////////////////////////////////////
552
553
    void _clear_cache_by_rowset(const BetaRowsetSharedPtr& rowset);
554
    void check_table_size_correctness();
555
    std::string get_segment_path(const RowsetMetaSharedPtr& rs_meta, int64_t seg_id);
556
    int64_t get_segment_file_size(const RowsetMetaSharedPtr& rs_meta);
557
    int64_t get_inverted_index_file_size(const RowsetMetaSharedPtr& rs_meta);
558
559
public:
560
    static const int64_t K_INVALID_CUMULATIVE_POINT = -1;
561
562
private:
563
    StorageEngine& _engine;
564
    DataDir* _data_dir = nullptr;
565
566
    std::string _tablet_path;
567
568
    DorisCallOnce<Status> _init_once;
569
    // meta store lock is used for prevent 2 threads do checkpoint concurrently
570
    // it will be used in econ-mode in the future
571
    std::shared_mutex _meta_store_lock;
572
    std::mutex _ingest_lock;
573
    std::mutex _base_compaction_lock;
574
    std::mutex _cumulative_compaction_lock;
575
    std::shared_timed_mutex _migration_lock;
576
    std::mutex _build_inverted_index_lock;
577
578
    // In unique key table with MoW, we should guarantee that only one
579
    // writer can update rowset and delete bitmap at the same time.
580
    // We use a separate lock rather than _meta_lock, to avoid blocking read queries
581
    // during publish_txn, which might take hundreds of milliseconds
582
    mutable std::mutex _rowset_update_lock;
583
584
    // if this tablet is broken, set to true. default is false
585
    std::atomic<bool> _is_bad;
586
    // timestamp of last cumu compaction failure
587
    std::atomic<int64_t> _last_cumu_compaction_failure_millis;
588
    // timestamp of last base compaction failure
589
    std::atomic<int64_t> _last_base_compaction_failure_millis;
590
    // timestamp of last full compaction failure
591
    std::atomic<int64_t> _last_full_compaction_failure_millis;
592
    // timestamp of last cumu compaction success
593
    std::atomic<int64_t> _last_cumu_compaction_success_millis;
594
    // timestamp of last base compaction success
595
    std::atomic<int64_t> _last_base_compaction_success_millis;
596
    // timestamp of last full compaction success
597
    std::atomic<int64_t> _last_full_compaction_success_millis;
598
    // timestamp of last cumu compaction schedule time
599
    std::atomic<int64_t> _last_cumu_compaction_schedule_millis;
600
    // timestamp of last base compaction schedule time
601
    std::atomic<int64_t> _last_base_compaction_schedule_millis;
602
    // timestamp of last full compaction schedule time
603
    std::atomic<int64_t> _last_full_compaction_schedule_millis;
604
    std::atomic<int64_t> _cumulative_point;
605
    std::atomic<int64_t> _cumulative_promotion_size;
606
    std::atomic<int32_t> _newly_created_rowset_num;
607
    std::atomic<int64_t> _last_checkpoint_time;
608
    std::string _last_cumu_compaction_status;
609
    std::string _last_base_compaction_status;
610
    std::string _last_full_compaction_status;
611
612
    // single replica compaction status
613
    std::string _last_single_compaction_failure_status;
614
    Version _last_fetched_version;
615
616
    // cumulative compaction policy
617
    std::shared_ptr<CumulativeCompactionPolicy> _cumulative_compaction_policy;
618
    std::string_view _cumulative_compaction_type;
619
620
    // use a separate thread to check all tablets paths existence
621
    std::atomic<bool> _is_tablet_path_exists;
622
623
    int64_t _last_missed_version;
624
    int64_t _last_missed_time_s;
625
626
    bool _skip_cumu_compaction = false;
627
    int64_t _skip_cumu_compaction_ts;
628
629
    bool _skip_base_compaction = false;
630
    int64_t _skip_base_compaction_ts;
631
632
    // cooldown related
633
    CooldownConf _cooldown_conf;
634
    // `_cooldown_conf_lock` is used to serialize update cooldown conf and all operations that:
635
    // 1. read cooldown conf
636
    // 2. upload rowsets to remote storage
637
    // 3. update cooldown meta id
638
    mutable std::shared_mutex _cooldown_conf_lock;
639
    // `_cold_compaction_lock` is used to serialize cold data compaction and all operations that
640
    // may delete compaction input rowsets.
641
    std::mutex _cold_compaction_lock;
642
    int64_t _last_failed_follow_cooldown_time = 0;
643
    // `_alter_failed` is used to indicate whether the tablet failed to perform a schema change
644
    std::atomic<bool> _alter_failed = false;
645
646
    int64_t _io_error_times = 0;
647
648
    // partition's visible version. it sync from fe, but not real-time.
649
    std::shared_ptr<const VersionWithTime> _visible_version;
650
651
    std::atomic_bool _is_full_compaction_running = false;
652
653
    int32_t _compaction_score = -1;
654
    int32_t _score_check_cnt = 0;
655
};
656
657
20
inline CumulativeCompactionPolicy* Tablet::cumulative_compaction_policy() {
658
20
    return _cumulative_compaction_policy.get();
659
20
}
660
661
285
inline bool Tablet::init_succeeded() {
662
285
    return _init_once.has_called() && _init_once.stored_result().ok();
663
285
}
664
665
263
inline bool Tablet::is_used() {
666
263
    return !_is_bad && _data_dir->is_used();
667
263
}
668
669
304
inline void Tablet::register_tablet_into_dir() {
670
304
    _data_dir->register_tablet(this);
671
304
}
672
673
255
inline void Tablet::deregister_tablet_from_dir() {
674
255
    _data_dir->deregister_tablet(this);
675
255
}
676
677
38
inline int64_t Tablet::cumulative_layer_point() const {
678
38
    return _cumulative_point;
679
38
}
680
681
329
inline void Tablet::set_cumulative_layer_point(int64_t new_point) {
682
    // cumulative point should only be reset to -1, or be increased
683
329
    CHECK(new_point == Tablet::K_INVALID_CUMULATIVE_POINT || new_point >= _cumulative_point)
684
0
            << "Unexpected cumulative point: " << new_point
685
0
            << ", origin: " << _cumulative_point.load();
686
329
    _cumulative_point = new_point;
687
329
}
688
689
30
inline int64_t Tablet::cumulative_promotion_size() const {
690
30
    return _cumulative_promotion_size;
691
30
}
692
693
47
inline void Tablet::set_cumulative_promotion_size(int64_t new_size) {
694
47
    _cumulative_promotion_size = new_size;
695
47
}
696
697
// TODO(lingbin): Why other methods that need to get information from _tablet_meta
698
// are not locked, here needs a comment to explain.
699
8
inline size_t Tablet::tablet_footprint() {
700
8
    std::shared_lock rdlock(_meta_lock);
701
8
    return _tablet_meta->tablet_footprint();
702
8
}
703
704
0
inline size_t Tablet::tablet_local_size() {
705
0
    std::shared_lock rdlock(_meta_lock);
706
0
    return _tablet_meta->tablet_local_size();
707
0
}
708
709
0
inline size_t Tablet::tablet_remote_size() {
710
0
    std::shared_lock rdlock(_meta_lock);
711
0
    return _tablet_meta->tablet_remote_size();
712
0
}
713
714
// TODO(lingbin): Why other methods which need to get information from _tablet_meta
715
// are not locked, here needs a comment to explain.
716
19
inline size_t Tablet::num_rows() {
717
19
    std::shared_lock rdlock(_meta_lock);
718
19
    return _tablet_meta->num_rows();
719
19
}
720
721
28
inline int Tablet::version_count() const {
722
28
    std::shared_lock rdlock(_meta_lock);
723
28
    return _tablet_meta->version_count();
724
28
}
725
726
28
inline int Tablet::stale_version_count() const {
727
28
    std::shared_lock rdlock(_meta_lock);
728
28
    return _tablet_meta->stale_version_count();
729
28
}
730
731
45
inline Version Tablet::max_version() const {
732
45
    std::shared_lock rdlock(_meta_lock);
733
45
    return _tablet_meta->max_version();
734
45
}
735
736
0
inline uint64_t Tablet::segment_count() const {
737
0
    std::shared_lock rdlock(_meta_lock);
738
0
    uint64_t segment_nums = 0;
739
0
    for (const auto& rs_meta : _tablet_meta->all_rs_metas()) {
740
0
        segment_nums += rs_meta->num_segments();
741
0
    }
742
0
    return segment_nums;
743
0
}
744
745
0
inline SortType Tablet::sort_type() const {
746
0
    return _tablet_meta->tablet_schema()->sort_type();
747
0
}
748
749
0
inline size_t Tablet::sort_col_num() const {
750
0
    return _tablet_meta->tablet_schema()->sort_col_num();
751
0
}
752
753
0
inline size_t Tablet::num_columns() const {
754
0
    return _tablet_meta->tablet_schema()->num_columns();
755
0
}
756
757
0
inline size_t Tablet::num_null_columns() const {
758
0
    return _tablet_meta->tablet_schema()->num_null_columns();
759
0
}
760
761
0
inline size_t Tablet::num_short_key_columns() const {
762
0
    return _tablet_meta->tablet_schema()->num_short_key_columns();
763
0
}
764
765
0
inline size_t Tablet::num_rows_per_row_block() const {
766
0
    return _tablet_meta->tablet_schema()->num_rows_per_row_block();
767
0
}
768
769
0
inline CompressKind Tablet::compress_kind() const {
770
0
    return _tablet_meta->tablet_schema()->compress_kind();
771
0
}
772
773
0
inline double Tablet::bloom_filter_fpp() const {
774
0
    return _tablet_meta->tablet_schema()->bloom_filter_fpp();
775
0
}
776
777
0
inline size_t Tablet::next_unique_id() const {
778
0
    return _tablet_meta->tablet_schema()->next_column_unique_id();
779
0
}
780
781
0
inline size_t Tablet::row_size() const {
782
0
    return _tablet_meta->tablet_schema()->row_size();
783
0
}
784
785
28
inline int64_t Tablet::avg_rs_meta_serialize_size() const {
786
28
    return _tablet_meta->avg_rs_meta_serialize_size();
787
28
}
788
789
} // namespace doris