Coverage Report

Created: 2026-07-29 14:09

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
/root/doris/be/src/olap/tablet.h
Line
Count
Source
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/atomic_shared_ptr.h"
38
#include "common/config.h"
39
#include "common/status.h"
40
#include "olap/base_tablet.h"
41
#include "olap/binlog_config.h"
42
#include "olap/data_dir.h"
43
#include "olap/olap_common.h"
44
#include "olap/partial_update_info.h"
45
#include "olap/rowset/rowset.h"
46
#include "olap/rowset/rowset_meta.h"
47
#include "olap/rowset/rowset_reader.h"
48
#include "olap/rowset/segment_v2/segment.h"
49
#include "olap/version_graph.h"
50
#include "segment_loader.h"
51
#include "util/metrics.h"
52
#include "util/once.h"
53
#include "util/slice.h"
54
55
namespace bvar {
56
template <typename T>
57
class Adder;
58
}
59
60
namespace doris {
61
#include "common/compile_check_begin.h"
62
63
class Tablet;
64
class CumulativeCompactionPolicy;
65
class CompactionMixin;
66
class SingleReplicaCompaction;
67
class RowsetWriter;
68
struct RowsetWriterContext;
69
class TTabletInfo;
70
class TabletMetaPB;
71
class TupleDescriptor;
72
class CalcDeleteBitmapToken;
73
enum CompressKind : int;
74
class RowsetBinlogMetasPB;
75
76
namespace io {
77
class RemoteFileSystem;
78
} // namespace io
79
namespace vectorized {
80
class Block;
81
} // namespace vectorized
82
struct RowLocation;
83
enum KeysType : int;
84
enum SortType : int;
85
86
enum TabletStorageType { STORAGE_TYPE_LOCAL, STORAGE_TYPE_REMOTE, STORAGE_TYPE_REMOTE_AND_LOCAL };
87
88
extern bvar::Adder<uint64_t> unused_remote_rowset_num;
89
90
static inline constexpr auto TRACE_TABLET_LOCK_THRESHOLD = std::chrono::seconds(1);
91
92
struct WriteCooldownMetaExecutors {
93
    WriteCooldownMetaExecutors(size_t executor_nums = 5);
94
95
    void stop();
96
97
    void submit(TabletSharedPtr tablet);
98
205k
    size_t _get_executor_pos(int64_t tablet_id) const {
99
205k
        return std::hash<int64_t>()(tablet_id) % _executor_nums;
100
205k
    };
101
    // Each executor is a mpsc to ensure uploads of the same tablet meta are not concurrent
102
    // FIXME(AlexYue): Use mpsc instead of `ThreadPool` with 1 thread
103
    // We use PriorityThreadPool since it would call status inside it's `shutdown` function.
104
    // Consider one situation where the StackTraceCache's singleton is detructed before
105
    // this WriteCooldownMetaExecutors's singleton, then invoking the status would also call
106
    // StackTraceCache which would then result in heap use after free like #23834
107
    std::vector<std::unique_ptr<PriorityThreadPool>> _executors;
108
    std::unordered_set<int64_t> _pending_tablets;
109
    std::mutex _latch;
110
    size_t _executor_nums;
111
};
112
113
class Tablet final : public BaseTablet {
114
public:
115
    Tablet(StorageEngine& engine, TabletMetaSharedPtr tablet_meta, DataDir* data_dir,
116
           const std::string_view& cumulative_compaction_type = "");
117
118
37.0M
    DataDir* data_dir() const { return _data_dir; }
119
1.29M
    int64_t replica_id() const { return _tablet_meta->replica_id(); }
120
121
44.7k
    std::string tablet_path() const override { return _tablet_path; }
122
123
    bool set_tablet_schema_into_rowset_meta();
124
    Status init();
125
    bool init_succeeded();
126
127
    bool is_used();
128
129
    void register_tablet_into_dir();
130
    void deregister_tablet_from_dir();
131
132
    void save_meta();
133
    // Used in clone task, to update local meta when finishing a clone job
134
    Status revise_tablet_meta(const std::vector<RowsetSharedPtr>& to_add,
135
                              const std::vector<RowsetSharedPtr>& to_delete,
136
                              bool is_incremental_clone);
137
138
    int64_t cumulative_layer_point() const;
139
    void set_cumulative_layer_point(int64_t new_point);
140
    inline int64_t cumulative_promotion_size() const;
141
    inline void set_cumulative_promotion_size(int64_t new_size);
142
143
    // Disk space occupied by tablet, contain local and remote.
144
    size_t tablet_footprint() override;
145
    // Local disk space occupied by tablet.
146
    size_t tablet_local_size();
147
    // Remote disk space occupied by tablet.
148
    size_t tablet_remote_size();
149
150
    size_t num_rows();
151
    size_t version_count() const;
152
    size_t stale_version_count() const;
153
    bool exceed_version_limit(int32_t limit) override;
154
    uint64_t segment_count() const;
155
    Version max_version() const;
156
    CumulativeCompactionPolicy* cumulative_compaction_policy();
157
158
    // properties encapsulated in TabletSchema
159
    SortType sort_type() const;
160
    size_t sort_col_num() const;
161
    size_t num_columns() const;
162
    size_t num_null_columns() const;
163
    size_t num_short_key_columns() const;
164
    size_t num_rows_per_row_block() const;
165
    double bloom_filter_fpp() const;
166
    size_t next_unique_id() 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
    // if quiet is true, no error log will be printed if there are missing versions
186
    Status check_version_integrity(const Version& version, bool quiet = false);
187
    bool check_version_exist(const Version& version) const;
188
    void acquire_version_and_rowsets(
189
            std::vector<std::pair<Version, RowsetSharedPtr>>* version_rowsets) const;
190
191
    // If skip_missing_version is true, skip versions if they are missing.
192
    Status capture_rs_readers(const Version& spec_version, std::vector<RowSetSplits>* rs_splits,
193
                              const CaptureRowsetOps& opts) override;
194
195
    // Find the missed versions until the spec_version.
196
    //
197
    // for example:
198
    //     [0-4][5-5][8-8][9-9][14-14]
199
    // if spec_version = 12, it will return [6, 6], [7, 7], [10, 10], [11, 11], [12, 12]
200
    Versions calc_missed_versions(int64_t spec_version, Versions existing_versions) const override;
201
202
    // meta lock
203
239k
    BthreadSharedMutex& get_header_lock() { return _meta_lock; }
204
642
    std::mutex& get_rowset_update_lock() { return _rowset_update_lock; }
205
632
    std::mutex& get_push_lock() { return _ingest_lock; }
206
6.97M
    std::mutex& get_base_compaction_lock() { return _base_compaction_lock; }
207
8.86M
    std::mutex& get_cumulative_compaction_lock() { return _cumulative_compaction_lock; }
208
1
    std::shared_mutex& get_meta_store_lock() { return _meta_store_lock; }
209
210
1.25k
    std::shared_timed_mutex& get_migration_lock() { return _migration_lock; }
211
212
22
    std::mutex& get_build_inverted_index_lock() { return _build_inverted_index_lock; }
213
214
    // operation for compaction
215
    bool can_do_compaction(size_t path_hash, CompactionType compaction_type);
216
    bool suitable_for_compaction(
217
            CompactionType compaction_type,
218
            std::shared_ptr<CumulativeCompactionPolicy> cumulative_compaction_policy);
219
220
    uint32_t calc_compaction_score();
221
222
    // This function to find max continuous version from the beginning.
223
    // For example: If there are 1, 2, 3, 5, 6, 7 versions belongs tablet, then 3 is target.
224
    // 3 will be saved in "version", and 7 will be saved in "max_version", if max_version != nullptr
225
    void max_continuous_version_from_beginning(Version* version, Version* max_version = nullptr);
226
227
0
    void set_bad(bool is_bad) { _is_bad = is_bad; }
228
229
17.0M
    int64_t last_cumu_compaction_failure_time() { return _last_cumu_compaction_failure_millis; }
230
100k
    void set_last_cumu_compaction_failure_time(int64_t millis) {
231
100k
        _last_cumu_compaction_failure_millis = millis;
232
100k
    }
233
234
6.97M
    int64_t last_base_compaction_failure_time() { return _last_base_compaction_failure_millis; }
235
32.3k
    void set_last_base_compaction_failure_time(int64_t millis) {
236
32.3k
        _last_base_compaction_failure_millis = millis;
237
32.3k
    }
238
239
0
    int64_t last_full_compaction_failure_time() { return _last_full_compaction_failure_millis; }
240
0
    void set_last_full_compaction_failure_time(int64_t millis) {
241
0
        _last_full_compaction_failure_millis = millis;
242
0
    }
243
244
96.7k
    int64_t last_cumu_compaction_success_time() { return _last_cumu_compaction_success_millis; }
245
2.56k
    void set_last_cumu_compaction_success_time(int64_t millis) {
246
2.56k
        _last_cumu_compaction_success_millis = millis;
247
2.56k
    }
248
249
96.7k
    int64_t last_base_compaction_success_time() { return _last_base_compaction_success_millis; }
250
2.51k
    void set_last_base_compaction_success_time(int64_t millis) {
251
2.51k
        _last_base_compaction_success_millis = millis;
252
2.51k
    }
253
254
0
    int64_t last_full_compaction_success_time() { return _last_full_compaction_success_millis; }
255
0
    void set_last_full_compaction_success_time(int64_t millis) {
256
0
        _last_full_compaction_success_millis = millis;
257
0
    }
258
259
0
    int64_t last_cumu_compaction_schedule_time() { return _last_cumu_compaction_schedule_millis; }
260
100k
    void set_last_cumu_compaction_schedule_time(int64_t millis) {
261
100k
        _last_cumu_compaction_schedule_millis = millis;
262
100k
    }
263
264
0
    int64_t last_base_compaction_schedule_time() { return _last_base_compaction_schedule_millis; }
265
16.1k
    void set_last_base_compaction_schedule_time(int64_t millis) {
266
16.1k
        _last_base_compaction_schedule_millis = millis;
267
16.1k
    }
268
269
0
    int64_t last_full_compaction_schedule_time() { return _last_full_compaction_schedule_millis; }
270
0
    void set_last_full_compaction_schedule_time(int64_t millis) {
271
0
        _last_full_compaction_schedule_millis = millis;
272
0
    }
273
274
0
    void set_last_single_compaction_failure_status(std::string status) {
275
0
        _last_single_compaction_failure_status = std::move(status);
276
0
    }
277
278
0
    void set_last_fetched_version(Version version) { _last_fetched_version = std::move(version); }
279
280
    void delete_all_files();
281
282
    void check_tablet_path_exists();
283
284
    std::vector<RowsetSharedPtr> pick_candidate_rowsets_to_cumulative_compaction();
285
    // MUST hold shared `_meta_lock`. Use this when the caller already holds the
286
    // header lock (e.g. the time-series cumulative score path under
287
    // suitable_for_compaction) to avoid recursive shared acquisition.
288
    std::vector<RowsetSharedPtr> pick_candidate_rowsets_to_cumulative_compaction_unlocked();
289
    std::vector<RowsetSharedPtr> pick_candidate_rowsets_to_base_compaction();
290
    std::vector<RowsetSharedPtr> pick_candidate_rowsets_to_full_compaction();
291
    std::vector<RowsetSharedPtr> pick_candidate_rowsets_to_build_inverted_index(
292
            const std::set<int64_t>& alter_index_uids, bool is_drop_op);
293
294
    // used for single compaction to get the local versions
295
    // Single compaction does not require remote rowsets and cannot violate the cooldown semantics
296
    std::vector<Version> get_all_local_versions();
297
298
    void calculate_cumulative_point();
299
    // TODO(ygl):
300
0
    bool is_primary_replica() { return false; }
301
302
    // return true if the checkpoint is actually done
303
    bool do_tablet_meta_checkpoint();
304
305
    // Check whether the rowset is useful or not, unuseful rowset can be swept up then.
306
    // Rowset which is under tablet's management is useful, i.e. rowset is in
307
    // _rs_version_map, or _stale_rs_version_map.
308
    // Rowset whose version range is not covered by this tablet is also useful.
309
    bool rowset_meta_is_useful(RowsetMetaSharedPtr rowset_meta);
310
311
    void build_tablet_report_info(TTabletInfo* tablet_info,
312
                                  bool enable_consecutive_missing_check = false,
313
                                  bool enable_path_check = false);
314
315
    // return a json string to show the compaction status of this tablet
316
    void get_compaction_status(std::string* json_result);
317
318
    static Status prepare_compaction_and_calculate_permits(
319
            CompactionType compaction_type, const TabletSharedPtr& tablet,
320
            std::shared_ptr<CompactionMixin>& compaction, int64_t& permits);
321
322
    void execute_compaction(CompactionMixin& compaction);
323
    void execute_single_replica_compaction(SingleReplicaCompaction& compaction);
324
325
    void set_cumulative_compaction_policy(
326
0
            std::shared_ptr<CumulativeCompactionPolicy> cumulative_compaction_policy) {
327
0
        _cumulative_compaction_policy = cumulative_compaction_policy;
328
0
    }
329
330
0
    std::shared_ptr<CumulativeCompactionPolicy> get_cumulative_compaction_policy() {
331
0
        return _cumulative_compaction_policy;
332
0
    }
333
334
100k
    void set_last_cumu_compaction_status(std::string status) {
335
100k
        _last_cumu_compaction_status = std::move(status);
336
100k
    }
337
338
0
    std::string get_last_cumu_compaction_status() { return _last_cumu_compaction_status; }
339
340
32.3k
    void set_last_base_compaction_status(std::string status) {
341
32.3k
        _last_base_compaction_status = std::move(status);
342
32.3k
    }
343
344
0
    std::string get_last_base_compaction_status() { return _last_base_compaction_status; }
345
346
0
    void set_last_full_compaction_status(std::string status) {
347
0
        _last_full_compaction_status = std::move(status);
348
0
    }
349
350
0
    std::string get_last_full_compaction_status() { return _last_full_compaction_status; }
351
352
    std::tuple<int64_t, int64_t> get_visible_version_and_time() const;
353
354
323k
    void set_visible_version(const std::shared_ptr<const VersionWithTime>& visible_version) {
355
323k
        _visible_version.store(visible_version);
356
323k
    }
357
358
    bool should_fetch_from_peer();
359
360
0
    inline bool all_beta() const {
361
0
        std::shared_lock rdlock(_meta_lock);
362
0
        return _tablet_meta->all_beta();
363
0
    }
364
365
0
    const TabletSchemaSPtr& tablet_schema_unlocked() const { return _max_version_schema; }
366
367
    Result<std::unique_ptr<RowsetWriter>> create_rowset_writer(RowsetWriterContext& context,
368
                                                               bool vertical) override;
369
370
    Result<std::unique_ptr<RowsetWriter>> create_transient_rowset_writer(
371
            const Rowset& rowset, std::shared_ptr<PartialUpdateInfo> partial_update_info,
372
            int64_t txn_expiration = 0) override;
373
    Result<std::unique_ptr<RowsetWriter>> create_transient_rowset_writer(
374
            RowsetWriterContext& context, const RowsetId& rowset_id);
375
376
    Status create_rowset(const RowsetMetaSharedPtr& rowset_meta, RowsetSharedPtr* rowset);
377
378
    // MUST hold EXCLUSIVE `_meta_lock`
379
    void add_rowsets(const std::vector<RowsetSharedPtr>& to_add);
380
    // MUST hold EXCLUSIVE `_meta_lock`
381
    Status delete_rowsets(const std::vector<RowsetSharedPtr>& to_delete, bool move_to_stale);
382
383
    // MUST hold SHARED `_meta_lock`
384
33.2k
    const auto& rowset_map() const { return _rs_version_map; }
385
    // MUST hold SHARED `_meta_lock`
386
33.2k
    const auto& stale_rowset_map() const { return _stale_rs_version_map; }
387
388
    ////////////////////////////////////////////////////////////////////////////
389
    // begin cooldown functions
390
    ////////////////////////////////////////////////////////////////////////////
391
859k
    int64_t storage_policy_id() const { return _tablet_meta->storage_policy_id(); }
392
    void set_storage_policy_id(int64_t id) { _tablet_meta->set_storage_policy_id(id); }
393
394
4.05M
    int64_t last_failed_follow_cooldown_time() const { return _last_failed_follow_cooldown_time; }
395
396
    // Cooldown to remote fs.
397
    Status cooldown(RowsetSharedPtr rowset = nullptr);
398
399
    RowsetSharedPtr pick_cooldown_rowset();
400
401
    RowsetSharedPtr need_cooldown(int64_t* cooldown_timestamp, size_t* file_size);
402
403
    struct CooldownConf {
404
        int64_t term = -1;
405
        int64_t cooldown_replica_id = -1;
406
    };
407
408
7
    CooldownConf cooldown_conf() const {
409
7
        std::shared_lock rlock(_cooldown_conf_lock);
410
7
        return _cooldown_conf;
411
7
    }
412
413
21
    CooldownConf cooldown_conf_unlocked() const { return _cooldown_conf; }
414
415
    // Return `true` if update success
416
    bool update_cooldown_conf(int64_t cooldown_term, int64_t cooldown_replica_id);
417
418
    Status remove_all_remote_rowsets();
419
420
    void record_unused_remote_rowset(const RowsetId& rowset_id, const std::string& resource,
421
                                     int64_t num_segments);
422
423
    uint32_t calc_cold_data_compaction_score() const;
424
425
22
    std::mutex& get_cold_compaction_lock() { return _cold_compaction_lock; }
426
427
0
    std::shared_mutex& get_cooldown_conf_lock() { return _cooldown_conf_lock; }
428
429
    static void async_write_cooldown_meta(TabletSharedPtr tablet);
430
    // Return `ABORTED` if should not to retry again
431
    Status write_cooldown_meta();
432
    ////////////////////////////////////////////////////////////////////////////
433
    // end cooldown functions
434
    ////////////////////////////////////////////////////////////////////////////
435
436
    CalcDeleteBitmapExecutor* calc_delete_bitmap_executor() override;
437
    Status save_delete_bitmap(const TabletTxnInfo* txn_info, int64_t txn_id,
438
                              DeleteBitmapPtr delete_bitmap, RowsetWriter* rowset_writer,
439
                              const RowsetIdUnorderedSet& cur_rowset_ids, int64_t lock_id = -1,
440
                              int64_t next_visible_version = -1) override;
441
442
    void merge_delete_bitmap(const DeleteBitmap& delete_bitmap);
443
    bool check_all_rowset_segment();
444
445
    void update_max_version_schema(const TabletSchemaSPtr& tablet_schema);
446
447
    void set_skip_compaction(bool skip,
448
                             CompactionType compaction_type = CompactionType::CUMULATIVE_COMPACTION,
449
                             int64_t start = -1);
450
    bool should_skip_compaction(CompactionType compaction_type, int64_t now);
451
452
    std::vector<std::string> get_binlog_filepath(std::string_view binlog_version) const;
453
    std::pair<std::string, int64_t> get_binlog_info(std::string_view binlog_version) const;
454
    std::string get_rowset_binlog_meta(std::string_view binlog_version,
455
                                       std::string_view rowset_id) const;
456
    Status get_rowset_binlog_metas(const std::vector<int64_t>& binlog_versions,
457
                                   RowsetBinlogMetasPB* metas_pb);
458
    Status get_rowset_binlog_metas(Version binlog_versions, RowsetBinlogMetasPB* metas_pb);
459
    std::string get_segment_filepath(std::string_view rowset_id,
460
                                     std::string_view segment_index) const;
461
    std::string get_segment_filepath(std::string_view rowset_id, int64_t segment_index) const;
462
    bool can_add_binlog(uint64_t total_binlog_size) const;
463
    void gc_binlogs(int64_t version);
464
    Status ingest_binlog_metas(RowsetBinlogMetasPB* metas_pb);
465
466
    void report_error(const Status& st);
467
468
0
    inline int64_t get_io_error_times() const { return _io_error_times; }
469
470
1.29M
    inline bool is_io_error_too_times() const {
471
1.29M
        return config::max_tablet_io_errors > 0 && _io_error_times >= config::max_tablet_io_errors;
472
1.29M
    }
473
474
677
    int64_t get_table_id() { return _tablet_meta->table_id(); }
475
476
    // binlog related functions
477
    bool is_enable_binlog();
478
0
    bool is_binlog_enabled() { return _tablet_meta->binlog_config().is_enable(); }
479
0
    int64_t binlog_ttl_ms() const { return _tablet_meta->binlog_config().ttl_seconds(); }
480
0
    int64_t binlog_max_bytes() const { return _tablet_meta->binlog_config().max_bytes(); }
481
482
    void set_binlog_config(BinlogConfig binlog_config);
483
484
0
    void set_is_full_compaction_running(bool is_full_compaction_running) {
485
0
        _is_full_compaction_running = is_full_compaction_running;
486
0
    }
487
0
    inline bool is_full_compaction_running() const { return _is_full_compaction_running; }
488
    void clear_cache() override;
489
490
    int32_t get_compaction_score() const { return _compaction_score; }
491
492
0
    void set_compaction_score(int32_t compaction_score) { _compaction_score = compaction_score; }
493
494
11.8k
    void add_compaction_score(int32_t score) {
495
11.8k
        if (_compaction_score < 0) {
496
11.2k
            return;
497
11.2k
        }
498
671
        _compaction_score += score;
499
671
    }
500
501
0
    void minus_compaction_score(int32_t score) {
502
0
        if (_compaction_score < 0) {
503
0
            return;
504
0
        }
505
0
        _compaction_score -= score;
506
0
    }
507
508
    Status prepare_txn(TPartitionId partition_id, TTransactionId transaction_id,
509
                       const PUniqueId& load_id, bool ingest);
510
    // TODO: commit_txn
511
512
private:
513
    Status _init_once_action();
514
    bool _contains_rowset(const RowsetId rowset_id);
515
    Status _contains_version(const Version& version);
516
517
    // Returns:
518
    // version: the max continuous version from beginning
519
    // max_version: the max version of this tablet
520
    void _max_continuous_version_from_beginning_unlocked(Version* version, Version* max_version,
521
                                                         bool* has_version_cross) const;
522
    RowsetSharedPtr _rowset_with_largest_size();
523
    /// Delete stale rowset by version. This method not only delete the version in expired rowset map,
524
    /// but also delete the version in rowset meta vector.
525
    void _delete_stale_rowset_by_version(const Version& version);
526
527
    uint32_t _calc_cumulative_compaction_score(
528
            std::shared_ptr<CumulativeCompactionPolicy> cumulative_compaction_policy);
529
    uint32_t _calc_base_compaction_score() const;
530
531
    std::vector<RowsetSharedPtr> _pick_visible_rowsets_to_compaction(int64_t min_start_version,
532
                                                                     int64_t max_start_version);
533
    // MUST hold shared `_meta_lock`.
534
    std::vector<RowsetSharedPtr> _pick_visible_rowsets_to_compaction_unlocked(
535
            int64_t min_start_version, int64_t max_start_version);
536
537
    void _init_context_common_fields(RowsetWriterContext& context);
538
539
    ////////////////////////////////////////////////////////////////////////////
540
    // begin cooldown functions
541
    ////////////////////////////////////////////////////////////////////////////
542
    Status _cooldown_data(RowsetSharedPtr rowset);
543
    Status _follow_cooldowned_data();
544
    Status _read_cooldown_meta(const StorageResource& storage_resource,
545
                               TabletMetaPB* tablet_meta_pb);
546
    bool _has_data_to_cooldown();
547
    int64_t _get_newest_cooldown_time(const RowsetSharedPtr& rowset);
548
    ////////////////////////////////////////////////////////////////////////////
549
    // end cooldown functions
550
    ////////////////////////////////////////////////////////////////////////////
551
552
    void _clear_cache_by_rowset(const BetaRowsetSharedPtr& rowset);
553
    void check_table_size_correctness();
554
    std::string get_segment_path(const RowsetMetaSharedPtr& rs_meta, int64_t seg_id);
555
    int64_t get_segment_file_size(const RowsetMetaSharedPtr& rs_meta);
556
    int64_t get_inverted_index_file_size(const RowsetMetaSharedPtr& rs_meta);
557
558
public:
559
    static const int64_t K_INVALID_CUMULATIVE_POINT = -1;
560
561
private:
562
    StorageEngine& _engine;
563
    DataDir* _data_dir = nullptr;
564
565
    std::string _tablet_path;
566
567
    DorisCallOnce<Status> _init_once;
568
    // meta store lock is used for prevent 2 threads do checkpoint concurrently
569
    // it will be used in econ-mode in the future
570
    std::shared_mutex _meta_store_lock;
571
    std::mutex _ingest_lock;
572
    std::mutex _base_compaction_lock;
573
    std::mutex _cumulative_compaction_lock;
574
    std::shared_timed_mutex _migration_lock;
575
    std::mutex _build_inverted_index_lock;
576
577
    // In unique key table with MoW, we should guarantee that only one
578
    // writer can update rowset and delete bitmap at the same time.
579
    // We use a separate lock rather than _meta_lock, to avoid blocking read queries
580
    // during publish_txn, which might take hundreds of milliseconds
581
    mutable std::mutex _rowset_update_lock;
582
583
    // if this tablet is broken, set to true. default is false
584
    std::atomic<bool> _is_bad;
585
    // timestamp of last cumu compaction failure
586
    std::atomic<int64_t> _last_cumu_compaction_failure_millis;
587
    // timestamp of last base compaction failure
588
    std::atomic<int64_t> _last_base_compaction_failure_millis;
589
    // timestamp of last full compaction failure
590
    std::atomic<int64_t> _last_full_compaction_failure_millis;
591
    // timestamp of last cumu compaction success
592
    std::atomic<int64_t> _last_cumu_compaction_success_millis;
593
    // timestamp of last base compaction success
594
    std::atomic<int64_t> _last_base_compaction_success_millis;
595
    // timestamp of last full compaction success
596
    std::atomic<int64_t> _last_full_compaction_success_millis;
597
    // timestamp of last cumu compaction schedule time
598
    std::atomic<int64_t> _last_cumu_compaction_schedule_millis;
599
    // timestamp of last base compaction schedule time
600
    std::atomic<int64_t> _last_base_compaction_schedule_millis;
601
    // timestamp of last full compaction schedule time
602
    std::atomic<int64_t> _last_full_compaction_schedule_millis;
603
    std::atomic<int64_t> _cumulative_point;
604
    std::atomic<int64_t> _cumulative_promotion_size;
605
    std::atomic<int32_t> _newly_created_rowset_num;
606
    std::atomic<int64_t> _last_checkpoint_time;
607
    std::string _last_cumu_compaction_status;
608
    std::string _last_base_compaction_status;
609
    std::string _last_full_compaction_status;
610
611
    // single replica compaction status
612
    std::string _last_single_compaction_failure_status;
613
    Version _last_fetched_version;
614
615
    // cumulative compaction policy
616
    std::shared_ptr<CumulativeCompactionPolicy> _cumulative_compaction_policy;
617
    std::string_view _cumulative_compaction_type;
618
619
    // use a separate thread to check all tablets paths existence
620
    std::atomic<bool> _is_tablet_path_exists;
621
622
    int64_t _last_missed_version;
623
    int64_t _last_missed_time_s;
624
625
    bool _skip_cumu_compaction = false;
626
    int64_t _skip_cumu_compaction_ts;
627
628
    bool _skip_base_compaction = false;
629
    int64_t _skip_base_compaction_ts;
630
631
    // cooldown related
632
    CooldownConf _cooldown_conf;
633
    // `_cooldown_conf_lock` is used to serialize update cooldown conf and all operations that:
634
    // 1. read cooldown conf
635
    // 2. upload rowsets to remote storage
636
    // 3. update cooldown meta id
637
    mutable std::shared_mutex _cooldown_conf_lock;
638
    // `_cold_compaction_lock` is used to serialize cold data compaction and all operations that
639
    // may delete compaction input rowsets.
640
    std::mutex _cold_compaction_lock;
641
    int64_t _last_failed_follow_cooldown_time = 0;
642
643
    int64_t _io_error_times = 0;
644
645
    // partition's visible version. it sync from fe, but not real-time.
646
    atomic_shared_ptr<const VersionWithTime> _visible_version;
647
648
    std::atomic_bool _is_full_compaction_running = false;
649
650
    int32_t _compaction_score = -1;
651
    int32_t _score_check_cnt = 0;
652
};
653
654
97.0k
inline CumulativeCompactionPolicy* Tablet::cumulative_compaction_policy() {
655
97.0k
    return _cumulative_compaction_policy.get();
656
97.0k
}
657
658
17.4M
inline bool Tablet::init_succeeded() {
659
17.4M
    return _init_once.has_called() && _init_once.stored_result().ok();
660
17.4M
}
661
662
19.7M
inline bool Tablet::is_used() {
663
19.7M
    return !_is_bad && _data_dir->is_used();
664
19.7M
}
665
666
322k
inline void Tablet::register_tablet_into_dir() {
667
322k
    _data_dir->register_tablet(this);
668
322k
}
669
670
942
inline void Tablet::deregister_tablet_from_dir() {
671
942
    _data_dir->deregister_tablet(this);
672
942
}
673
674
720k
inline int64_t Tablet::cumulative_layer_point() const {
675
720k
    return _cumulative_point;
676
720k
}
677
678
2.97k
inline void Tablet::set_cumulative_layer_point(int64_t new_point) {
679
    // cumulative point should only be reset to -1, or be increased
680
2.97k
    CHECK(new_point == Tablet::K_INVALID_CUMULATIVE_POINT || new_point >= _cumulative_point)
681
0
            << "Unexpected cumulative point: " << new_point
682
0
            << ", origin: " << _cumulative_point.load();
683
2.97k
    _cumulative_point = new_point;
684
2.97k
}
685
686
96.9k
inline int64_t Tablet::cumulative_promotion_size() const {
687
96.9k
    return _cumulative_promotion_size;
688
96.9k
}
689
690
135k
inline void Tablet::set_cumulative_promotion_size(int64_t new_size) {
691
135k
    _cumulative_promotion_size = new_size;
692
135k
}
693
694
// TODO(lingbin): Why other methods that need to get information from _tablet_meta
695
// are not locked, here needs a comment to explain.
696
2.69k
inline size_t Tablet::tablet_footprint() {
697
2.69k
    std::shared_lock rdlock(_meta_lock);
698
2.69k
    return _tablet_meta->tablet_footprint();
699
2.69k
}
700
701
3.08M
inline size_t Tablet::tablet_local_size() {
702
3.08M
    std::shared_lock rdlock(_meta_lock);
703
3.08M
    return _tablet_meta->tablet_local_size();
704
3.08M
}
705
706
3.08M
inline size_t Tablet::tablet_remote_size() {
707
3.08M
    std::shared_lock rdlock(_meta_lock);
708
3.08M
    return _tablet_meta->tablet_remote_size();
709
3.08M
}
710
711
// TODO(lingbin): Why other methods which need to get information from _tablet_meta
712
// are not locked, here needs a comment to explain.
713
inline size_t Tablet::num_rows() {
714
    std::shared_lock rdlock(_meta_lock);
715
    return _tablet_meta->num_rows();
716
}
717
718
5.02M
inline size_t Tablet::version_count() const {
719
5.02M
    std::shared_lock rdlock(_meta_lock);
720
5.02M
    return _tablet_meta->version_count();
721
5.02M
}
722
723
628
inline size_t Tablet::stale_version_count() const {
724
628
    std::shared_lock rdlock(_meta_lock);
725
628
    return _tablet_meta->stale_version_count();
726
628
}
727
728
419k
inline Version Tablet::max_version() const {
729
419k
    std::shared_lock rdlock(_meta_lock);
730
419k
    return _tablet_meta->max_version();
731
419k
}
732
733
4.87M
inline uint64_t Tablet::segment_count() const {
734
4.87M
    std::shared_lock rdlock(_meta_lock);
735
4.87M
    uint64_t segment_nums = 0;
736
8.74M
    for (const auto& [_, rs_meta] : _tablet_meta->all_rs_metas()) {
737
8.74M
        segment_nums += rs_meta->num_segments();
738
8.74M
    }
739
4.87M
    return segment_nums;
740
4.87M
}
741
742
0
inline SortType Tablet::sort_type() const {
743
0
    return _tablet_meta->tablet_schema()->sort_type();
744
0
}
745
746
0
inline size_t Tablet::sort_col_num() const {
747
0
    return _tablet_meta->tablet_schema()->sort_col_num();
748
0
}
749
750
0
inline size_t Tablet::num_columns() const {
751
0
    return _tablet_meta->tablet_schema()->num_columns();
752
0
}
753
754
0
inline size_t Tablet::num_null_columns() const {
755
0
    return _tablet_meta->tablet_schema()->num_null_columns();
756
0
}
757
758
0
inline size_t Tablet::num_short_key_columns() const {
759
0
    return _tablet_meta->tablet_schema()->num_short_key_columns();
760
0
}
761
762
0
inline size_t Tablet::num_rows_per_row_block() const {
763
0
    return _tablet_meta->tablet_schema()->num_rows_per_row_block();
764
0
}
765
766
0
inline double Tablet::bloom_filter_fpp() const {
767
0
    return _tablet_meta->tablet_schema()->bloom_filter_fpp();
768
0
}
769
770
0
inline size_t Tablet::next_unique_id() const {
771
0
    return _tablet_meta->tablet_schema()->next_column_unique_id();
772
0
}
773
774
628
inline int64_t Tablet::avg_rs_meta_serialize_size() const {
775
628
    return _tablet_meta->avg_rs_meta_serialize_size();
776
628
}
777
778
#include "common/compile_check_end.h"
779
} // namespace doris