Coverage Report

Created: 2025-06-18 22:27

/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
3.80k
    DataDir* data_dir() const { return _data_dir; }
117
268
    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
    uint32_t calc_compaction_score(
226
            CompactionType compaction_type,
227
            std::shared_ptr<CumulativeCompactionPolicy> cumulative_compaction_policy);
228
229
    // This function to find max continuous version from the beginning.
230
    // For example: If there are 1, 2, 3, 5, 6, 7 versions belongs tablet, then 3 is target.
231
    // 3 will be saved in "version", and 7 will be saved in "max_version", if max_version != nullptr
232
    void max_continuous_version_from_beginning(Version* version, Version* max_version = nullptr);
233
234
0
    void set_bad(bool is_bad) { _is_bad = is_bad; }
235
236
255
    int64_t last_cumu_compaction_failure_time() { return _last_cumu_compaction_failure_millis; }
237
1
    void set_last_cumu_compaction_failure_time(int64_t millis) {
238
1
        _last_cumu_compaction_failure_millis = millis;
239
1
    }
240
241
0
    int64_t last_base_compaction_failure_time() { return _last_base_compaction_failure_millis; }
242
0
    void set_last_base_compaction_failure_time(int64_t millis) {
243
0
        _last_base_compaction_failure_millis = millis;
244
0
    }
245
246
0
    int64_t last_full_compaction_failure_time() { return _last_full_compaction_failure_millis; }
247
0
    void set_last_full_compaction_failure_time(int64_t millis) {
248
0
        _last_full_compaction_failure_millis = millis;
249
0
    }
250
251
0
    int64_t last_cumu_compaction_success_time() { return _last_cumu_compaction_success_millis; }
252
2
    void set_last_cumu_compaction_success_time(int64_t millis) {
253
2
        _last_cumu_compaction_success_millis = millis;
254
2
    }
255
256
0
    int64_t last_base_compaction_success_time() { return _last_base_compaction_success_millis; }
257
0
    void set_last_base_compaction_success_time(int64_t millis) {
258
0
        _last_base_compaction_success_millis = millis;
259
0
    }
260
261
0
    int64_t last_full_compaction_success_time() { return _last_full_compaction_success_millis; }
262
0
    void set_last_full_compaction_success_time(int64_t millis) {
263
0
        _last_full_compaction_success_millis = millis;
264
0
    }
265
266
0
    int64_t last_cumu_compaction_schedule_time() { return _last_cumu_compaction_schedule_millis; }
267
0
    void set_last_cumu_compaction_schedule_time(int64_t millis) {
268
0
        _last_cumu_compaction_schedule_millis = millis;
269
0
    }
270
271
0
    int64_t last_base_compaction_schedule_time() { return _last_base_compaction_schedule_millis; }
272
0
    void set_last_base_compaction_schedule_time(int64_t millis) {
273
0
        _last_base_compaction_schedule_millis = millis;
274
0
    }
275
276
0
    int64_t last_full_compaction_schedule_time() { return _last_full_compaction_schedule_millis; }
277
0
    void set_last_full_compaction_schedule_time(int64_t millis) {
278
0
        _last_full_compaction_schedule_millis = millis;
279
0
    }
280
281
0
    void set_last_single_compaction_failure_status(std::string status) {
282
0
        _last_single_compaction_failure_status = std::move(status);
283
0
    }
284
285
0
    void set_last_fetched_version(Version version) { _last_fetched_version = std::move(version); }
286
287
    void delete_all_files();
288
289
    void check_tablet_path_exists();
290
291
    std::vector<RowsetSharedPtr> pick_candidate_rowsets_to_cumulative_compaction();
292
    std::vector<RowsetSharedPtr> pick_candidate_rowsets_to_base_compaction();
293
    std::vector<RowsetSharedPtr> pick_candidate_rowsets_to_full_compaction();
294
    std::vector<RowsetSharedPtr> pick_candidate_rowsets_to_build_inverted_index(
295
            const std::set<int64_t>& alter_index_uids, bool is_drop_op);
296
297
    // used for single compaction to get the local versions
298
    // Single compaction does not require remote rowsets and cannot violate the cooldown semantics
299
    std::vector<Version> get_all_local_versions();
300
301
    void calculate_cumulative_point();
302
    // TODO(ygl):
303
0
    bool is_primary_replica() { return false; }
304
305
    // return true if the checkpoint is actually done
306
    bool do_tablet_meta_checkpoint();
307
308
    // Check whether the rowset is useful or not, unuseful rowset can be swept up then.
309
    // Rowset which is under tablet's management is useful, i.e. rowset is in
310
    // _rs_version_map, or _stale_rs_version_map.
311
    // Rowset whose version range is not covered by this tablet is also useful.
312
    bool rowset_meta_is_useful(RowsetMetaSharedPtr rowset_meta);
313
314
    void build_tablet_report_info(TTabletInfo* tablet_info,
315
                                  bool enable_consecutive_missing_check = false,
316
                                  bool enable_path_check = false);
317
318
    // return a json string to show the compaction status of this tablet
319
    void get_compaction_status(std::string* json_result);
320
321
    static Status prepare_compaction_and_calculate_permits(
322
            CompactionType compaction_type, const TabletSharedPtr& tablet,
323
            std::shared_ptr<CompactionMixin>& compaction, int64_t& permits);
324
325
    void execute_compaction(CompactionMixin& compaction);
326
    void execute_single_replica_compaction(SingleReplicaCompaction& compaction);
327
328
    void set_cumulative_compaction_policy(
329
0
            std::shared_ptr<CumulativeCompactionPolicy> cumulative_compaction_policy) {
330
0
        _cumulative_compaction_policy = cumulative_compaction_policy;
331
0
    }
332
333
0
    std::shared_ptr<CumulativeCompactionPolicy> get_cumulative_compaction_policy() {
334
0
        return _cumulative_compaction_policy;
335
0
    }
336
337
1
    void set_last_cumu_compaction_status(std::string status) {
338
1
        _last_cumu_compaction_status = std::move(status);
339
1
    }
340
341
0
    std::string get_last_cumu_compaction_status() { return _last_cumu_compaction_status; }
342
343
0
    void set_last_base_compaction_status(std::string status) {
344
0
        _last_base_compaction_status = std::move(status);
345
0
    }
346
347
0
    std::string get_last_base_compaction_status() { return _last_base_compaction_status; }
348
349
0
    void set_last_full_compaction_status(std::string status) {
350
0
        _last_full_compaction_status = std::move(status);
351
0
    }
352
353
0
    std::string get_last_full_compaction_status() { return _last_full_compaction_status; }
354
355
    std::tuple<int64_t, int64_t> get_visible_version_and_time() const;
356
357
559
    void set_visible_version(const std::shared_ptr<const VersionWithTime>& visible_version) {
358
559
        std::atomic_store_explicit(&_visible_version, visible_version, std::memory_order_relaxed);
359
559
    }
360
361
    bool should_fetch_from_peer();
362
363
0
    inline bool all_beta() const {
364
0
        std::shared_lock rdlock(_meta_lock);
365
0
        return _tablet_meta->all_beta();
366
0
    }
367
368
0
    const TabletSchemaSPtr& tablet_schema_unlocked() const { return _max_version_schema; }
369
370
    Result<std::unique_ptr<RowsetWriter>> create_rowset_writer(RowsetWriterContext& context,
371
                                                               bool vertical) override;
372
373
    Result<std::unique_ptr<RowsetWriter>> create_transient_rowset_writer(
374
            const Rowset& rowset, std::shared_ptr<PartialUpdateInfo> partial_update_info,
375
            int64_t txn_expiration = 0) override;
376
    Result<std::unique_ptr<RowsetWriter>> create_transient_rowset_writer(
377
            RowsetWriterContext& context, const RowsetId& rowset_id);
378
379
    Status create_rowset(const RowsetMetaSharedPtr& rowset_meta, RowsetSharedPtr* rowset);
380
381
    // MUST hold EXCLUSIVE `_meta_lock`
382
    void add_rowsets(const std::vector<RowsetSharedPtr>& to_add);
383
    // MUST hold EXCLUSIVE `_meta_lock`
384
    Status delete_rowsets(const std::vector<RowsetSharedPtr>& to_delete, bool move_to_stale);
385
386
    // MUST hold SHARED `_meta_lock`
387
485
    const auto& rowset_map() const { return _rs_version_map; }
388
    // MUST hold SHARED `_meta_lock`
389
480
    const auto& stale_rowset_map() const { return _stale_rs_version_map; }
390
391
    ////////////////////////////////////////////////////////////////////////////
392
    // begin cooldown functions
393
    ////////////////////////////////////////////////////////////////////////////
394
22
    int64_t storage_policy_id() const { return _tablet_meta->storage_policy_id(); }
395
4
    void set_storage_policy_id(int64_t id) { _tablet_meta->set_storage_policy_id(id); }
396
397
0
    int64_t last_failed_follow_cooldown_time() const { return _last_failed_follow_cooldown_time; }
398
399
    // Cooldown to remote fs.
400
    Status cooldown(RowsetSharedPtr rowset = nullptr);
401
402
    RowsetSharedPtr pick_cooldown_rowset();
403
404
    RowsetSharedPtr need_cooldown(int64_t* cooldown_timestamp, size_t* file_size);
405
406
    struct CooldownConf {
407
        int64_t term = -1;
408
        int64_t cooldown_replica_id = -1;
409
    };
410
411
0
    CooldownConf cooldown_conf() const {
412
0
        std::shared_lock rlock(_cooldown_conf_lock);
413
0
        return _cooldown_conf;
414
0
    }
415
416
0
    CooldownConf cooldown_conf_unlocked() const { return _cooldown_conf; }
417
418
    // Return `true` if update success
419
    bool update_cooldown_conf(int64_t cooldown_term, int64_t cooldown_replica_id);
420
421
    Status remove_all_remote_rowsets();
422
423
    void record_unused_remote_rowset(const RowsetId& rowset_id, const std::string& resource,
424
                                     int64_t num_segments);
425
426
    uint32_t calc_cold_data_compaction_score() const;
427
428
19
    std::mutex& get_cold_compaction_lock() { return _cold_compaction_lock; }
429
430
0
    std::shared_mutex& get_cooldown_conf_lock() { return _cooldown_conf_lock; }
431
432
    static void async_write_cooldown_meta(TabletSharedPtr tablet);
433
    // Return `ABORTED` if should not to retry again
434
    Status write_cooldown_meta();
435
    ////////////////////////////////////////////////////////////////////////////
436
    // end cooldown functions
437
    ////////////////////////////////////////////////////////////////////////////
438
439
    CalcDeleteBitmapExecutor* calc_delete_bitmap_executor() override;
440
    Status save_delete_bitmap(const TabletTxnInfo* txn_info, int64_t txn_id,
441
                              DeleteBitmapPtr delete_bitmap, RowsetWriter* rowset_writer,
442
                              const RowsetIdUnorderedSet& cur_rowset_ids,
443
                              int64_t next_visible_version = -1) override;
444
445
    void merge_delete_bitmap(const DeleteBitmap& delete_bitmap);
446
    bool check_all_rowset_segment();
447
448
    void update_max_version_schema(const TabletSchemaSPtr& tablet_schema);
449
450
    void set_skip_compaction(bool skip,
451
                             CompactionType compaction_type = CompactionType::CUMULATIVE_COMPACTION,
452
                             int64_t start = -1);
453
    bool should_skip_compaction(CompactionType compaction_type, int64_t now);
454
455
    std::vector<std::string> get_binlog_filepath(std::string_view binlog_version) const;
456
    std::pair<std::string, int64_t> get_binlog_info(std::string_view binlog_version) const;
457
    std::string get_rowset_binlog_meta(std::string_view binlog_version,
458
                                       std::string_view rowset_id) const;
459
    Status get_rowset_binlog_metas(const std::vector<int64_t>& binlog_versions,
460
                                   RowsetBinlogMetasPB* metas_pb);
461
    Status get_rowset_binlog_metas(Version binlog_versions, RowsetBinlogMetasPB* metas_pb);
462
    std::string get_segment_filepath(std::string_view rowset_id,
463
                                     std::string_view segment_index) const;
464
    std::string get_segment_filepath(std::string_view rowset_id, int64_t segment_index) const;
465
    bool can_add_binlog(uint64_t total_binlog_size) const;
466
    void gc_binlogs(int64_t version);
467
    Status ingest_binlog_metas(RowsetBinlogMetasPB* metas_pb);
468
469
    void report_error(const Status& st);
470
471
0
    inline int64_t get_io_error_times() const { return _io_error_times; }
472
473
0
    inline bool is_io_error_too_times() const {
474
0
        return config::max_tablet_io_errors > 0 && _io_error_times >= config::max_tablet_io_errors;
475
0
    }
476
477
0
    int64_t get_table_id() { return _tablet_meta->table_id(); }
478
479
    // binlog related functions
480
    bool is_enable_binlog();
481
0
    bool is_binlog_enabled() { return _tablet_meta->binlog_config().is_enable(); }
482
0
    int64_t binlog_ttl_ms() const { return _tablet_meta->binlog_config().ttl_seconds(); }
483
0
    int64_t binlog_max_bytes() const { return _tablet_meta->binlog_config().max_bytes(); }
484
485
    void set_binlog_config(BinlogConfig binlog_config);
486
487
0
    void set_alter_failed(bool alter_failed) { _alter_failed = alter_failed; }
488
0
    bool is_alter_failed() { return _alter_failed; }
489
490
0
    void set_is_full_compaction_running(bool is_full_compaction_running) {
491
0
        _is_full_compaction_running = is_full_compaction_running;
492
0
    }
493
0
    inline bool is_full_compaction_running() const { return _is_full_compaction_running; }
494
    void clear_cache() override;
495
496
private:
497
    Status _init_once_action();
498
    bool _contains_rowset(const RowsetId rowset_id);
499
    Status _contains_version(const Version& version);
500
501
    // Returns:
502
    // version: the max continuous version from beginning
503
    // max_version: the max version of this tablet
504
    void _max_continuous_version_from_beginning_unlocked(Version* version, Version* max_version,
505
                                                         bool* has_version_cross) const;
506
    RowsetSharedPtr _rowset_with_largest_size();
507
    /// Delete stale rowset by version. This method not only delete the version in expired rowset map,
508
    /// but also delete the version in rowset meta vector.
509
    void _delete_stale_rowset_by_version(const Version& version);
510
511
    uint32_t _calc_cumulative_compaction_score(
512
            std::shared_ptr<CumulativeCompactionPolicy> cumulative_compaction_policy);
513
    uint32_t _calc_base_compaction_score() const;
514
515
    std::vector<RowsetSharedPtr> _pick_visible_rowsets_to_compaction(int64_t min_start_version,
516
                                                                     int64_t max_start_version);
517
518
    void _init_context_common_fields(RowsetWriterContext& context);
519
520
    ////////////////////////////////////////////////////////////////////////////
521
    // begin cooldown functions
522
    ////////////////////////////////////////////////////////////////////////////
523
    Status _cooldown_data(RowsetSharedPtr rowset);
524
    Status _follow_cooldowned_data();
525
    Status _read_cooldown_meta(const StorageResource& storage_resource,
526
                               TabletMetaPB* tablet_meta_pb);
527
    bool _has_data_to_cooldown();
528
    int64_t _get_newest_cooldown_time(const RowsetSharedPtr& rowset);
529
    ////////////////////////////////////////////////////////////////////////////
530
    // end cooldown functions
531
    ////////////////////////////////////////////////////////////////////////////
532
533
    void _clear_cache_by_rowset(const BetaRowsetSharedPtr& rowset);
534
    void check_table_size_correctness();
535
    std::string get_segment_path(const RowsetMetaSharedPtr& rs_meta, int64_t seg_id);
536
    int64_t get_segment_file_size(const RowsetMetaSharedPtr& rs_meta);
537
    int64_t get_inverted_index_file_size(const RowsetMetaSharedPtr& rs_meta);
538
539
public:
540
    static const int64_t K_INVALID_CUMULATIVE_POINT = -1;
541
542
private:
543
    StorageEngine& _engine;
544
    DataDir* _data_dir = nullptr;
545
546
    std::string _tablet_path;
547
548
    DorisCallOnce<Status> _init_once;
549
    // meta store lock is used for prevent 2 threads do checkpoint concurrently
550
    // it will be used in econ-mode in the future
551
    std::shared_mutex _meta_store_lock;
552
    std::mutex _ingest_lock;
553
    std::mutex _base_compaction_lock;
554
    std::mutex _cumulative_compaction_lock;
555
    std::shared_timed_mutex _migration_lock;
556
    std::mutex _build_inverted_index_lock;
557
558
    // In unique key table with MoW, we should guarantee that only one
559
    // writer can update rowset and delete bitmap at the same time.
560
    // We use a separate lock rather than _meta_lock, to avoid blocking read queries
561
    // during publish_txn, which might take hundreds of milliseconds
562
    mutable std::mutex _rowset_update_lock;
563
564
    // if this tablet is broken, set to true. default is false
565
    std::atomic<bool> _is_bad;
566
    // timestamp of last cumu compaction failure
567
    std::atomic<int64_t> _last_cumu_compaction_failure_millis;
568
    // timestamp of last base compaction failure
569
    std::atomic<int64_t> _last_base_compaction_failure_millis;
570
    // timestamp of last full compaction failure
571
    std::atomic<int64_t> _last_full_compaction_failure_millis;
572
    // timestamp of last cumu compaction success
573
    std::atomic<int64_t> _last_cumu_compaction_success_millis;
574
    // timestamp of last base compaction success
575
    std::atomic<int64_t> _last_base_compaction_success_millis;
576
    // timestamp of last full compaction success
577
    std::atomic<int64_t> _last_full_compaction_success_millis;
578
    // timestamp of last cumu compaction schedule time
579
    std::atomic<int64_t> _last_cumu_compaction_schedule_millis;
580
    // timestamp of last base compaction schedule time
581
    std::atomic<int64_t> _last_base_compaction_schedule_millis;
582
    // timestamp of last full compaction schedule time
583
    std::atomic<int64_t> _last_full_compaction_schedule_millis;
584
    std::atomic<int64_t> _cumulative_point;
585
    std::atomic<int64_t> _cumulative_promotion_size;
586
    std::atomic<int32_t> _newly_created_rowset_num;
587
    std::atomic<int64_t> _last_checkpoint_time;
588
    std::string _last_cumu_compaction_status;
589
    std::string _last_base_compaction_status;
590
    std::string _last_full_compaction_status;
591
592
    // single replica compaction status
593
    std::string _last_single_compaction_failure_status;
594
    Version _last_fetched_version;
595
596
    // cumulative compaction policy
597
    std::shared_ptr<CumulativeCompactionPolicy> _cumulative_compaction_policy;
598
    std::string_view _cumulative_compaction_type;
599
600
    // use a separate thread to check all tablets paths existence
601
    std::atomic<bool> _is_tablet_path_exists;
602
603
    int64_t _last_missed_version;
604
    int64_t _last_missed_time_s;
605
606
    bool _skip_cumu_compaction = false;
607
    int64_t _skip_cumu_compaction_ts;
608
609
    bool _skip_base_compaction = false;
610
    int64_t _skip_base_compaction_ts;
611
612
    // cooldown related
613
    CooldownConf _cooldown_conf;
614
    // `_cooldown_conf_lock` is used to serialize update cooldown conf and all operations that:
615
    // 1. read cooldown conf
616
    // 2. upload rowsets to remote storage
617
    // 3. update cooldown meta id
618
    mutable std::shared_mutex _cooldown_conf_lock;
619
    // `_cold_compaction_lock` is used to serialize cold data compaction and all operations that
620
    // may delete compaction input rowsets.
621
    std::mutex _cold_compaction_lock;
622
    int64_t _last_failed_follow_cooldown_time = 0;
623
    // `_alter_failed` is used to indicate whether the tablet failed to perform a schema change
624
    std::atomic<bool> _alter_failed = false;
625
626
    int64_t _io_error_times = 0;
627
628
    // partition's visible version. it sync from fe, but not real-time.
629
    std::shared_ptr<const VersionWithTime> _visible_version;
630
631
    std::atomic_bool _is_full_compaction_running = false;
632
};
633
634
21
inline CumulativeCompactionPolicy* Tablet::cumulative_compaction_policy() {
635
21
    return _cumulative_compaction_policy.get();
636
21
}
637
638
285
inline bool Tablet::init_succeeded() {
639
285
    return _init_once.has_called() && _init_once.stored_result().ok();
640
285
}
641
642
262
inline bool Tablet::is_used() {
643
262
    return !_is_bad && _data_dir->is_used();
644
262
}
645
646
304
inline void Tablet::register_tablet_into_dir() {
647
304
    _data_dir->register_tablet(this);
648
304
}
649
650
255
inline void Tablet::deregister_tablet_from_dir() {
651
255
    _data_dir->deregister_tablet(this);
652
255
}
653
654
290
inline int64_t Tablet::cumulative_layer_point() const {
655
290
    return _cumulative_point;
656
290
}
657
658
329
inline void Tablet::set_cumulative_layer_point(int64_t new_point) {
659
    // cumulative point should only be reset to -1, or be increased
660
329
    CHECK(new_point == Tablet::K_INVALID_CUMULATIVE_POINT || new_point >= _cumulative_point)
661
0
            << "Unexpected cumulative point: " << new_point
662
0
            << ", origin: " << _cumulative_point.load();
663
329
    _cumulative_point = new_point;
664
329
}
665
666
31
inline int64_t Tablet::cumulative_promotion_size() const {
667
31
    return _cumulative_promotion_size;
668
31
}
669
670
299
inline void Tablet::set_cumulative_promotion_size(int64_t new_size) {
671
299
    _cumulative_promotion_size = new_size;
672
299
}
673
674
// TODO(lingbin): Why other methods that need to get information from _tablet_meta
675
// are not locked, here needs a comment to explain.
676
8
inline size_t Tablet::tablet_footprint() {
677
8
    std::shared_lock rdlock(_meta_lock);
678
8
    return _tablet_meta->tablet_footprint();
679
8
}
680
681
0
inline size_t Tablet::tablet_local_size() {
682
0
    std::shared_lock rdlock(_meta_lock);
683
0
    return _tablet_meta->tablet_local_size();
684
0
}
685
686
0
inline size_t Tablet::tablet_remote_size() {
687
0
    std::shared_lock rdlock(_meta_lock);
688
0
    return _tablet_meta->tablet_remote_size();
689
0
}
690
691
// TODO(lingbin): Why other methods which need to get information from _tablet_meta
692
// are not locked, here needs a comment to explain.
693
19
inline size_t Tablet::num_rows() {
694
19
    std::shared_lock rdlock(_meta_lock);
695
19
    return _tablet_meta->num_rows();
696
19
}
697
698
28
inline int Tablet::version_count() const {
699
28
    std::shared_lock rdlock(_meta_lock);
700
28
    return _tablet_meta->version_count();
701
28
}
702
703
28
inline int Tablet::stale_version_count() const {
704
28
    std::shared_lock rdlock(_meta_lock);
705
28
    return _tablet_meta->stale_version_count();
706
28
}
707
708
46
inline Version Tablet::max_version() const {
709
46
    std::shared_lock rdlock(_meta_lock);
710
46
    return _tablet_meta->max_version();
711
46
}
712
713
0
inline uint64_t Tablet::segment_count() const {
714
0
    std::shared_lock rdlock(_meta_lock);
715
0
    uint64_t segment_nums = 0;
716
0
    for (const auto& rs_meta : _tablet_meta->all_rs_metas()) {
717
0
        segment_nums += rs_meta->num_segments();
718
0
    }
719
0
    return segment_nums;
720
0
}
721
722
0
inline SortType Tablet::sort_type() const {
723
0
    return _tablet_meta->tablet_schema()->sort_type();
724
0
}
725
726
0
inline size_t Tablet::sort_col_num() const {
727
0
    return _tablet_meta->tablet_schema()->sort_col_num();
728
0
}
729
730
0
inline size_t Tablet::num_columns() const {
731
0
    return _tablet_meta->tablet_schema()->num_columns();
732
0
}
733
734
0
inline size_t Tablet::num_null_columns() const {
735
0
    return _tablet_meta->tablet_schema()->num_null_columns();
736
0
}
737
738
0
inline size_t Tablet::num_short_key_columns() const {
739
0
    return _tablet_meta->tablet_schema()->num_short_key_columns();
740
0
}
741
742
0
inline size_t Tablet::num_rows_per_row_block() const {
743
0
    return _tablet_meta->tablet_schema()->num_rows_per_row_block();
744
0
}
745
746
0
inline CompressKind Tablet::compress_kind() const {
747
0
    return _tablet_meta->tablet_schema()->compress_kind();
748
0
}
749
750
0
inline double Tablet::bloom_filter_fpp() const {
751
0
    return _tablet_meta->tablet_schema()->bloom_filter_fpp();
752
0
}
753
754
0
inline size_t Tablet::next_unique_id() const {
755
0
    return _tablet_meta->tablet_schema()->next_column_unique_id();
756
0
}
757
758
0
inline size_t Tablet::row_size() const {
759
0
    return _tablet_meta->tablet_schema()->row_size();
760
0
}
761
762
28
inline int64_t Tablet::avg_rs_meta_serialize_size() const {
763
28
    return _tablet_meta->avg_rs_meta_serialize_size();
764
28
}
765
766
} // namespace doris