Coverage Report

Created: 2025-06-03 11:52

/root/doris/be/src/olap/base_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 <memory>
21
#include <mutex>
22
#include <shared_mutex>
23
#include <string>
24
25
#include "common/status.h"
26
#include "olap/iterators.h"
27
#include "olap/olap_common.h"
28
#include "olap/partial_update_info.h"
29
#include "olap/rowset/segment_v2/segment.h"
30
#include "olap/tablet_fwd.h"
31
#include "olap/tablet_meta.h"
32
#include "olap/tablet_schema.h"
33
#include "olap/version_graph.h"
34
#include "util/metrics.h"
35
36
namespace doris {
37
struct RowSetSplits;
38
struct RowsetWriterContext;
39
class RowsetWriter;
40
class CalcDeleteBitmapToken;
41
class SegmentCacheHandle;
42
class RowIdConversion;
43
struct PartialUpdateInfo;
44
class FixedReadPlan;
45
46
struct TabletWithVersion {
47
    BaseTabletSPtr tablet;
48
    int64_t version;
49
};
50
51
enum class CompactionStage { NOT_SCHEDULED, PENDING, EXECUTING };
52
53
// Base class for all tablet classes
54
class BaseTablet {
55
public:
56
    explicit BaseTablet(TabletMetaSharedPtr tablet_meta);
57
    virtual ~BaseTablet();
58
    BaseTablet(const BaseTablet&) = delete;
59
    BaseTablet& operator=(const BaseTablet&) = delete;
60
61
9.27k
    TabletState tablet_state() const { return _tablet_meta->tablet_state(); }
62
    Status set_tablet_state(TabletState state);
63
3
    int64_t table_id() const { return _tablet_meta->table_id(); }
64
0
    int64_t index_id() const { return _tablet_meta->index_id(); }
65
943
    int64_t partition_id() const { return _tablet_meta->partition_id(); }
66
17.9k
    int64_t tablet_id() const { return _tablet_meta->tablet_id(); }
67
1.47k
    int32_t schema_hash() const { return _tablet_meta->schema_hash(); }
68
1.23k
    KeysType keys_type() const { return _tablet_meta->tablet_schema()->keys_type(); }
69
128
    size_t num_key_columns() const { return _tablet_meta->tablet_schema()->num_key_columns(); }
70
331
    int64_t ttl_seconds() const { return _tablet_meta->ttl_seconds(); }
71
    // currently used by schema change, inverted index building, and cooldown
72
18
    std::timed_mutex& get_schema_change_lock() { return _schema_change_lock; }
73
1.17k
    bool enable_unique_key_merge_on_write() const {
74
1.17k
#ifdef BE_TEST
75
1.17k
        if (_tablet_meta == nullptr) {
76
0
            return false;
77
0
        }
78
1.17k
#endif
79
1.17k
        return _tablet_meta->enable_unique_key_merge_on_write();
80
1.17k
    }
81
82
    // Property encapsulated in TabletMeta
83
2.40k
    const TabletMetaSharedPtr& tablet_meta() { return _tablet_meta; }
84
85
    int32 max_version_config();
86
87
    // FIXME(plat1ko): It is not appropriate to expose this lock
88
66
    std::shared_mutex& get_header_lock() { return _meta_lock; }
89
90
    void update_max_version_schema(const TabletSchemaSPtr& tablet_schema);
91
92
    Status update_by_least_common_schema(const TabletSchemaSPtr& update_schema);
93
94
11.5k
    TabletSchemaSPtr tablet_schema() const {
95
11.5k
        std::shared_lock rlock(_meta_lock);
96
11.5k
        return _max_version_schema;
97
11.5k
    }
98
99
    virtual bool exceed_version_limit(int32_t limit) = 0;
100
101
    virtual Result<std::unique_ptr<RowsetWriter>> create_rowset_writer(RowsetWriterContext& context,
102
                                                                       bool vertical) = 0;
103
104
    virtual Status capture_consistent_rowsets_unlocked(
105
            const Version& spec_version, std::vector<RowsetSharedPtr>* rowsets) const = 0;
106
107
    virtual Status capture_rs_readers(const Version& spec_version,
108
                                      std::vector<RowSetSplits>* rs_splits,
109
                                      bool skip_missing_version) = 0;
110
111
    virtual size_t tablet_footprint() = 0;
112
113
    // this method just return the compaction sum on each rowset
114
    // note(tsy): we should unify the compaction score calculation finally
115
    uint32_t get_real_compaction_score() const;
116
117
    // MUST hold shared meta lock
118
    Status capture_rs_readers_unlocked(const Versions& version_path,
119
                                       std::vector<RowSetSplits>* rs_splits) const;
120
121
    // _rs_version_map and _stale_rs_version_map should be protected by _meta_lock
122
    // The caller must call hold _meta_lock when call this three function.
123
    RowsetSharedPtr get_rowset_by_version(const Version& version, bool find_is_stale = false) const;
124
    RowsetSharedPtr get_stale_rowset_by_version(const Version& version) const;
125
    RowsetSharedPtr get_rowset_with_max_version() const;
126
127
    Status get_all_rs_id(int64_t max_version, RowsetIdUnorderedSet* rowset_ids) const;
128
    Status get_all_rs_id_unlocked(int64_t max_version, RowsetIdUnorderedSet* rowset_ids) const;
129
130
    // Get the missed versions until the spec_version.
131
    Versions get_missed_versions(int64_t spec_version) const;
132
    Versions get_missed_versions_unlocked(int64_t spec_version) const;
133
134
    void generate_tablet_meta_copy(TabletMeta& new_tablet_meta) const;
135
    void generate_tablet_meta_copy_unlocked(TabletMeta& new_tablet_meta) const;
136
137
36
    virtual int64_t max_version_unlocked() const { return _tablet_meta->max_version().second; }
138
139
    static TabletSchemaSPtr tablet_schema_with_merged_max_schema_version(
140
            const std::vector<RowsetMetaSharedPtr>& rowset_metas);
141
142
    ////////////////////////////////////////////////////////////////////////////
143
    // begin MoW functions
144
    ////////////////////////////////////////////////////////////////////////////
145
    std::vector<RowsetSharedPtr> get_rowset_by_ids(
146
            const RowsetIdUnorderedSet* specified_rowset_ids);
147
148
    // Lookup a row with TupleDescriptor and fill Block
149
    Status lookup_row_data(const Slice& encoded_key, const RowLocation& row_location,
150
                           RowsetSharedPtr rowset, OlapReaderStatistics& stats, std::string& values,
151
                           bool write_to_cache = false);
152
153
    // Lookup the row location of `encoded_key`, the function sets `row_location` on success.
154
    // NOTE: the method only works in unique key model with primary key index, you will got a
155
    //       not supported error in other data model.
156
    Status lookup_row_key(const Slice& encoded_key, TabletSchema* latest_schema, bool with_seq_col,
157
                          const std::vector<RowsetSharedPtr>& specified_rowsets,
158
                          RowLocation* row_location, uint32_t version,
159
                          std::vector<std::unique_ptr<SegmentCacheHandle>>& segment_caches,
160
                          RowsetSharedPtr* rowset = nullptr, bool with_rowid = true,
161
                          std::string* encoded_seq_value = nullptr,
162
                          OlapReaderStatistics* stats = nullptr,
163
                          DeleteBitmapPtr tablet_delete_bitmap = nullptr);
164
165
    // calc delete bitmap when flush memtable, use a fake version to calc
166
    // For example, cur max version is 5, and we use version 6 to calc but
167
    // finally this rowset publish version with 8, we should make up data
168
    // for rowset 6-7. Also, if a compaction happens between commit_txn and
169
    // publish_txn, we should remove compaction input rowsets' delete_bitmap
170
    // and build newly generated rowset's delete_bitmap
171
    static Status calc_delete_bitmap(const BaseTabletSPtr& tablet, RowsetSharedPtr rowset,
172
                                     const std::vector<segment_v2::SegmentSharedPtr>& segments,
173
                                     const std::vector<RowsetSharedPtr>& specified_rowsets,
174
                                     DeleteBitmapPtr delete_bitmap, int64_t version,
175
                                     CalcDeleteBitmapToken* token,
176
                                     RowsetWriter* rowset_writer = nullptr,
177
                                     DeleteBitmapPtr tablet_delete_bitmap = nullptr);
178
179
    Status calc_segment_delete_bitmap(RowsetSharedPtr rowset,
180
                                      const segment_v2::SegmentSharedPtr& seg,
181
                                      const std::vector<RowsetSharedPtr>& specified_rowsets,
182
                                      DeleteBitmapPtr delete_bitmap, int64_t end_version,
183
                                      RowsetWriter* rowset_writer,
184
                                      DeleteBitmapPtr tablet_delete_bitmap = nullptr);
185
186
    Status calc_delete_bitmap_between_segments(
187
            RowsetId rowset_id, const std::vector<segment_v2::SegmentSharedPtr>& segments,
188
            DeleteBitmapPtr delete_bitmap);
189
190
    static Status commit_phase_update_delete_bitmap(
191
            const BaseTabletSPtr& tablet, const RowsetSharedPtr& rowset,
192
            RowsetIdUnorderedSet& pre_rowset_ids, DeleteBitmapPtr delete_bitmap,
193
            const std::vector<segment_v2::SegmentSharedPtr>& segments, int64_t txn_id,
194
            CalcDeleteBitmapToken* token, RowsetWriter* rowset_writer = nullptr);
195
196
    static void add_sentinel_mark_to_delete_bitmap(DeleteBitmap* delete_bitmap,
197
                                                   const RowsetIdUnorderedSet& rowsetids);
198
199
    Status check_delete_bitmap_correctness(DeleteBitmapPtr delete_bitmap, int64_t max_version,
200
                                           int64_t txn_id, const RowsetIdUnorderedSet& rowset_ids,
201
                                           std::vector<RowsetSharedPtr>* rowsets = nullptr);
202
203
    static const signed char* get_delete_sign_column_data(const vectorized::Block& block,
204
                                                          size_t rows_at_least = 0);
205
206
    static Status generate_default_value_block(const TabletSchema& schema,
207
                                               const std::vector<uint32_t>& cids,
208
                                               const std::vector<std::string>& default_values,
209
                                               const vectorized::Block& ref_block,
210
                                               vectorized::Block& default_value_block);
211
212
    static Status generate_new_block_for_partial_update(
213
            TabletSchemaSPtr rowset_schema, const PartialUpdateInfo* partial_update_info,
214
            const FixedReadPlan& read_plan_ori, const FixedReadPlan& read_plan_update,
215
            const std::map<RowsetId, RowsetSharedPtr>& rsid_to_rowset,
216
            vectorized::Block* output_block);
217
218
    static Status generate_new_block_for_flexible_partial_update(
219
            TabletSchemaSPtr rowset_schema, const PartialUpdateInfo* partial_update_info,
220
            std::set<uint32_t>& rids_be_overwritten, const FixedReadPlan& read_plan_ori,
221
            const FixedReadPlan& read_plan_update,
222
            const std::map<RowsetId, RowsetSharedPtr>& rsid_to_rowset,
223
            vectorized::Block* output_block);
224
225
    // We use the TabletSchema from the caller because the TabletSchema in the rowset'meta
226
    // may be outdated due to schema change. Also note that the the cids should indicate the indexes
227
    // of the columns in the TabletSchema passed in.
228
    static Status fetch_value_through_row_column(RowsetSharedPtr input_rowset,
229
                                                 const TabletSchema& tablet_schema, uint32_t segid,
230
                                                 const std::vector<uint32_t>& rowids,
231
                                                 const std::vector<uint32_t>& cids,
232
                                                 vectorized::Block& block);
233
234
    static Status fetch_value_by_rowids(RowsetSharedPtr input_rowset, uint32_t segid,
235
                                        const std::vector<uint32_t>& rowids,
236
                                        const TabletColumn& tablet_column,
237
                                        vectorized::MutableColumnPtr& dst);
238
239
    virtual Result<std::unique_ptr<RowsetWriter>> create_transient_rowset_writer(
240
            const Rowset& rowset, std::shared_ptr<PartialUpdateInfo> partial_update_info,
241
            int64_t txn_expiration = 0) = 0;
242
243
    static Status update_delete_bitmap(const BaseTabletSPtr& self, TabletTxnInfo* txn_info,
244
                                       int64_t txn_id, int64_t txn_expiration = 0,
245
                                       DeleteBitmapPtr tablet_delete_bitmap = nullptr);
246
    virtual Status save_delete_bitmap(const TabletTxnInfo* txn_info, int64_t txn_id,
247
                                      DeleteBitmapPtr delete_bitmap, RowsetWriter* rowset_writer,
248
                                      const RowsetIdUnorderedSet& cur_rowset_ids,
249
                                      int64_t lock_id = -1, int64_t next_visible_version = -1) = 0;
250
    virtual CalcDeleteBitmapExecutor* calc_delete_bitmap_executor() = 0;
251
252
    void calc_compaction_output_rowset_delete_bitmap(
253
            const std::vector<RowsetSharedPtr>& input_rowsets,
254
            const RowIdConversion& rowid_conversion, uint64_t start_version, uint64_t end_version,
255
            std::set<RowLocation>* missed_rows,
256
            std::map<RowsetSharedPtr, std::list<std::pair<RowLocation, RowLocation>>>* location_map,
257
            const DeleteBitmap& input_delete_bitmap, DeleteBitmap* output_rowset_delete_bitmap);
258
259
    Status check_rowid_conversion(
260
            RowsetSharedPtr dst_rowset,
261
            const std::map<RowsetSharedPtr, std::list<std::pair<RowLocation, RowLocation>>>&
262
                    location_map);
263
264
    static Status update_delete_bitmap_without_lock(
265
            const BaseTabletSPtr& self, const RowsetSharedPtr& rowset,
266
            const std::vector<RowsetSharedPtr>* specified_base_rowsets = nullptr);
267
268
    using DeleteBitmapKeyRanges =
269
            std::vector<std::tuple<DeleteBitmap::BitmapKey, DeleteBitmap::BitmapKey>>;
270
    void agg_delete_bitmap_for_stale_rowsets(
271
            Version version, DeleteBitmapKeyRanges& remove_delete_bitmap_key_ranges);
272
    void check_agg_delete_bitmap_for_stale_rowsets(int64_t& useless_rowset_count,
273
                                                   int64_t& useless_rowset_version_count);
274
    ////////////////////////////////////////////////////////////////////////////
275
    // end MoW functions
276
    ////////////////////////////////////////////////////////////////////////////
277
278
    RowsetSharedPtr get_rowset(const RowsetId& rowset_id);
279
280
    std::vector<RowsetSharedPtr> get_snapshot_rowset(bool include_stale_rowset = false) const;
281
282
    virtual void clear_cache() = 0;
283
284
    // Find the first consecutive empty rowsets. output->size() >= limit
285
    void calc_consecutive_empty_rowsets(std::vector<RowsetSharedPtr>* empty_rowsets,
286
                                        const std::vector<RowsetSharedPtr>& candidate_rowsets,
287
                                        int64_t limit);
288
289
    // Return the merged schema of all rowsets
290
0
    virtual TabletSchemaSPtr merged_tablet_schema() const {
291
0
        std::shared_lock rlock(_meta_lock);
292
0
        return _max_version_schema;
293
0
    }
294
295
    void traverse_rowsets(std::function<void(const RowsetSharedPtr&)> visitor,
296
11
                          bool include_stale = false) {
297
11
        std::shared_lock rlock(_meta_lock);
298
62
        for (auto& [v, rs] : _rs_version_map) {
299
62
            visitor(rs);
300
62
        }
301
11
        if (!include_stale) return;
302
10
        for (auto& [v, rs] : _stale_rs_version_map) {
303
0
            visitor(rs);
304
0
        }
305
10
    }
306
307
    Status calc_file_crc(uint32_t* crc_value, int64_t start_version, int64_t end_version,
308
                         uint32_t* rowset_count, int64_t* file_count);
309
310
    Status show_nested_index_file(std::string* json_meta);
311
312
12.4k
    TabletUid tablet_uid() const { return _tablet_meta->tablet_uid(); }
313
561
    TabletInfo get_tablet_info() const { return TabletInfo(tablet_id(), tablet_uid()); }
314
315
    void get_base_rowset_delete_bitmap_count(
316
            uint64_t* max_base_rowset_delete_bitmap_score,
317
            int64_t* max_base_rowset_delete_bitmap_score_tablet_id);
318
319
3
    virtual Status check_delete_bitmap_cache(int64_t txn_id, DeleteBitmap* expected_delete_bitmap) {
320
3
        return Status::OK();
321
3
    }
322
323
protected:
324
    // Find the missed versions until the spec_version.
325
    //
326
    // for example:
327
    //     [0-4][5-5][8-8][9-9][14-14]
328
    // for cloud, if spec_version = 12, it will return [6-7],[10-12]
329
    // for local, if spec_version = 12, it will return [6, 6], [7, 7], [10, 10], [11, 11], [12, 12]
330
    virtual Versions calc_missed_versions(int64_t spec_version,
331
                                          Versions existing_versions) const = 0;
332
333
    void _print_missed_versions(const Versions& missed_versions) const;
334
    bool _reconstruct_version_tracker_if_necessary();
335
336
    static void _rowset_ids_difference(const RowsetIdUnorderedSet& cur,
337
                                       const RowsetIdUnorderedSet& pre,
338
                                       RowsetIdUnorderedSet* to_add, RowsetIdUnorderedSet* to_del);
339
340
    Status _capture_consistent_rowsets_unlocked(const std::vector<Version>& version_path,
341
                                                std::vector<RowsetSharedPtr>* rowsets) const;
342
343
    Status sort_block(vectorized::Block& in_block, vectorized::Block& output_block);
344
345
    mutable std::shared_mutex _meta_lock;
346
    TimestampedVersionTracker _timestamped_version_tracker;
347
    // After version 0.13, all newly created rowsets are saved in _rs_version_map.
348
    // And if rowset being compacted, the old rowsets will be saved in _stale_rs_version_map;
349
    std::unordered_map<Version, RowsetSharedPtr, HashOfVersion> _rs_version_map;
350
    // This variable _stale_rs_version_map is used to record these rowsets which are be compacted.
351
    // These _stale rowsets are been removed when rowsets' pathVersion is expired,
352
    // this policy is judged and computed by TimestampedVersionTracker.
353
    std::unordered_map<Version, RowsetSharedPtr, HashOfVersion> _stale_rs_version_map;
354
    const TabletMetaSharedPtr _tablet_meta;
355
    TabletSchemaSPtr _max_version_schema;
356
357
    // metrics of this tablet
358
    std::shared_ptr<MetricEntity> _metric_entity;
359
360
protected:
361
    std::timed_mutex _schema_change_lock;
362
363
public:
364
    IntCounter* query_scan_bytes = nullptr;
365
    IntCounter* query_scan_rows = nullptr;
366
    IntCounter* query_scan_count = nullptr;
367
    IntCounter* flush_bytes = nullptr;
368
    IntCounter* flush_finish_count = nullptr;
369
    std::atomic<int64_t> published_count = 0;
370
    std::atomic<int64_t> read_block_count = 0;
371
    std::atomic<int64_t> write_count = 0;
372
    std::atomic<int64_t> compaction_count = 0;
373
374
    CompactionStage compaction_stage = CompactionStage::NOT_SCHEDULED;
375
    std::mutex sample_info_lock;
376
    std::vector<CompactionSampleInfo> sample_infos;
377
    Status last_compaction_status = Status::OK();
378
};
379
380
} /* namespace doris */