Coverage Report

Created: 2025-06-20 19:47

/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 <shared_mutex>
22
#include <string>
23
24
#include "common/status.h"
25
#include "olap/iterators.h"
26
#include "olap/olap_common.h"
27
#include "olap/rowset/segment_v2/segment.h"
28
#include "olap/tablet_fwd.h"
29
#include "olap/tablet_meta.h"
30
#include "olap/tablet_schema.h"
31
#include "olap/version_graph.h"
32
#include "util/metrics.h"
33
34
namespace doris {
35
struct RowSetSplits;
36
struct RowsetWriterContext;
37
class RowsetWriter;
38
class CalcDeleteBitmapToken;
39
class SegmentCacheHandle;
40
class RowIdConversion;
41
struct PartialUpdateInfo;
42
class PartialUpdateReadPlan;
43
44
struct TabletWithVersion {
45
    BaseTabletSPtr tablet;
46
    int64_t version;
47
};
48
49
enum class CompactionStage { NOT_SCHEDULED, PENDING, EXECUTING };
50
51
// Base class for all tablet classes
52
class BaseTablet {
53
public:
54
    explicit BaseTablet(TabletMetaSharedPtr tablet_meta);
55
    virtual ~BaseTablet();
56
    BaseTablet(const BaseTablet&) = delete;
57
    BaseTablet& operator=(const BaseTablet&) = delete;
58
59
9.30k
    TabletState tablet_state() const { return _tablet_meta->tablet_state(); }
60
    Status set_tablet_state(TabletState state);
61
3
    int64_t table_id() const { return _tablet_meta->table_id(); }
62
0
    int64_t index_id() const { return _tablet_meta->index_id(); }
63
947
    int64_t partition_id() const { return _tablet_meta->partition_id(); }
64
17.9k
    int64_t tablet_id() const { return _tablet_meta->tablet_id(); }
65
1.48k
    int32_t schema_hash() const { return _tablet_meta->schema_hash(); }
66
1.24k
    KeysType keys_type() const { return _tablet_meta->tablet_schema()->keys_type(); }
67
128
    size_t num_key_columns() const { return _tablet_meta->tablet_schema()->num_key_columns(); }
68
331
    int64_t ttl_seconds() const { return _tablet_meta->ttl_seconds(); }
69
18
    std::mutex& get_schema_change_lock() { return _schema_change_lock; }
70
1.17k
    bool enable_unique_key_merge_on_write() const {
71
1.17k
#ifdef BE_TEST
72
1.17k
        if (_tablet_meta == nullptr) {
73
0
            return false;
74
0
        }
75
1.17k
#endif
76
1.17k
        return _tablet_meta->enable_unique_key_merge_on_write();
77
1.17k
    }
78
79
    // Property encapsulated in TabletMeta
80
2.63k
    const TabletMetaSharedPtr& tablet_meta() { return _tablet_meta; }
81
82
    // FIXME(plat1ko): It is not appropriate to expose this lock
83
66
    std::shared_mutex& get_header_lock() { return _meta_lock; }
84
85
    void update_max_version_schema(const TabletSchemaSPtr& tablet_schema);
86
87
    Status update_by_least_common_schema(const TabletSchemaSPtr& update_schema);
88
89
11.5k
    TabletSchemaSPtr tablet_schema() const {
90
11.5k
        std::shared_lock rlock(_meta_lock);
91
11.5k
        return _max_version_schema;
92
11.5k
    }
93
94
    virtual bool exceed_version_limit(int32_t limit) = 0;
95
96
    virtual Result<std::unique_ptr<RowsetWriter>> create_rowset_writer(RowsetWriterContext& context,
97
                                                                       bool vertical) = 0;
98
99
    virtual Status capture_consistent_rowsets_unlocked(
100
            const Version& spec_version, std::vector<RowsetSharedPtr>* rowsets) const = 0;
101
102
    virtual Status capture_rs_readers(const Version& spec_version,
103
                                      std::vector<RowSetSplits>* rs_splits,
104
                                      bool skip_missing_version) = 0;
105
106
    virtual size_t tablet_footprint() = 0;
107
108
    // this method just return the compaction sum on each rowset
109
    // note(tsy): we should unify the compaction score calculation finally
110
    uint32_t get_real_compaction_score() const;
111
112
    // MUST hold shared meta lock
113
    Status capture_rs_readers_unlocked(const Versions& version_path,
114
                                       std::vector<RowSetSplits>* rs_splits) const;
115
116
    // _rs_version_map and _stale_rs_version_map should be protected by _meta_lock
117
    // The caller must call hold _meta_lock when call this three function.
118
    RowsetSharedPtr get_rowset_by_version(const Version& version, bool find_is_stale = false) const;
119
    RowsetSharedPtr get_stale_rowset_by_version(const Version& version) const;
120
    RowsetSharedPtr get_rowset_with_max_version() const;
121
122
    Status get_all_rs_id(int64_t max_version, RowsetIdUnorderedSet* rowset_ids) const;
123
    Status get_all_rs_id_unlocked(int64_t max_version, RowsetIdUnorderedSet* rowset_ids) const;
124
125
    // Get the missed versions until the spec_version.
126
    Versions get_missed_versions(int64_t spec_version) const;
127
    Versions get_missed_versions_unlocked(int64_t spec_version) const;
128
129
    void generate_tablet_meta_copy(TabletMeta& new_tablet_meta) const;
130
    void generate_tablet_meta_copy_unlocked(TabletMeta& new_tablet_meta) const;
131
132
36
    virtual int64_t max_version_unlocked() const { return _tablet_meta->max_version().second; }
133
134
    static TabletSchemaSPtr tablet_schema_with_merged_max_schema_version(
135
            const std::vector<RowsetMetaSharedPtr>& rowset_metas);
136
137
    ////////////////////////////////////////////////////////////////////////////
138
    // begin MoW functions
139
    ////////////////////////////////////////////////////////////////////////////
140
    std::vector<RowsetSharedPtr> get_rowset_by_ids(
141
            const RowsetIdUnorderedSet* specified_rowset_ids);
142
143
    // Lookup a row with TupleDescriptor and fill Block
144
    Status lookup_row_data(const Slice& encoded_key, const RowLocation& row_location,
145
                           RowsetSharedPtr rowset, const TupleDescriptor* desc,
146
                           OlapReaderStatistics& stats, std::string& values,
147
                           bool write_to_cache = false);
148
    // Lookup the row location of `encoded_key`, the function sets `row_location` on success.
149
    // NOTE: the method only works in unique key model with primary key index, you will got a
150
    //       not supported error in other data model.
151
    Status lookup_row_key(const Slice& encoded_key, TabletSchema* latest_schema, bool with_seq_col,
152
                          const std::vector<RowsetSharedPtr>& specified_rowsets,
153
                          RowLocation* row_location, uint32_t version,
154
                          std::vector<std::unique_ptr<SegmentCacheHandle>>& segment_caches,
155
                          RowsetSharedPtr* rowset = nullptr, bool with_rowid = true,
156
                          OlapReaderStatistics* stats = nullptr);
157
158
    // calc delete bitmap when flush memtable, use a fake version to calc
159
    // For example, cur max version is 5, and we use version 6 to calc but
160
    // finally this rowset publish version with 8, we should make up data
161
    // for rowset 6-7. Also, if a compaction happens between commit_txn and
162
    // publish_txn, we should remove compaction input rowsets' delete_bitmap
163
    // and build newly generated rowset's delete_bitmap
164
    static Status calc_delete_bitmap(const BaseTabletSPtr& tablet, RowsetSharedPtr rowset,
165
                                     const std::vector<segment_v2::SegmentSharedPtr>& segments,
166
                                     const std::vector<RowsetSharedPtr>& specified_rowsets,
167
                                     DeleteBitmapPtr delete_bitmap, int64_t version,
168
                                     CalcDeleteBitmapToken* token,
169
                                     RowsetWriter* rowset_writer = nullptr);
170
171
    Status calc_segment_delete_bitmap(RowsetSharedPtr rowset,
172
                                      const segment_v2::SegmentSharedPtr& seg,
173
                                      const std::vector<RowsetSharedPtr>& specified_rowsets,
174
                                      DeleteBitmapPtr delete_bitmap, int64_t end_version,
175
                                      RowsetWriter* rowset_writer);
176
177
    Status calc_delete_bitmap_between_segments(
178
            const RowsetId& rowset_id, const std::vector<segment_v2::SegmentSharedPtr>& segments,
179
            DeleteBitmapPtr delete_bitmap);
180
181
    static Status commit_phase_update_delete_bitmap(
182
            const BaseTabletSPtr& tablet, const RowsetSharedPtr& rowset,
183
            RowsetIdUnorderedSet& pre_rowset_ids, DeleteBitmapPtr delete_bitmap,
184
            const std::vector<segment_v2::SegmentSharedPtr>& segments, int64_t txn_id,
185
            CalcDeleteBitmapToken* token, RowsetWriter* rowset_writer = nullptr);
186
187
    static void add_sentinel_mark_to_delete_bitmap(DeleteBitmap* delete_bitmap,
188
                                                   const RowsetIdUnorderedSet& rowsetids);
189
190
    Status check_delete_bitmap_correctness(DeleteBitmapPtr delete_bitmap, int64_t max_version,
191
                                           int64_t txn_id, const RowsetIdUnorderedSet& rowset_ids,
192
                                           std::vector<RowsetSharedPtr>* rowsets = nullptr);
193
194
    static const signed char* get_delete_sign_column_data(vectorized::Block& block,
195
                                                          size_t rows_at_least = 0);
196
197
    static Status generate_default_value_block(const TabletSchema& schema,
198
                                               const std::vector<uint32_t>& cids,
199
                                               const std::vector<std::string>& default_values,
200
                                               const vectorized::Block& ref_block,
201
                                               vectorized::Block& default_value_block);
202
203
    static Status generate_new_block_for_partial_update(
204
            TabletSchemaSPtr rowset_schema, const PartialUpdateInfo* partial_update_info,
205
            const PartialUpdateReadPlan& read_plan_ori,
206
            const PartialUpdateReadPlan& read_plan_update,
207
            const std::map<RowsetId, RowsetSharedPtr>& rsid_to_rowset,
208
            vectorized::Block* output_block);
209
210
    // We use the TabletSchema from the caller because the TabletSchema in the rowset'meta
211
    // may be outdated due to schema change. Also note that the the cids should indicate the indexes
212
    // of the columns in the TabletSchema passed in.
213
    static Status fetch_value_through_row_column(RowsetSharedPtr input_rowset,
214
                                                 const TabletSchema& tablet_schema, uint32_t segid,
215
                                                 const std::vector<uint32_t>& rowids,
216
                                                 const std::vector<uint32_t>& cids,
217
                                                 vectorized::Block& block);
218
219
    static Status fetch_value_by_rowids(RowsetSharedPtr input_rowset, uint32_t segid,
220
                                        const std::vector<uint32_t>& rowids,
221
                                        const TabletColumn& tablet_column,
222
                                        vectorized::MutableColumnPtr& dst);
223
224
    virtual Result<std::unique_ptr<RowsetWriter>> create_transient_rowset_writer(
225
            const Rowset& rowset, std::shared_ptr<PartialUpdateInfo> partial_update_info,
226
            int64_t txn_expiration = 0) = 0;
227
228
    static Status update_delete_bitmap(const BaseTabletSPtr& self, TabletTxnInfo* txn_info,
229
                                       int64_t txn_id, int64_t txn_expiration = 0);
230
    virtual Status save_delete_bitmap(const TabletTxnInfo* txn_info, int64_t txn_id,
231
                                      DeleteBitmapPtr delete_bitmap, RowsetWriter* rowset_writer,
232
                                      const RowsetIdUnorderedSet& cur_rowset_ids,
233
                                      int64_t next_visible_version = -1) = 0;
234
    virtual CalcDeleteBitmapExecutor* calc_delete_bitmap_executor() = 0;
235
236
    void calc_compaction_output_rowset_delete_bitmap(
237
            const std::vector<RowsetSharedPtr>& input_rowsets,
238
            const RowIdConversion& rowid_conversion, uint64_t start_version, uint64_t end_version,
239
            std::set<RowLocation>* missed_rows,
240
            std::map<RowsetSharedPtr, std::list<std::pair<RowLocation, RowLocation>>>* location_map,
241
            const DeleteBitmap& input_delete_bitmap, DeleteBitmap* output_rowset_delete_bitmap);
242
243
    Status check_rowid_conversion(
244
            RowsetSharedPtr dst_rowset,
245
            const std::map<RowsetSharedPtr, std::list<std::pair<RowLocation, RowLocation>>>&
246
                    location_map);
247
248
    static Status update_delete_bitmap_without_lock(
249
            const BaseTabletSPtr& self, const RowsetSharedPtr& rowset,
250
            const std::vector<RowsetSharedPtr>* specified_base_rowsets = nullptr);
251
252
    ////////////////////////////////////////////////////////////////////////////
253
    // end MoW functions
254
    ////////////////////////////////////////////////////////////////////////////
255
256
    RowsetSharedPtr get_rowset(const RowsetId& rowset_id);
257
258
    std::vector<RowsetSharedPtr> get_snapshot_rowset(bool include_stale_rowset = false) const;
259
260
    virtual void clear_cache() = 0;
261
262
    // Find the first consecutive empty rowsets. output->size() >= limit
263
    void calc_consecutive_empty_rowsets(std::vector<RowsetSharedPtr>* empty_rowsets,
264
                                        const std::vector<RowsetSharedPtr>& candidate_rowsets,
265
                                        int64_t limit);
266
267
    // Return the merged schema of all rowsets
268
0
    virtual TabletSchemaSPtr merged_tablet_schema() const {
269
0
        std::shared_lock rlock(_meta_lock);
270
0
        return _max_version_schema;
271
0
    }
272
273
    void traverse_rowsets(std::function<void(const RowsetSharedPtr&)> visitor,
274
11
                          bool include_stale = false) {
275
11
        std::shared_lock rlock(_meta_lock);
276
62
        for (auto& [v, rs] : _rs_version_map) {
277
62
            visitor(rs);
278
62
        }
279
11
        if (!include_stale) return;
280
10
        for (auto& [v, rs] : _stale_rs_version_map) {
281
0
            visitor(rs);
282
0
        }
283
10
    }
284
285
    Status calc_file_crc(uint32_t* crc_value, int64_t start_version, int64_t end_version,
286
                         int32_t* rowset_count, int64_t* file_count);
287
288
    Status show_nested_index_file(std::string* json_meta);
289
290
12.4k
    TabletUid tablet_uid() const { return _tablet_meta->tablet_uid(); }
291
561
    TabletInfo get_tablet_info() const { return TabletInfo(tablet_id(), tablet_uid()); }
292
293
    void get_base_rowset_delete_bitmap_count(
294
            uint64_t* max_base_rowset_delete_bitmap_score,
295
            int64_t* max_base_rowset_delete_bitmap_score_tablet_id);
296
297
3
    virtual Status check_delete_bitmap_cache(int64_t txn_id, DeleteBitmap* expected_delete_bitmap) {
298
3
        return Status::OK();
299
3
    }
300
301
protected:
302
    // Find the missed versions until the spec_version.
303
    //
304
    // for example:
305
    //     [0-4][5-5][8-8][9-9][14-14]
306
    // for cloud, if spec_version = 12, it will return [6-7],[10-12]
307
    // for local, if spec_version = 12, it will return [6, 6], [7, 7], [10, 10], [11, 11], [12, 12]
308
    virtual Versions calc_missed_versions(int64_t spec_version,
309
                                          Versions existing_versions) const = 0;
310
311
    void _print_missed_versions(const Versions& missed_versions) const;
312
    bool _reconstruct_version_tracker_if_necessary();
313
314
    static void _rowset_ids_difference(const RowsetIdUnorderedSet& cur,
315
                                       const RowsetIdUnorderedSet& pre,
316
                                       RowsetIdUnorderedSet* to_add, RowsetIdUnorderedSet* to_del);
317
318
    Status _capture_consistent_rowsets_unlocked(const std::vector<Version>& version_path,
319
                                                std::vector<RowsetSharedPtr>* rowsets) const;
320
321
    Status sort_block(vectorized::Block& in_block, vectorized::Block& output_block);
322
323
    mutable std::shared_mutex _meta_lock;
324
    TimestampedVersionTracker _timestamped_version_tracker;
325
    // After version 0.13, all newly created rowsets are saved in _rs_version_map.
326
    // And if rowset being compacted, the old rowsets will be saved in _stale_rs_version_map;
327
    std::unordered_map<Version, RowsetSharedPtr, HashOfVersion> _rs_version_map;
328
    // This variable _stale_rs_version_map is used to record these rowsets which are be compacted.
329
    // These _stale rowsets are been removed when rowsets' pathVersion is expired,
330
    // this policy is judged and computed by TimestampedVersionTracker.
331
    std::unordered_map<Version, RowsetSharedPtr, HashOfVersion> _stale_rs_version_map;
332
    const TabletMetaSharedPtr _tablet_meta;
333
    TabletSchemaSPtr _max_version_schema;
334
335
    // metrics of this tablet
336
    std::shared_ptr<MetricEntity> _metric_entity;
337
338
protected:
339
    std::mutex _schema_change_lock;
340
341
public:
342
    IntCounter* query_scan_bytes = nullptr;
343
    IntCounter* query_scan_rows = nullptr;
344
    IntCounter* query_scan_count = nullptr;
345
    IntCounter* flush_bytes = nullptr;
346
    IntCounter* flush_finish_count = nullptr;
347
    std::atomic<int64_t> published_count = 0;
348
    std::atomic<int64_t> read_block_count = 0;
349
    std::atomic<int64_t> write_count = 0;
350
    std::atomic<int64_t> compaction_count = 0;
351
352
    CompactionStage compaction_stage = CompactionStage::NOT_SCHEDULED;
353
    std::mutex sample_info_lock;
354
    std::vector<CompactionSampleInfo> sample_infos;
355
    Status last_compaction_status = Status::OK();
356
};
357
358
} /* namespace doris */