Coverage Report

Created: 2026-08-14 01:26

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
be/src/storage/segment/vertical_segment_writer.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 <gen_cpp/olap_file.pb.h>
21
#include <gen_cpp/segment_v2.pb.h>
22
23
#include <cstddef>
24
#include <cstdint>
25
#include <functional>
26
#include <map>
27
#include <memory> // unique_ptr
28
#include <string>
29
#include <unordered_set>
30
#include <vector>
31
32
#include "common/status.h" // Status
33
#include "storage/index/index_file_writer.h"
34
#include "storage/key/row_key_encoder.h"
35
#include "storage/olap_define.h"
36
#include "storage/partial_update_info.h"
37
#include "storage/segment/column_writer.h"
38
#include "storage/segment/segment_index_file_cache_loader.h"
39
#include "storage/tablet/tablet.h"
40
#include "storage/tablet/tablet_schema.h"
41
#include "util/faststring.h"
42
#include "util/slice.h"
43
44
namespace doris {
45
class Block;
46
class IOlapColumnDataAccessor;
47
class OlapBlockDataConvertor;
48
49
class DataDir;
50
class MemTracker;
51
class ShortKeyIndexBuilder;
52
class PrimaryKeyIndexBuilder;
53
class KeyCoder;
54
struct RowsetWriterContext;
55
56
namespace io {
57
class FileWriter;
58
class FileSystem;
59
} // namespace io
60
namespace segment_v2 {
61
class IndexFileWriter;
62
class MowKeyProbe;
63
64
struct VerticalSegmentWriterOptions {
65
    uint32_t num_rows_per_block = 1024;
66
    bool enable_unique_key_merge_on_write = false;
67
    CompressionTypePB compression_type = UNKNOWN_COMPRESSION;
68
69
    RowsetWriterContext* rowset_ctx = nullptr;
70
    DataWriteType write_type = DataWriteType::TYPE_DEFAULT;
71
    std::shared_ptr<MowContext> mow_ctx;
72
};
73
74
class DerivedColumnGenerator;
75
// Matches block_transform.h: at most one derived column (the row-store column)
76
// for each flush, held as a {cid, generator} pair; null generator means none.
77
using DerivedColumn = std::pair<uint32_t, std::shared_ptr<const DerivedColumnGenerator>>;
78
79
struct RowsInBlock {
80
    const Block* block;
81
    size_t row_pos;
82
    size_t num_rows;
83
};
84
85
class VerticalSegmentWriter {
86
public:
87
    explicit VerticalSegmentWriter(io::FileWriter* file_writer, uint32_t segment_id,
88
                                   TabletSchemaSPtr tablet_schema, BaseTabletSPtr tablet,
89
                                   DataDir* data_dir, const VerticalSegmentWriterOptions& opts,
90
                                   IndexFileWriter* index_file_writer);
91
    ~VerticalSegmentWriter();
92
93
    VerticalSegmentWriter(const VerticalSegmentWriter&) = delete;
94
    const VerticalSegmentWriter& operator=(const VerticalSegmentWriter&) = delete;
95
96
    Status init();
97
98
    // Add one block to batch, memory is owned by the caller.
99
    // The batched blocks will be flushed in write_batch.
100
    // Once write_batch is called, no more blocks shoud be added.
101
    Status batch_block(const Block* block, size_t row_pos, size_t num_rows);
102
    Status write_batch();
103
104
137
    void set_derived_column(DerivedColumn derived_column) {
105
137
        _derived_column = std::move(derived_column);
106
137
    }
107
108
0
    [[nodiscard]] std::string data_dir_path() const {
109
0
        return _data_dir == nullptr ? "" : _data_dir->path();
110
0
    }
111
112
137
    [[nodiscard]] uint32_t num_rows_written() const { return _num_rows_written; }
113
114
    // for partial update
115
137
    [[nodiscard]] int64_t num_rows_updated() const { return _num_rows_updated; }
116
137
    [[nodiscard]] int64_t num_rows_deleted() const { return _num_rows_deleted; }
117
137
    [[nodiscard]] int64_t num_rows_new_added() const { return _num_rows_new_added; }
118
137
    [[nodiscard]] int64_t num_rows_filtered() const { return _num_rows_filtered; }
119
0
    [[nodiscard]] uint32_t row_count() const { return _row_count; }
120
137
    [[nodiscard]] uint32_t segment_id() const { return _segment_id; }
121
122
    Status finalize(uint64_t* segment_file_size, uint64_t* index_size,
123
                    SegmentIndexFileCacheInfo* index_file_cache_info = nullptr);
124
125
    Status finalize_columns_index(uint64_t* index_size);
126
    Status finalize_footer(uint64_t* segment_file_size,
127
                           SegmentIndexFileCacheInfo* index_file_cache_info = nullptr);
128
129
    Slice min_encoded_key();
130
    Slice max_encoded_key();
131
132
    void clear();
133
134
137
    Status close_inverted_index(int64_t* inverted_index_file_size) {
135
        // no inverted index
136
137
        if (_index_file_writer == nullptr) {
137
115
            *inverted_index_file_size = 0;
138
115
            return Status::OK();
139
115
        }
140
22
        RETURN_IF_ERROR(_index_file_writer->begin_close());
141
22
        *inverted_index_file_size = _index_file_writer->get_index_file_total_size();
142
22
        return Status::OK();
143
22
    }
144
145
private:
146
    void _init_column_meta(ColumnMetaPB* meta, uint32_t column_id, const TabletColumn& column,
147
                           const ColumnWriterOptions& opts);
148
    Status _create_column_writer(uint32_t cid, const TabletColumn& column,
149
                                 const TabletSchemaSPtr& schema);
150
    uint64_t _estimated_remaining_size();
151
    Status _write_ordinal_index();
152
    Status _write_zone_map();
153
    Status _write_inverted_index();
154
    Status _write_ann_index();
155
    Status _write_bloom_filter_index();
156
    Status _write_short_key_index();
157
    Status _write_primary_key_index();
158
    Status _write_footer();
159
    Status _write_raw_data(const std::vector<Slice>& slices);
160
    void _set_min_max_key(const Slice& key);
161
    void _set_min_key(const Slice& key);
162
    void _set_max_key(const Slice& key);
163
    Status _append_row_store_column(const Block& block, size_t row_pos, size_t num_rows,
164
                                    uint32_t cid);
165
    Status _append_generated_column(const DerivedColumnGenerator& generator, const Block& block,
166
                                    size_t row_pos, size_t num_rows, uint32_t cid);
167
    // Thin wrapper over MowKeyProbe that translates a ProbeOutcome back into the out-parameters the
168
    // partial update fill loops use. `found_cb` receives the rowset that holds `loc`: the fixed
169
    // path pins it in its HistoricalRowFetcher, the flexible path in `_rsid_to_rowset`, which its
170
    // fill still reads from.
171
    Status _probe_key_for_mow(
172
            const MowKeyProbe& probe, std::string key, std::size_t segment_pos,
173
            bool have_input_seq_column, bool have_delete_sign,
174
            const std::vector<RowsetSharedPtr>& specified_rowsets,
175
            std::vector<std::unique_ptr<SegmentCacheHandle>>& segment_caches,
176
            bool& has_default_or_nullable, std::vector<bool>& use_default_or_null_flag,
177
            const std::function<void(const RowLocation& loc, const RowsetSharedPtr& rowset)>&
178
                    found_cb,
179
            const std::function<Status()>& not_found_cb, PartialUpdateStats& stats);
180
    Status _partial_update_preconditions_check(size_t row_pos, bool is_flexible_update);
181
    Status _append_block_with_partial_content(RowsInBlock& data, Block& full_block);
182
    Status _append_block_with_flexible_partial_content(RowsInBlock& data, Block& full_block);
183
    Status _generate_encoded_default_seq_value(const TabletSchema& tablet_schema,
184
                                               const PartialUpdateInfo& info,
185
                                               std::string* encoded_value);
186
    Status _generate_flexible_read_plan(
187
            FlexibleReadPlan& read_plan, RowsInBlock& data, size_t segment_start_pos,
188
            bool schema_has_sequence_col, int32_t seq_map_col_unique_id,
189
            std::vector<BitmapValue>* skip_bitmaps,
190
            const std::vector<IOlapColumnDataAccessor*>& key_columns,
191
            IOlapColumnDataAccessor* seq_column, const signed char* delete_signs,
192
            const std::vector<RowsetSharedPtr>& specified_rowsets,
193
            std::vector<std::unique_ptr<SegmentCacheHandle>>& segment_caches,
194
            bool& has_default_or_nullable, std::vector<bool>& use_default_or_null_flag,
195
            PartialUpdateStats& stats);
196
    Status _generate_key_index(RowsInBlock& data,
197
                               std::vector<IOlapColumnDataAccessor*>& key_columns,
198
                               IOlapColumnDataAccessor* seq_column,
199
                               std::map<uint32_t, IOlapColumnDataAccessor*>& cid_to_column);
200
    Status _generate_primary_key_index(
201
            const std::vector<IOlapColumnDataAccessor*>& primary_key_columns,
202
            IOlapColumnDataAccessor* seq_column, size_t num_rows, bool need_sort);
203
    Status _generate_short_key_index(std::vector<IOlapColumnDataAccessor*>& key_columns,
204
                                     size_t num_rows, const std::vector<size_t>& short_key_pos);
205
    Status _check_column_writer_disk_capacity(size_t cid);
206
    Status _finalize_column_writer_and_update_meta(size_t cid);
207
208
2.15k
    bool _is_mow() {
209
2.15k
        return _tablet_schema->keys_type() == UNIQUE_KEYS && _opts.enable_unique_key_merge_on_write;
210
2.15k
    }
211
1.58k
    bool _is_mow_with_cluster_key() {
212
1.58k
        return _is_mow() && !_tablet_schema->cluster_key_uids().empty();
213
1.58k
    }
214
215
private:
216
    friend class ::doris::BlockAggregator;
217
    uint32_t _segment_id;
218
    TabletSchemaSPtr _tablet_schema;
219
    BaseTabletSPtr _tablet;
220
    DataDir* _data_dir = nullptr;
221
    VerticalSegmentWriterOptions _opts;
222
223
    // Not owned. owned by RowsetWriter
224
    io::FileWriter* _file_writer = nullptr;
225
    // Not owned. owned by RowsetWriter or SegmentFlusher
226
    IndexFileWriter* _index_file_writer = nullptr;
227
228
    SegmentFooterPB _footer;
229
    SegmentIndexFileCacheInfo _index_file_cache_info;
230
    size_t _num_short_key_columns;
231
232
    std::unique_ptr<ShortKeyIndexBuilder> _short_key_index_builder;
233
    std::unique_ptr<PrimaryKeyIndexBuilder> _primary_key_index_builder;
234
    std::vector<std::unique_ptr<ColumnWriter>> _column_writers;
235
    std::unique_ptr<MemTracker> _mem_tracker;
236
237
    std::unique_ptr<OlapBlockDataConvertor> _olap_data_convertor;
238
    // used for building short key index or primary key index during vectorized write.
239
    // NOTE: must stay declared after _tablet_schema and _opts, the constructor
240
    // init list reads both through _is_mow().
241
    RowKeyEncoder _key_encoder;
242
    size_t _short_key_row_pos = 0;
243
244
    // _num_rows_written means row count already written in this current column group
245
    uint32_t _num_rows_written = 0;
246
247
    /** for partial update stats **/
248
    int64_t _num_rows_updated = 0;
249
    int64_t _num_rows_new_added = 0;
250
    int64_t _num_rows_deleted = 0;
251
    // number of rows filtered in strict mode partial update
252
    int64_t _num_rows_filtered = 0;
253
254
    // _row_count means total row count of this segment
255
    // In vertical compaction row count is recorded when key columns group finish
256
    //  and _num_rows_written will be updated in value column group
257
    uint32_t _row_count = 0;
258
259
    bool _is_first_row = true;
260
    faststring _min_key;
261
    faststring _max_key;
262
263
    std::shared_ptr<MowContext> _mow_context;
264
    // group every rowset-segment row id to speed up reader
265
    std::map<RowsetId, RowsetSharedPtr> _rsid_to_rowset;
266
267
    std::vector<RowsInBlock> _batched_blocks;
268
269
    // the derived column the transform chain hands off to this writer's bounded pump
270
    DerivedColumn _derived_column;
271
272
    BlockAggregator _block_aggregator;
273
};
274
275
} // namespace segment_v2
276
} // namespace doris