be/src/storage/segment/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 <butil/macros.h> |
21 | | #include <gen_cpp/olap_file.pb.h> |
22 | | #include <gen_cpp/segment_v2.pb.h> |
23 | | #include <stddef.h> |
24 | | |
25 | | #include <cstdint> |
26 | | #include <functional> |
27 | | #include <map> |
28 | | #include <memory> // unique_ptr |
29 | | #include <string> |
30 | | #include <vector> |
31 | | |
32 | | #include "common/status.h" // Status |
33 | | #include "storage/index/index_file_writer.h" |
34 | | #include "storage/olap_define.h" |
35 | | #include "storage/segment/column_writer.h" |
36 | | #include "storage/tablet/tablet.h" |
37 | | #include "storage/tablet/tablet_schema.h" |
38 | | #include "util/faststring.h" |
39 | | #include "util/slice.h" |
40 | | |
41 | | namespace doris { |
42 | | class Block; |
43 | | class IOlapColumnDataAccessor; |
44 | | class OlapBlockDataConvertor; |
45 | | |
46 | | // TODO(lingbin): Should be a conf that can be dynamically adjusted, or a member in the context |
47 | | const uint32_t MAX_SEGMENT_SIZE = static_cast<uint32_t>(OLAP_MAX_COLUMN_SEGMENT_FILE_SIZE * |
48 | | OLAP_COLUMN_FILE_SEGMENT_SIZE_SCALE); |
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 | | } // namespace io |
59 | | |
60 | | namespace segment_v2 { |
61 | | |
62 | | extern const char* k_segment_magic; |
63 | | extern const uint32_t k_segment_magic_length; |
64 | | |
65 | | class VariantStatsCaculator; |
66 | | |
67 | | struct SegmentWriterOptions { |
68 | | uint32_t num_rows_per_block = 1024; |
69 | | uint32_t max_rows_per_segment = UINT32_MAX; |
70 | | bool enable_unique_key_merge_on_write = false; |
71 | | CompressionTypePB compression_type = UNKNOWN_COMPRESSION; |
72 | | |
73 | | RowsetWriterContext* rowset_ctx = nullptr; |
74 | | DataWriteType write_type = DataWriteType::TYPE_DEFAULT; |
75 | | std::shared_ptr<MowContext> mow_ctx; |
76 | | }; |
77 | | |
78 | | using TabletSharedPtr = std::shared_ptr<Tablet>; |
79 | | |
80 | | class SegmentWriter { |
81 | | public: |
82 | | explicit SegmentWriter(io::FileWriter* file_writer, uint32_t segment_id, |
83 | | TabletSchemaSPtr tablet_schema, BaseTabletSPtr tablet, DataDir* data_dir, |
84 | | const SegmentWriterOptions& opts, IndexFileWriter* inverted_file_writer); |
85 | | ~SegmentWriter(); |
86 | | |
87 | | Status init(); |
88 | | |
89 | | // for vertical compaction |
90 | | Status init(const std::vector<uint32_t>& col_ids, bool has_key); |
91 | | |
92 | | Status append_row(const RowCursor& row); |
93 | | |
94 | | Status append_block(const Block* block, size_t row_pos, size_t num_rows); |
95 | | Status probe_key_for_mow(std::string key, std::size_t segment_pos, bool have_input_seq_column, |
96 | | bool have_delete_sign, |
97 | | const std::vector<RowsetSharedPtr>& specified_rowsets, |
98 | | std::vector<std::unique_ptr<SegmentCacheHandle>>& segment_caches, |
99 | | bool& has_default_or_nullable, |
100 | | std::vector<bool>& use_default_or_null_flag, |
101 | | const std::function<void(const RowLocation& loc)>& found_cb, |
102 | | const std::function<Status()>& not_found_cb, |
103 | | PartialUpdateStats& stats); |
104 | | Status partial_update_preconditions_check(size_t row_pos); |
105 | | Status append_block_with_partial_content(const Block* block, size_t row_pos, size_t num_rows); |
106 | | |
107 | | int64_t max_row_to_add(size_t row_avg_size_in_bytes); |
108 | | |
109 | | uint64_t estimate_segment_size(); |
110 | | |
111 | 45.7k | uint32_t num_rows_written() const { return _num_rows_written; } |
112 | | |
113 | | // for partial update |
114 | 5.35k | int64_t num_rows_updated() const { return _num_rows_updated; } |
115 | 5.35k | int64_t num_rows_deleted() const { return _num_rows_deleted; } |
116 | 5.35k | int64_t num_rows_new_added() const { return _num_rows_new_added; } |
117 | 5.35k | int64_t num_rows_filtered() const { return _num_rows_filtered; } |
118 | | |
119 | 50.5k | uint32_t row_count() const { return _row_count; } |
120 | | |
121 | | Status finalize(uint64_t* segment_file_size, uint64_t* index_size); |
122 | | |
123 | 5.36k | uint32_t get_segment_id() const { return _segment_id; } |
124 | | |
125 | | Status finalize_columns_data(); |
126 | | Status finalize_columns_index(uint64_t* index_size); |
127 | | Status finalize_footer(uint64_t* segment_file_size); |
128 | | |
129 | | void init_column_meta(ColumnMetaPB* meta, uint32_t column_id, const TabletColumn& column, |
130 | | TabletSchemaSPtr tablet_schema); |
131 | | Slice min_encoded_key(); |
132 | | Slice max_encoded_key(); |
133 | | |
134 | 0 | bool is_unique_key() { return _tablet_schema->keys_type() == UNIQUE_KEYS; } |
135 | | |
136 | | void clear(); |
137 | | |
138 | | void set_mow_context(std::shared_ptr<MowContext> mow_context); |
139 | | |
140 | 5.36k | Status close_inverted_index(int64_t* inverted_index_file_size) { |
141 | | // no inverted index |
142 | 5.36k | if (_index_file_writer == nullptr) { |
143 | 4.85k | *inverted_index_file_size = 0; |
144 | 4.85k | return Status::OK(); |
145 | 4.85k | } |
146 | 505 | RETURN_IF_ERROR(_index_file_writer->begin_close()); |
147 | 505 | *inverted_index_file_size = _index_file_writer->get_index_file_total_size(); |
148 | 505 | return Status::OK(); |
149 | 505 | } |
150 | | |
151 | 6 | uint64_t primary_keys_size() const { return _primary_keys_size; } |
152 | | |
153 | | private: |
154 | | DISALLOW_COPY_AND_ASSIGN(SegmentWriter); |
155 | | Status _create_column_writer(uint32_t cid, const TabletColumn& column, |
156 | | const TabletSchemaSPtr& schema); |
157 | | Status _create_writers(const TabletSchemaSPtr& tablet_schema, |
158 | | const std::vector<uint32_t>& col_ids); |
159 | | Status _write_data(); |
160 | | Status _write_ordinal_index(); |
161 | | Status _write_zone_map(); |
162 | | Status _write_inverted_index(); |
163 | | Status _write_ann_index(); |
164 | | Status _write_bloom_filter_index(); |
165 | | Status _write_short_key_index(); |
166 | | Status _write_primary_key_index(); |
167 | | Status _write_footer(); |
168 | | Status _write_raw_data(const std::vector<Slice>& slices); |
169 | | void _maybe_invalid_row_cache(const std::string& key); |
170 | | std::string _encode_keys(const std::vector<IOlapColumnDataAccessor*>& key_columns, size_t pos); |
171 | | // used for unique-key with merge on write and segment min_max key |
172 | | std::string _full_encode_keys(const std::vector<IOlapColumnDataAccessor*>& key_columns, |
173 | | size_t pos, bool null_first = true); |
174 | | |
175 | | static std::string _full_encode_keys(const std::vector<const KeyCoder*>& key_coders, |
176 | | const std::vector<IOlapColumnDataAccessor*>& key_columns, |
177 | | size_t pos, bool null_first = true); |
178 | | |
179 | | // used for unique-key with merge on write |
180 | | void _encode_seq_column(const IOlapColumnDataAccessor* seq_column, size_t pos, |
181 | | std::string* encoded_keys); |
182 | | void _encode_rowid(const uint32_t rowid, std::string* encoded_keys); |
183 | | void set_min_max_key(const Slice& key); |
184 | | void set_min_key(const Slice& key); |
185 | | void set_max_key(const Slice& key); |
186 | | void _serialize_block_to_row_column(const Block& block); |
187 | | Status _generate_primary_key_index( |
188 | | const std::vector<const KeyCoder*>& primary_key_coders, |
189 | | const std::vector<IOlapColumnDataAccessor*>& primary_key_columns, |
190 | | IOlapColumnDataAccessor* seq_column, size_t num_rows, bool need_sort); |
191 | | Status _generate_short_key_index(std::vector<IOlapColumnDataAccessor*>& key_columns, |
192 | | size_t num_rows, const std::vector<size_t>& short_key_pos); |
193 | | bool _is_mow(); |
194 | | bool _is_mow_with_cluster_key(); |
195 | | |
196 | | private: |
197 | | uint32_t _segment_id; |
198 | | TabletSchemaSPtr _tablet_schema; |
199 | | BaseTabletSPtr _tablet; |
200 | | DataDir* _data_dir = nullptr; |
201 | | SegmentWriterOptions _opts; |
202 | | |
203 | | // Not owned. owned by RowsetWriter or SegmentFlusher |
204 | | io::FileWriter* _file_writer = nullptr; |
205 | | // Not owned. owned by RowsetWriter or SegmentFlusher |
206 | | IndexFileWriter* _index_file_writer = nullptr; |
207 | | |
208 | | SegmentFooterPB _footer; |
209 | | // for mow tables with cluster key, the sort key is the cluster keys not unique keys |
210 | | // for other tables, the sort key is the keys |
211 | | size_t _num_sort_key_columns; |
212 | | size_t _num_short_key_columns; |
213 | | |
214 | | std::unique_ptr<ShortKeyIndexBuilder> _short_key_index_builder; |
215 | | std::unique_ptr<PrimaryKeyIndexBuilder> _primary_key_index_builder; |
216 | | std::vector<std::unique_ptr<ColumnWriter>> _column_writers; |
217 | | std::unique_ptr<MemTracker> _mem_tracker; |
218 | | |
219 | | std::unique_ptr<OlapBlockDataConvertor> _olap_data_convertor; |
220 | | // used for building short key index or primary key index during vectorized write. |
221 | | // for mow table with cluster keys, this is cluster keys |
222 | | std::vector<const KeyCoder*> _key_coders; |
223 | | // for mow table with cluster keys, this is primary keys |
224 | | std::vector<const KeyCoder*> _primary_key_coders; |
225 | | const KeyCoder* _seq_coder = nullptr; |
226 | | const KeyCoder* _rowid_coder = nullptr; |
227 | | std::vector<uint16_t> _key_index_size; |
228 | | size_t _short_key_row_pos = 0; |
229 | | |
230 | | std::vector<uint32_t> _column_ids; |
231 | | bool _has_key = true; |
232 | | // _num_rows_written means row count already written in this current column group |
233 | | uint32_t _num_rows_written = 0; |
234 | | |
235 | | /** for partial update stats **/ |
236 | | int64_t _num_rows_updated = 0; |
237 | | int64_t _num_rows_new_added = 0; |
238 | | int64_t _num_rows_deleted = 0; |
239 | | // number of rows filtered in strict mode partial update |
240 | | int64_t _num_rows_filtered = 0; |
241 | | // _row_count means total row count of this segment |
242 | | // In vertical compaction row count is recorded when key columns group finish |
243 | | // and _num_rows_written will be updated in value column group |
244 | | uint32_t _row_count = 0; |
245 | | |
246 | | bool _is_first_row = true; |
247 | | faststring _min_key; |
248 | | faststring _max_key; |
249 | | |
250 | | std::shared_ptr<MowContext> _mow_context; |
251 | | // group every rowset-segment row id to speed up reader |
252 | | std::map<RowsetId, RowsetSharedPtr> _rsid_to_rowset; |
253 | | std::vector<std::string> _primary_keys; |
254 | | uint64_t _primary_keys_size = 0; |
255 | | // variant statistics calculator for efficient stats collection |
256 | | std::unique_ptr<VariantStatsCaculator> _variant_stats_calculator; |
257 | | }; |
258 | | |
259 | | } // namespace segment_v2 |
260 | | } // namespace doris |