be/src/storage/rowid_conversion.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 <algorithm> |
21 | | #include <map> |
22 | | #include <memory> |
23 | | #include <utility> |
24 | | #include <vector> |
25 | | |
26 | | #include "common/cast_set.h" |
27 | | #include "common/check.h" |
28 | | #include "runtime/thread_context.h" |
29 | | #include "storage/olap_common.h" |
30 | | #include "storage/utils.h" |
31 | | |
32 | | namespace doris { |
33 | | |
34 | | // For unique key merge on write table, we should update delete bitmap |
35 | | // of destination rowset when compaction finished. |
36 | | // Through the row id correspondence between the source rowset and the |
37 | | // destination rowset, we can quickly update the delete bitmap of the |
38 | | // destination rowset. |
39 | | class RowIdConversion { |
40 | | public: |
41 | | enum class Mode { DENSE, LAZY_CHUNKED }; |
42 | | |
43 | | struct DestinationRowId { |
44 | | uint32_t segment_pos; |
45 | | uint32_t row_id; |
46 | | }; |
47 | | |
48 | 273 | explicit RowIdConversion(Mode mode = Mode::DENSE) : _mode(mode) {} |
49 | 273 | ~RowIdConversion() { RELEASE_THREAD_MEM_TRACKER(_seg_rowid_map_mem_used); } |
50 | | |
51 | | Status init_segment_map(const RowsetId& src_rowset_id, const std::vector<uint32_t>& segment_ids, |
52 | 424 | const std::vector<uint32_t>& num_rows) { |
53 | 424 | DCHECK_EQ(segment_ids.size(), num_rows.size()); |
54 | 1.16k | for (size_t i = 0; i < num_rows.size(); i++) { |
55 | 739 | auto src_segment = std::pair<RowsetId, uint32_t> {src_rowset_id, segment_ids[i]}; |
56 | 739 | auto iter = _segment_to_id_map.find(src_segment); |
57 | | // A segment-group reader can be reopened, so reuse existing source-segment maps. |
58 | 739 | if (iter != _segment_to_id_map.end()) { |
59 | 22 | DORIS_CHECK_LT(iter->second, _segment_num_rows.size()); |
60 | 22 | DORIS_CHECK_EQ(_segment_num_rows[iter->second], num_rows[i]); |
61 | 22 | continue; |
62 | 22 | } |
63 | | |
64 | 717 | constexpr size_t RESERVED_MEMORY = 10 * 1024 * 1024; // 10M |
65 | 717 | RETURN_IF_ERROR(check_memory_limit(RESERVED_MEMORY)); |
66 | | |
67 | 717 | uint32_t id = cast_set<uint32_t>(_segment_num_rows.size()); |
68 | 717 | auto insert_result = _segment_to_id_map.emplace(src_segment, id); |
69 | 717 | DORIS_CHECK(insert_result.second); |
70 | 717 | _id_to_segment_map.push_back(src_segment); |
71 | 717 | _segment_num_rows.push_back(num_rows[i]); |
72 | 717 | if (_mode == Mode::LAZY_CHUNKED) { |
73 | 4 | _lazy_segments_rowid_map.emplace_back(); |
74 | 4 | auto& chunks = _lazy_segments_rowid_map.back(); |
75 | 4 | chunks.resize((cast_set<size_t>(num_rows[i]) + ROWS_PER_CHUNK - 1) / |
76 | 4 | ROWS_PER_CHUNK); |
77 | 4 | track_lazy_mem_usage(chunks.capacity() * sizeof(std::unique_ptr<RowIdPair[]>)); |
78 | 4 | continue; |
79 | 4 | } |
80 | | |
81 | 713 | std::vector<std::pair<uint32_t, uint32_t>> vec( |
82 | 713 | num_rows[i], std::pair<uint32_t, uint32_t>(UINT32_MAX, UINT32_MAX)); |
83 | | |
84 | | //NOTE: manually count _segments_rowid_map's memory here, because _segments_rowid_map could be used by indexCompaction. |
85 | | // indexCompaction is a thridparty code, it's too complex to modify it. |
86 | | // refer compact_column. |
87 | 713 | track_mem_usage(vec.capacity()); |
88 | 713 | _segments_rowid_map.emplace_back(std::move(vec)); |
89 | 713 | } |
90 | 424 | return Status::OK(); |
91 | 424 | } |
92 | | |
93 | | // set dst rowset id |
94 | 150 | void set_dst_rowset_id(const RowsetId& dst_rowset_id) { _dst_rowst_id = dst_rowset_id; } |
95 | 27 | const RowsetId& get_dst_rowset_id() const { return _dst_rowst_id; } |
96 | | |
97 | | // add row id to the map |
98 | | Status add(const std::vector<RowLocation>& rss_row_ids, |
99 | 1.78k | const std::vector<uint32_t>& dst_segments_num_row) { |
100 | 4.75M | for (auto& item : rss_row_ids) { |
101 | 4.75M | if (item.row_id == -1) { |
102 | 0 | continue; |
103 | 0 | } |
104 | 4.75M | uint32_t id = _segment_to_id_map.at( |
105 | 4.75M | std::pair<RowsetId, uint32_t> {item.rowset_id, item.segment_id}); |
106 | 4.75M | if (_cur_dst_segment_pos < dst_segments_num_row.size() && |
107 | 4.75M | _cur_dst_segment_rowid >= dst_segments_num_row[_cur_dst_segment_pos]) { |
108 | 1.15k | _cur_dst_segment_pos++; |
109 | 1.15k | _cur_dst_segment_rowid = 0; |
110 | 1.15k | } |
111 | 4.75M | if (_mode == Mode::DENSE) { |
112 | 4.75M | _segments_rowid_map[id][item.row_id] = std::pair<uint32_t, uint32_t> { |
113 | 4.75M | _cur_dst_segment_pos, _cur_dst_segment_rowid++}; |
114 | 4.75M | continue; |
115 | 4.75M | } |
116 | 6 | RowIdPair* destination = nullptr; |
117 | 6 | RETURN_IF_ERROR(get_or_create_lazy_destination(id, item.row_id, &destination)); |
118 | 6 | *destination = {_cur_dst_segment_pos, _cur_dst_segment_rowid++}; |
119 | 6 | } |
120 | 1.78k | return Status::OK(); |
121 | 1.78k | } |
122 | | |
123 | | // Get the destination segment position and row id. The physical destination segment id is |
124 | | // resolved only after the output rowset is built. |
125 | | // return non-zero if the src RowLocation does not exist |
126 | 1.77M | int get(const RowLocation& src, DestinationRowId* dst) const { |
127 | 1.77M | auto iter = _segment_to_id_map.find({src.rowset_id, src.segment_id}); |
128 | 1.77M | if (iter == _segment_to_id_map.end()) { |
129 | 2 | return -1; |
130 | 2 | } |
131 | 1.77M | const RowIdPair* destination = nullptr; |
132 | 1.77M | if (_mode == Mode::DENSE) { |
133 | 1.77M | const auto& rowid_map = _segments_rowid_map[iter->second]; |
134 | 1.77M | if (src.row_id >= rowid_map.size()) { |
135 | 1 | return -1; |
136 | 1 | } |
137 | 1.77M | destination = &rowid_map[src.row_id]; |
138 | 1.77M | } else { |
139 | 10 | const auto id = iter->second; |
140 | 10 | if (src.row_id >= _segment_num_rows[id]) { |
141 | 1 | return -1; |
142 | 1 | } |
143 | 9 | destination = get_lazy_destination(id, src.row_id); |
144 | 9 | } |
145 | 1.77M | if (destination == nullptr) { |
146 | 2 | return -1; |
147 | 2 | } |
148 | 1.77M | const auto& [dst_segment_pos, dst_rowid] = *destination; |
149 | 1.77M | if (dst_segment_pos == UINT32_MAX && dst_rowid == UINT32_MAX) { |
150 | 764k | return -1; |
151 | 764k | } |
152 | | |
153 | 1.00M | dst->segment_pos = dst_segment_pos; |
154 | 1.00M | dst->row_id = dst_rowid; |
155 | 1.00M | return 0; |
156 | 1.77M | } |
157 | | |
158 | | const std::vector<std::vector<std::pair<uint32_t, uint32_t>>>& get_rowid_conversion_map() |
159 | 29 | const { |
160 | 29 | DORIS_CHECK(_mode == Mode::DENSE); |
161 | 29 | return _segments_rowid_map; |
162 | 29 | } |
163 | | |
164 | 3 | size_t memory_usage() const { return _seg_rowid_map_mem_used; } |
165 | | |
166 | 30 | const std::map<std::pair<RowsetId, uint32_t>, uint32_t>& get_src_segment_to_id_map() const { |
167 | 30 | return _segment_to_id_map; |
168 | 30 | } |
169 | | |
170 | 0 | std::pair<RowsetId, uint32_t> get_segment_by_id(uint32_t id) const { |
171 | 0 | DCHECK_GT(_id_to_segment_map.size(), id); |
172 | 0 | return _id_to_segment_map.at(id); |
173 | 0 | } |
174 | | |
175 | 0 | uint32_t get_id_by_segment(const std::pair<RowsetId, uint32_t>& segment) const { |
176 | 0 | return _segment_to_id_map.at(segment); |
177 | 0 | } |
178 | | |
179 | | private: |
180 | | using RowIdPair = std::pair<uint32_t, uint32_t>; |
181 | | using LazySegmentRowIdMap = std::vector<std::unique_ptr<RowIdPair[]>>; |
182 | | // A 4096-row chunk uses 32 KiB, balancing sparse-range waste and allocation overhead. |
183 | | static constexpr uint32_t ROWS_PER_CHUNK = 4096; |
184 | | |
185 | 723 | Status check_memory_limit(size_t reserved_memory) const { |
186 | 723 | if (!doris::GlobalMemoryArbitrator::is_exceed_hard_mem_limit(reserved_memory)) { |
187 | 723 | return Status::OK(); |
188 | 723 | } |
189 | 0 | return Status::MemoryLimitExceeded(fmt::format( |
190 | 0 | "RowIdConversion allocation failed, process memory exceed limit or sys available " |
191 | 0 | "memory less than low water mark, {}, consuming tracker:<{}>, peak used {}, " |
192 | 0 | "current used {}.", |
193 | 0 | doris::GlobalMemoryArbitrator::process_mem_log_str(), |
194 | 0 | doris::thread_context()->thread_mem_tracker_mgr->limiter_mem_tracker()->label(), |
195 | 0 | doris::thread_context() |
196 | 0 | ->thread_mem_tracker_mgr->limiter_mem_tracker() |
197 | 0 | ->peak_consumption(), |
198 | 0 | doris::thread_context() |
199 | 0 | ->thread_mem_tracker_mgr->limiter_mem_tracker() |
200 | 0 | ->consumption())); |
201 | 723 | } |
202 | | |
203 | | Status get_or_create_lazy_destination(uint32_t segment_id, uint32_t row_id, |
204 | 6 | RowIdPair** destination) { |
205 | 6 | DORIS_CHECK_LT(segment_id, _segment_num_rows.size()); |
206 | 6 | DORIS_CHECK_LT(row_id, _segment_num_rows[segment_id]); |
207 | 6 | auto& chunks = _lazy_segments_rowid_map[segment_id]; |
208 | 6 | auto& chunk = chunks[row_id / ROWS_PER_CHUNK]; |
209 | 6 | if (chunk == nullptr) { |
210 | 6 | constexpr size_t CHUNK_BYTES = ROWS_PER_CHUNK * sizeof(RowIdPair); |
211 | 6 | RETURN_IF_ERROR(check_memory_limit(CHUNK_BYTES)); |
212 | 6 | chunk = std::make_unique<RowIdPair[]>(ROWS_PER_CHUNK); |
213 | 6 | std::fill_n(chunk.get(), ROWS_PER_CHUNK, RowIdPair {UINT32_MAX, UINT32_MAX}); |
214 | 6 | track_lazy_mem_usage(CHUNK_BYTES); |
215 | 6 | } |
216 | 6 | *destination = &chunk[row_id % ROWS_PER_CHUNK]; |
217 | 6 | return Status::OK(); |
218 | 6 | } |
219 | | |
220 | 9 | const RowIdPair* get_lazy_destination(uint32_t segment_id, uint32_t row_id) const { |
221 | 9 | const auto& chunks = _lazy_segments_rowid_map[segment_id]; |
222 | 9 | const auto& chunk = chunks[row_id / ROWS_PER_CHUNK]; |
223 | 9 | return chunk == nullptr ? nullptr : &chunk[row_id % ROWS_PER_CHUNK]; |
224 | 9 | } |
225 | | |
226 | 713 | void track_mem_usage(size_t delta_std_pair_cap) { |
227 | 713 | _std_pair_cap += delta_std_pair_cap; |
228 | | |
229 | 713 | size_t new_size = |
230 | 713 | _std_pair_cap * sizeof(std::pair<uint32_t, uint32_t>) + |
231 | 713 | _segments_rowid_map.capacity() * sizeof(std::vector<std::pair<uint32_t, uint32_t>>); |
232 | 713 | CONSUME_THREAD_MEM_TRACKER(new_size - _seg_rowid_map_mem_used); |
233 | 0 | _seg_rowid_map_mem_used = new_size; |
234 | 713 | } |
235 | | |
236 | 10 | void track_lazy_mem_usage(size_t bytes) { |
237 | 10 | CONSUME_THREAD_MEM_TRACKER(bytes); |
238 | 0 | _seg_rowid_map_mem_used += bytes; |
239 | 10 | } |
240 | | |
241 | | private: |
242 | | // the first level vector: index indicates src segment. |
243 | | // the second level vector: index indicates row id of source segment, |
244 | | // value indicates destination segment position and row id. |
245 | | // <UINT32_MAX, UINT32_MAX> indicates current row not exist. |
246 | | std::vector<std::vector<std::pair<uint32_t, uint32_t>>> _segments_rowid_map; |
247 | | // The first-level index indicates the internal source segment id. |
248 | | // The second-level index is source row_id / ROWS_PER_CHUNK and selects a lazy chunk. |
249 | | // The chunk offset is source row_id % ROWS_PER_CHUNK. |
250 | | // The value indicates destination segment position and row id. |
251 | | std::vector<LazySegmentRowIdMap> _lazy_segments_rowid_map; |
252 | | std::vector<uint32_t> _segment_num_rows; |
253 | | size_t _seg_rowid_map_mem_used {0}; |
254 | | size_t _std_pair_cap {0}; |
255 | | Mode _mode; |
256 | | |
257 | | // Map source segment to 0 to n |
258 | | std::map<std::pair<RowsetId, uint32_t>, uint32_t> _segment_to_id_map; |
259 | | |
260 | | // Map 0 to n to source segment |
261 | | std::vector<std::pair<RowsetId, uint32_t>> _id_to_segment_map; |
262 | | |
263 | | // dst rowset id |
264 | | RowsetId _dst_rowst_id; |
265 | | |
266 | | // current dst segment position |
267 | | std::uint32_t _cur_dst_segment_pos = 0; |
268 | | |
269 | | // current rowid of dst segment |
270 | | std::uint32_t _cur_dst_segment_rowid = 0; |
271 | | }; |
272 | | |
273 | | } // namespace doris |