be/src/exec/scan/scanner.cpp
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 | | #include "exec/scan/scanner.h" |
19 | | |
20 | | #include <glog/logging.h> |
21 | | |
22 | | #include "common/config.h" |
23 | | #include "common/status.h" |
24 | | #include "core/block/column_with_type_and_name.h" |
25 | | #include "core/column/column_nothing.h" |
26 | | #include "exec/operator/scan_operator.h" |
27 | | #include "exec/scan/scan_node.h" |
28 | | #include "exprs/vexpr_context.h" |
29 | | #include "runtime/descriptors.h" |
30 | | #include "runtime/runtime_profile.h" |
31 | | #include "util/concurrency_stats.h" |
32 | | #include "util/defer_op.h" |
33 | | |
34 | | namespace doris { |
35 | | |
36 | | Scanner::Scanner(RuntimeState* state, ScanLocalStateBase* local_state, int64_t limit, |
37 | | RuntimeProfile* profile) |
38 | 1.01M | : _state(state), |
39 | 1.01M | _local_state(local_state), |
40 | 1.01M | _limit(limit), |
41 | 1.01M | _profile(profile), |
42 | 1.01M | _output_tuple_desc(_local_state->output_tuple_desc()), |
43 | 1.01M | _output_row_descriptor(_local_state->_parent->output_row_descriptor()), |
44 | 1.01M | _has_prepared(false), |
45 | 1.01M | _remote_slow_task_threshold(state->query_options().__isset.remote_slow_task_threshold |
46 | 1.01M | ? state->query_options().remote_slow_task_threshold |
47 | 1.01M | : INT64_MAX) { |
48 | 1.01M | _total_rf_num = cast_set<int>(_local_state->_helper.runtime_filter_nums()); |
49 | 1.01M | DorisMetrics::instance()->scanner_cnt->increment(1); |
50 | 1.01M | } |
51 | | |
52 | 0 | bool Scanner::is_slow_task() const { |
53 | 0 | return _file_cache_profile_reporter && |
54 | 0 | _file_cache_profile_reporter->num_remote_io_total->value() > _remote_slow_task_threshold; |
55 | 0 | } |
56 | | |
57 | 1.01M | Status Scanner::init(RuntimeState* state, const VExprContextSPtrs& conjuncts) { |
58 | 1.01M | if (!conjuncts.empty()) { |
59 | 14.3k | _conjuncts.resize(conjuncts.size()); |
60 | 32.1k | for (size_t i = 0; i != conjuncts.size(); ++i) { |
61 | 17.8k | RETURN_IF_ERROR(conjuncts[i]->clone(state, _conjuncts[i])); |
62 | 17.8k | } |
63 | 14.3k | } |
64 | | |
65 | 1.01M | const auto& projections = _local_state->_projections; |
66 | 1.01M | if (!projections.empty()) { |
67 | 946k | _projections.resize(projections.size()); |
68 | 8.72M | for (size_t i = 0; i != projections.size(); ++i) { |
69 | 7.78M | RETURN_IF_ERROR(projections[i]->clone(state, _projections[i])); |
70 | 7.78M | } |
71 | 946k | } |
72 | | |
73 | 1.01M | const auto& intermediate_projections = _local_state->_intermediate_projections; |
74 | 1.01M | if (!intermediate_projections.empty()) { |
75 | 10.7k | _intermediate_projections.resize(intermediate_projections.size()); |
76 | 24.2k | for (int i = 0; i < intermediate_projections.size(); i++) { |
77 | 13.5k | _intermediate_projections[i].resize(intermediate_projections[i].size()); |
78 | 81.0k | for (int j = 0; j < intermediate_projections[i].size(); j++) { |
79 | 67.5k | RETURN_IF_ERROR(intermediate_projections[i][j]->clone( |
80 | 67.5k | state, _intermediate_projections[i][j])); |
81 | 67.5k | } |
82 | 13.5k | } |
83 | 10.7k | } |
84 | | |
85 | 1.01M | return Status::OK(); |
86 | 1.01M | } |
87 | | |
88 | 1.05M | Status Scanner::get_block_after_projects(RuntimeState* state, Block* block, bool* eos) { |
89 | 1.05M | SCOPED_CONCURRENCY_COUNT(ConcurrencyStatsManager::instance().vscanner_get_block); |
90 | 1.05M | auto& row_descriptor = _local_state->_parent->row_descriptor(); |
91 | 1.05M | if (_output_row_descriptor) { |
92 | 954k | if (_alreay_eos) { |
93 | 0 | *eos = true; |
94 | 0 | _padding_block.swap(_origin_block); |
95 | 954k | } else { |
96 | 954k | _origin_block.clear_column_data(row_descriptor.num_materialized_slots()); |
97 | 954k | const auto min_batch_size = std::max(state->batch_size() / 2, 1); |
98 | 2.11M | while (_padding_block.rows() < min_batch_size && !*eos) { |
99 | 1.17M | RETURN_IF_ERROR(get_block(state, &_origin_block, eos)); |
100 | 1.17M | if (_origin_block.rows() >= min_batch_size) { |
101 | 15.6k | break; |
102 | 15.6k | } |
103 | | |
104 | 1.16M | if (_origin_block.rows() + _padding_block.rows() <= state->batch_size()) { |
105 | 1.16M | RETURN_IF_ERROR(_merge_padding_block()); |
106 | 1.16M | _origin_block.clear_column_data(row_descriptor.num_materialized_slots()); |
107 | 1.16M | } else { |
108 | 998 | if (_origin_block.rows() < _padding_block.rows()) { |
109 | 0 | _padding_block.swap(_origin_block); |
110 | 0 | } |
111 | 998 | break; |
112 | 998 | } |
113 | 1.16M | } |
114 | 954k | } |
115 | | |
116 | | // first output the origin block change eos = false, next time output padding block |
117 | | // set the eos to true |
118 | 954k | if (*eos && !_padding_block.empty() && !_origin_block.empty()) { |
119 | 0 | _alreay_eos = true; |
120 | 0 | *eos = false; |
121 | 0 | } |
122 | 954k | if (_origin_block.empty() && !_padding_block.empty()) { |
123 | 184k | _padding_block.swap(_origin_block); |
124 | 184k | } |
125 | 954k | return _do_projections(&_origin_block, block); |
126 | 954k | } else { |
127 | 99.7k | return get_block(state, block, eos); |
128 | 99.7k | } |
129 | 1.05M | } |
130 | | |
131 | 1.27M | Status Scanner::get_block(RuntimeState* state, Block* block, bool* eof) { |
132 | | // only empty block should be here |
133 | 1.27M | DCHECK(block->rows() == 0); |
134 | | // scanner running time |
135 | 1.27M | SCOPED_RAW_TIMER(&_per_scanner_timer); |
136 | 1.27M | int64_t rows_read_threshold = _num_rows_read + config::doris_scanner_row_num; |
137 | 1.27M | if (!block->mem_reuse()) { |
138 | 8.73M | for (auto* const slot_desc : _output_tuple_desc->slots()) { |
139 | 8.73M | block->insert(ColumnWithTypeAndName(slot_desc->get_empty_mutable_column(), |
140 | 8.73M | slot_desc->get_data_type_ptr(), |
141 | 8.73M | slot_desc->col_name())); |
142 | 8.73M | } |
143 | 1.12M | } |
144 | | |
145 | 1.27M | { |
146 | 1.28M | do { |
147 | | // 1. Get input block from scanner |
148 | 1.28M | { |
149 | | // get block time |
150 | 1.28M | SCOPED_TIMER(_local_state->_scan_timer); |
151 | 1.28M | RETURN_IF_ERROR(_get_block_impl(state, block, eof)); |
152 | 1.28M | if (*eof) { |
153 | 1.00M | DCHECK(block->rows() == 0); |
154 | 1.00M | break; |
155 | 1.00M | } |
156 | 279k | _num_rows_read += block->rows(); |
157 | 279k | _num_byte_read += block->allocated_bytes(); |
158 | 279k | } |
159 | | |
160 | | // 2. Filter the output block finally. |
161 | 0 | { |
162 | 279k | SCOPED_TIMER(_local_state->_filter_timer); |
163 | 279k | RETURN_IF_ERROR(_filter_output_block(block)); |
164 | 279k | } |
165 | | // record rows return (after filter) for _limit check |
166 | 279k | _num_rows_return += block->rows(); |
167 | 279k | } while (!_should_stop && !state->is_cancelled() && block->rows() == 0 && !(*eof) && |
168 | 279k | _num_rows_read < rows_read_threshold); |
169 | 1.27M | } |
170 | | |
171 | 1.27M | if (state->is_cancelled()) { |
172 | | // TODO: Should return the specific ErrorStatus instead of just Cancelled. |
173 | 10 | return Status::Cancelled("cancelled"); |
174 | 10 | } |
175 | 1.27M | *eof = *eof || _should_stop; |
176 | | // set eof to true if per scanner limit is reached |
177 | | // currently for query: ORDER BY key LIMIT n |
178 | 1.27M | *eof = *eof || (_limit > 0 && _num_rows_return >= _limit); |
179 | | |
180 | 1.27M | return Status::OK(); |
181 | 1.27M | } |
182 | | |
183 | 278k | Status Scanner::_filter_output_block(Block* block) { |
184 | 278k | auto old_rows = block->rows(); |
185 | 278k | Status st = VExprContext::filter_block(_conjuncts, block, block->columns()); |
186 | 278k | _counter.num_rows_unselected += old_rows - block->rows(); |
187 | 278k | return st; |
188 | 278k | } |
189 | | |
190 | 956k | Status Scanner::_do_projections(Block* origin_block, Block* output_block) { |
191 | 956k | SCOPED_RAW_TIMER(&_per_scanner_timer); |
192 | 956k | SCOPED_RAW_TIMER(&_projection_timer); |
193 | | |
194 | 956k | const size_t rows = origin_block->rows(); |
195 | 956k | if (rows == 0) { |
196 | 756k | return Status::OK(); |
197 | 756k | } |
198 | 199k | Block input_block = *origin_block; |
199 | | |
200 | 199k | std::vector<int> result_column_ids; |
201 | 199k | for (auto& projections : _intermediate_projections) { |
202 | 14.2k | result_column_ids.resize(projections.size()); |
203 | 90.2k | for (int i = 0; i < projections.size(); i++) { |
204 | 75.9k | RETURN_IF_ERROR(projections[i]->execute(&input_block, &result_column_ids[i])); |
205 | 75.9k | } |
206 | 14.2k | input_block.shuffle_columns(result_column_ids); |
207 | 14.2k | } |
208 | | |
209 | 199k | DCHECK_EQ(rows, input_block.rows()); |
210 | 199k | MutableBlock mutable_block = |
211 | 199k | VectorizedUtils::build_mutable_mem_reuse_block(output_block, *_output_row_descriptor); |
212 | | |
213 | 199k | auto& mutable_columns = mutable_block.mutable_columns(); |
214 | | |
215 | 199k | DCHECK_EQ(mutable_columns.size(), _projections.size()); |
216 | | |
217 | 738k | for (int i = 0; i < mutable_columns.size(); ++i) { |
218 | 540k | ColumnPtr column_ptr; |
219 | 540k | RETURN_IF_ERROR(_projections[i]->execute(&input_block, column_ptr)); |
220 | 539k | column_ptr = column_ptr->convert_to_full_column_if_const(); |
221 | 539k | if (mutable_columns[i]->is_nullable() != column_ptr->is_nullable()) { |
222 | 0 | throw Exception(ErrorCode::INTERNAL_ERROR, "Nullable mismatch"); |
223 | 0 | } |
224 | 539k | mutable_columns[i] = column_ptr->assume_mutable(); |
225 | 539k | } |
226 | | |
227 | 198k | output_block->set_columns(std::move(mutable_columns)); |
228 | | |
229 | | // origin columns was moved into output_block, so we need to set origin_block to empty columns |
230 | 198k | auto empty_columns = origin_block->clone_empty_columns(); |
231 | 198k | origin_block->set_columns(std::move(empty_columns)); |
232 | 198k | DCHECK_EQ(output_block->rows(), rows); |
233 | | |
234 | 198k | return Status::OK(); |
235 | 199k | } |
236 | | |
237 | 1.01M | Status Scanner::try_append_late_arrival_runtime_filter() { |
238 | 1.01M | if (_applied_rf_num == _total_rf_num) { |
239 | 961k | return Status::OK(); |
240 | 961k | } |
241 | 1.01M | DCHECK(_applied_rf_num < _total_rf_num); |
242 | 51.9k | int arrived_rf_num = 0; |
243 | 51.9k | RETURN_IF_ERROR(_local_state->update_late_arrival_runtime_filter(_state, arrived_rf_num)); |
244 | | |
245 | 51.9k | if (arrived_rf_num == _applied_rf_num) { |
246 | | // No newly arrived runtime filters, just return; |
247 | 1.08k | return Status::OK(); |
248 | 1.08k | } |
249 | | |
250 | | // avoid conjunct destroy in used by storage layer |
251 | 50.8k | _conjuncts.clear(); |
252 | 50.8k | RETURN_IF_ERROR(_local_state->clone_conjunct_ctxs(_conjuncts)); |
253 | 50.8k | return Status::OK(); |
254 | 50.8k | } |
255 | | |
256 | 1.01M | Status Scanner::close(RuntimeState* state) { |
257 | 1.01M | #ifndef BE_TEST |
258 | 1.01M | COUNTER_UPDATE(_local_state->_scanner_wait_worker_timer, _scanner_wait_worker_timer); |
259 | 1.01M | #endif |
260 | 1.01M | return Status::OK(); |
261 | 1.01M | } |
262 | | |
263 | 1.01M | bool Scanner::_try_close() { |
264 | 1.01M | bool expected = false; |
265 | 1.01M | return _is_closed.compare_exchange_strong(expected, true); |
266 | 1.01M | } |
267 | | |
268 | 1.00M | void Scanner::_collect_profile_before_close() { |
269 | 1.00M | COUNTER_UPDATE(_local_state->_scan_cpu_timer, _scan_cpu_timer); |
270 | 1.00M | COUNTER_UPDATE(_local_state->_rows_read_counter, _num_rows_read); |
271 | | |
272 | | // Update stats for load |
273 | 1.00M | _state->update_num_rows_load_filtered(_counter.num_rows_filtered); |
274 | 1.00M | _state->update_num_rows_load_unselected(_counter.num_rows_unselected); |
275 | 1.00M | } |
276 | | |
277 | 1.00M | void Scanner::update_scan_cpu_timer() { |
278 | 1.00M | int64_t cpu_time = _cpu_watch.elapsed_time(); |
279 | 1.00M | _scan_cpu_timer += cpu_time; |
280 | 1.00M | if (_state && _state->get_query_ctx()) { |
281 | 1.00M | _state->get_query_ctx()->resource_ctx()->cpu_context()->update_cpu_cost_ms(cpu_time); |
282 | 1.00M | } |
283 | 1.00M | } |
284 | | |
285 | | } // namespace doris |