be/src/exec/scan/scanner.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 <stdint.h> |
21 | | |
22 | | #include <algorithm> |
23 | | #include <atomic> |
24 | | #include <functional> |
25 | | #include <optional> |
26 | | #include <vector> |
27 | | |
28 | | #include "common/status.h" |
29 | | #include "core/block/block.h" |
30 | | #include "runtime/exec_env.h" |
31 | | #include "runtime/runtime_state.h" |
32 | | #include "storage/tablet/tablet.h" |
33 | | #include "util/stopwatch.hpp" |
34 | | |
35 | | namespace doris { |
36 | | class RuntimeProfile; |
37 | | class TupleDescriptor; |
38 | | |
39 | | class VExprContext; |
40 | | |
41 | | class ScanLocalStateBase; |
42 | | } // namespace doris |
43 | | |
44 | | namespace doris { |
45 | | |
46 | | // Counter for load |
47 | | struct ScannerCounter { |
48 | 21.1k | ScannerCounter() : num_rows_filtered(0), num_rows_unselected(0) {} |
49 | | |
50 | | int64_t num_rows_filtered; // unqualified rows (unmatched the dest schema, or no partition) |
51 | | int64_t num_rows_unselected; // rows filtered by predicates |
52 | | }; |
53 | | |
54 | | class Scanner { |
55 | | public: |
56 | | Scanner(RuntimeState* state, ScanLocalStateBase* local_state, int64_t limit, |
57 | | RuntimeProfile* profile); |
58 | | |
59 | | //only used for FileScanner read one line. |
60 | | Scanner(RuntimeState* state, RuntimeProfile* profile) |
61 | 19 | : _state(state), _limit(1), _profile(profile), _total_rf_num(0), _has_prepared(false) { |
62 | 19 | DorisMetrics::instance()->scanner_cnt->increment(1); |
63 | 19 | }; |
64 | | |
65 | 21.1k | virtual ~Scanner() { |
66 | 21.1k | SCOPED_SWITCH_THREAD_MEM_TRACKER_LIMITER(_state->query_mem_tracker()); |
67 | 21.1k | _input_block.clear(); |
68 | 21.1k | _conjuncts.clear(); |
69 | 21.1k | _projections.clear(); |
70 | 21.1k | _origin_block.clear(); |
71 | 21.1k | _common_expr_ctxs_push_down.clear(); |
72 | 21.1k | DorisMetrics::instance()->scanner_cnt->increment(-1); |
73 | 21.1k | } |
74 | | |
75 | | virtual Status init(RuntimeState* state, const VExprContextSPtrs& conjuncts); |
76 | 21.1k | Status prepare() { |
77 | 21.1k | SCOPED_RAW_TIMER(&_per_scanner_timer); |
78 | 21.1k | SCOPED_RAW_TIMER(&_per_scanner_prepare_timer); |
79 | 21.1k | return _prepare_impl(); |
80 | 21.1k | } |
81 | | |
82 | 21.1k | Status open(RuntimeState* state) { |
83 | 21.1k | SCOPED_RAW_TIMER(&_per_scanner_timer); |
84 | 21.1k | SCOPED_RAW_TIMER(&_per_scanner_open_timer); |
85 | 21.1k | return _open_impl(state); |
86 | 21.1k | } |
87 | | |
88 | | Status get_block(RuntimeState* state, Block* block, bool* eos); |
89 | | Status get_block_after_projects(RuntimeState* state, Block* block, bool* eos); |
90 | | |
91 | | virtual Status close(RuntimeState* state); |
92 | | |
93 | | // Try to stop scanner, and all running readers. |
94 | 21.1k | virtual void try_stop() { _should_stop = true; }; |
95 | | |
96 | 0 | virtual std::string get_name() { return ""; } |
97 | | |
98 | | // return the readable name of current scan range. |
99 | | // eg, for file scanner, return the current file path. |
100 | 0 | virtual std::string get_current_scan_range_name() { return "not implemented"; } |
101 | | |
102 | | #ifdef BE_TEST |
103 | | static uint64_t TEST_build_condition_cache_digest(uint64_t seed, |
104 | | const VExprContextSPtrs& conjuncts); |
105 | | #endif |
106 | | |
107 | | protected: |
108 | | // Rebuild the condition-cache digest from the scanner's current conjunct snapshot. The local |
109 | | // state's digest is used only as a safety gate: zero means condition cache was disabled during |
110 | | // scan-node open (for example by TopN or an expression without a reliable digest). |
111 | | uint64_t _current_condition_cache_digest() const; |
112 | | static uint64_t _build_condition_cache_digest(uint64_t seed, |
113 | | const VExprContextSPtrs& conjuncts); |
114 | | |
115 | 6 | virtual Status _prepare_impl() { |
116 | 6 | _has_prepared = true; |
117 | 6 | return Status::OK(); |
118 | 6 | } |
119 | | |
120 | 21.1k | virtual Status _open_impl(RuntimeState* state) { |
121 | 21.1k | _block_avg_bytes = state->batch_size() * 8; |
122 | 21.1k | return Status::OK(); |
123 | 21.1k | } |
124 | | |
125 | | // Subclass should implement this to return data. |
126 | | virtual Status _get_block_impl(RuntimeState* state, Block* block, bool* eof) = 0; |
127 | | |
128 | 20.9k | virtual bool _can_merge_padding_blocks(const Block& /*left*/, const Block& /*right*/) const { |
129 | 20.9k | return true; |
130 | 20.9k | } |
131 | | |
132 | 26.0k | Status _merge_padding_block() { |
133 | 26.0k | if (_padding_block.empty()) { |
134 | 20.9k | _padding_block.swap(_origin_block); |
135 | 20.9k | } else if (_origin_block.rows()) { |
136 | 273 | ScopedMutableBlock scoped_mutable_block(&_padding_block); |
137 | 273 | auto& mutable_block = scoped_mutable_block.mutable_block(); |
138 | 273 | RETURN_IF_ERROR(mutable_block.merge(_origin_block)); |
139 | 273 | } |
140 | 26.0k | return Status::OK(); |
141 | 26.0k | } |
142 | | |
143 | | // Update the counters before closing this scanner |
144 | | virtual void _collect_profile_before_close(); |
145 | | |
146 | | // Whether rows filtered/unselected by this scanner should be reported to the load |
147 | | // counters in RuntimeState. Only the scanner reading the load source data should |
148 | | // report, otherwise rows filtered by query predicates (e.g. in INSERT INTO ... SELECT |
149 | | // or DELETE FROM ... WHERE) would be mixed into load counters and make |
150 | | // num_rows_load_success() negative. |
151 | 21.1k | virtual bool _should_update_load_counters() const { return _is_load; } |
152 | | |
153 | | // Check if scanner is already closed, if not, mark it as closed. |
154 | | // Returns true if the scanner was successfully marked as closed (first time). |
155 | | // Returns false if the scanner was already closed. |
156 | | bool _try_close(); |
157 | | |
158 | | // Filter the output block finally. |
159 | | virtual Status _filter_output_block(Block* block); |
160 | | |
161 | | Status _do_projections(Block* origin_block, Block* output_block); |
162 | | |
163 | | private: |
164 | 21.2k | void _start_scan_cpu_timer() { |
165 | 21.2k | _cpu_watch.reset(); |
166 | 21.2k | _cpu_watch.start(); |
167 | 21.2k | } |
168 | | |
169 | 21.2k | void _update_wait_worker_timer() { _scanner_wait_worker_timer += _watch.elapsed_time(); } |
170 | | void _update_scan_cpu_timer(); |
171 | | |
172 | | public: |
173 | | // Call start_wait_worker_timer() when submit the scanner to the thread pool. |
174 | | // And call update_wait_worker_timer() when it is actually being executed. |
175 | 42.4k | void start_wait_worker_timer() { |
176 | 42.4k | _watch.reset(); |
177 | 42.4k | _watch.start(); |
178 | 42.4k | } |
179 | | |
180 | 21.2k | void resume() { |
181 | 21.2k | _update_wait_worker_timer(); |
182 | 21.2k | _start_scan_cpu_timer(); |
183 | 21.2k | } |
184 | 21.2k | void pause() { |
185 | 21.2k | _update_scan_cpu_timer(); |
186 | 21.2k | start_wait_worker_timer(); |
187 | 21.2k | } |
188 | 0 | int64_t get_time_cost_ns() const { return _per_scanner_timer; } |
189 | 0 | int64_t get_prepare_time_cost_ns() const { return _per_scanner_prepare_timer; } |
190 | 0 | int64_t get_open_time_cost_ns() const { return _per_scanner_open_timer; } |
191 | | |
192 | 0 | int64_t projection_time() const { return _projection_timer; } |
193 | 0 | int64_t get_rows_read() const { return _num_rows_read; } |
194 | | |
195 | 21.2k | bool has_prepared() const { return _has_prepared; } |
196 | | |
197 | | Status try_append_late_arrival_runtime_filter(); |
198 | | |
199 | 0 | int64_t get_scanner_wait_worker_timer() const { return _scanner_wait_worker_timer; } |
200 | | |
201 | | // Some counters need to be updated realtime, for example, workload group policy need |
202 | | // scan bytes to cancel the query exceed limit. |
203 | 0 | virtual void update_realtime_counters() {} |
204 | | |
205 | 42.5k | RuntimeState* runtime_state() { return _state; } |
206 | | |
207 | 21.2k | bool is_open() const { return _is_open; } |
208 | 21.1k | void set_opened() { _is_open = true; } |
209 | | |
210 | 12 | virtual doris::TabletStorageType get_storage_type() { |
211 | 12 | return doris::TabletStorageType::STORAGE_TYPE_REMOTE; |
212 | 12 | } |
213 | | |
214 | | // Returns true if this scanner's partition has been pruned by a runtime filter. |
215 | | // Overridden by OlapScanner to check partition pruning state. |
216 | 24 | virtual bool check_partition_pruned() const { return false; } |
217 | | |
218 | 0 | bool need_to_close() const { return _need_to_close; } |
219 | | |
220 | 21.1k | void mark_to_need_to_close() { |
221 | | // If the scanner is failed during init or open, then not need update counters |
222 | | // because the query is fail and the counter is useless. And it may core during |
223 | | // update counters. For example, update counters depend on scanner's tablet, but |
224 | | // the tablet == null when init failed. |
225 | 21.1k | if (_is_open) { |
226 | 21.1k | _collect_profile_before_close(); |
227 | 21.1k | } |
228 | 21.1k | _need_to_close = true; |
229 | 21.1k | } |
230 | | |
231 | 0 | void set_status_on_failure(const Status& st) { _status = st; } |
232 | | |
233 | 21.2k | int64_t limit() const { return _limit; } |
234 | | |
235 | 0 | auto get_block_avg_bytes() const { return _block_avg_bytes; } |
236 | | |
237 | 4.93k | void update_block_avg_bytes(size_t block_avg_bytes) { _block_avg_bytes = block_avg_bytes; } |
238 | | |
239 | | protected: |
240 | | RuntimeState* _state = nullptr; |
241 | | ScanLocalStateBase* _local_state = nullptr; |
242 | | |
243 | | // Set if scan node has sort limit info |
244 | | int64_t _limit = -1; |
245 | | |
246 | | RuntimeProfile* _profile = nullptr; |
247 | | |
248 | | const TupleDescriptor* _output_tuple_desc = nullptr; |
249 | | std::optional<std::reference_wrapper<const RowDescriptor>> _projection_output_row_descriptor; |
250 | | bool _has_projection = false; |
251 | | |
252 | | // If _input_tuple_desc is set, the scanner will read data into |
253 | | // this _input_block first, then convert to the output block. |
254 | | Block _input_block; |
255 | | |
256 | | bool _is_open = false; |
257 | | std::atomic<bool> _is_closed {false}; |
258 | | bool _need_to_close = false; |
259 | | Status _status; |
260 | | |
261 | | // If _applied_rf_num == _total_rf_num |
262 | | // means all runtime filters are arrived and applied. |
263 | | int _applied_rf_num = 0; |
264 | | int _total_rf_num = 0; |
265 | | // Cloned from _conjuncts of scan node. |
266 | | // It includes predicate in SQL and runtime filters. |
267 | | VExprContextSPtrs _conjuncts; |
268 | | VExprContextSPtrs _projections; |
269 | | // Used in common subexpression elimination to compute intermediate results. |
270 | | std::vector<VExprContextSPtrs> _intermediate_projections; |
271 | | Block _origin_block; |
272 | | Block _padding_block; |
273 | | |
274 | | VExprContextSPtrs _common_expr_ctxs_push_down; |
275 | | |
276 | | // num of rows read from scanner |
277 | | int64_t _num_rows_read = 0; |
278 | | |
279 | | int64_t _num_byte_read = 0; |
280 | | |
281 | | // num of rows return from scanner, after filter block |
282 | | int64_t _num_rows_return = 0; |
283 | | |
284 | | size_t _block_avg_bytes = 0; |
285 | | |
286 | | // Set true after counter is updated finally |
287 | | bool _has_updated_counter = false; |
288 | | |
289 | | // watch to count the time wait for scanner thread |
290 | | MonotonicStopWatch _watch; |
291 | | // Do not use ScopedTimer. There is no guarantee that, the counter |
292 | | ThreadCpuStopWatch _cpu_watch; |
293 | | int64_t _scanner_wait_worker_timer = 0; |
294 | | int64_t _scan_cpu_timer = 0; |
295 | | |
296 | | bool _is_load = false; |
297 | | |
298 | | bool _has_prepared = false; |
299 | | |
300 | | ScannerCounter _counter; |
301 | | int64_t _per_scanner_timer = 0; |
302 | | int64_t _per_scanner_prepare_timer = 0; |
303 | | int64_t _per_scanner_open_timer = 0; |
304 | | int64_t _projection_timer = 0; |
305 | | |
306 | | bool _should_stop = false; |
307 | | |
308 | | // Cached pointer to ScanOperator's remaining-limit counter. Null when |
309 | | // this scanner is on the topn path or the query has no LIMIT. |
310 | | std::atomic<int64_t>* _shared_scan_limit = nullptr; |
311 | | }; |
312 | | |
313 | | using ScannerSPtr = std::shared_ptr<Scanner>; |
314 | | |
315 | | } // namespace doris |