/root/doris/be/src/runtime/query_context.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/PaloInternalService_types.h> |
21 | | #include <gen_cpp/RuntimeProfile_types.h> |
22 | | #include <gen_cpp/Types_types.h> |
23 | | #include <glog/logging.h> |
24 | | |
25 | | #include <atomic> |
26 | | #include <cstdint> |
27 | | #include <memory> |
28 | | #include <mutex> |
29 | | #include <string> |
30 | | #include <unordered_map> |
31 | | |
32 | | #include "common/config.h" |
33 | | #include "common/factory_creator.h" |
34 | | #include "common/object_pool.h" |
35 | | #include "runtime/exec_env.h" |
36 | | #include "runtime/memory/mem_tracker_limiter.h" |
37 | | #include "runtime/runtime_predicate.h" |
38 | | #include "runtime/workload_management/resource_context.h" |
39 | | #include "runtime_filter/runtime_filter_mgr.h" |
40 | | #include "util/hash_util.hpp" |
41 | | #include "util/threadpool.h" |
42 | | #include "vec/exec/scan/scanner_scheduler.h" |
43 | | #include "workload_group/workload_group.h" |
44 | | |
45 | | namespace doris { |
46 | | |
47 | | namespace pipeline { |
48 | | class PipelineFragmentContext; |
49 | | class PipelineTask; |
50 | | class Dependency; |
51 | | } // namespace pipeline |
52 | | |
53 | | struct ReportStatusRequest { |
54 | | const Status status; |
55 | | std::vector<RuntimeState*> runtime_states; |
56 | | bool done; |
57 | | TNetworkAddress coord_addr; |
58 | | TUniqueId query_id; |
59 | | int fragment_id; |
60 | | TUniqueId fragment_instance_id; |
61 | | int backend_num; |
62 | | RuntimeState* runtime_state; |
63 | | std::string load_error_url; |
64 | | std::string first_error_msg; |
65 | | std::function<void(const Status&)> cancel_fn; |
66 | | }; |
67 | | |
68 | | enum class QuerySource { |
69 | | INTERNAL_FRONTEND, |
70 | | STREAM_LOAD, |
71 | | GROUP_COMMIT_LOAD, |
72 | | ROUTINE_LOAD, |
73 | | EXTERNAL_CONNECTOR |
74 | | }; |
75 | | |
76 | | const std::string toString(QuerySource query_source); |
77 | | |
78 | | // Save the common components of fragments in a query. |
79 | | // Some components like DescriptorTbl may be very large |
80 | | // that will slow down each execution of fragments when DeSer them every time. |
81 | | class DescriptorTbl; |
82 | | class QueryContext : public std::enable_shared_from_this<QueryContext> { |
83 | | ENABLE_FACTORY_CREATOR(QueryContext); |
84 | | |
85 | | public: |
86 | | static std::shared_ptr<QueryContext> create(TUniqueId query_id, ExecEnv* exec_env, |
87 | | const TQueryOptions& query_options, |
88 | | TNetworkAddress coord_addr, bool is_nereids, |
89 | | TNetworkAddress current_connect_fe, |
90 | | QuerySource query_type); |
91 | | |
92 | | // use QueryContext::create, cannot be made private because of ENABLE_FACTORY_CREATOR::create_shared. |
93 | | QueryContext(TUniqueId query_id, ExecEnv* exec_env, const TQueryOptions& query_options, |
94 | | TNetworkAddress coord_addr, bool is_nereids, TNetworkAddress current_connect_fe, |
95 | | QuerySource query_type); |
96 | | |
97 | | ~QueryContext(); |
98 | | |
99 | | void init_query_task_controller(); |
100 | | |
101 | 0 | ExecEnv* exec_env() const { return _exec_env; } |
102 | | |
103 | 0 | bool is_timeout(timespec now) const { |
104 | 0 | if (_timeout_second <= 0) { |
105 | 0 | return false; |
106 | 0 | } |
107 | 0 | return _query_watcher.elapsed_time_seconds(now) > _timeout_second; |
108 | 0 | } |
109 | | |
110 | 0 | int64_t get_remaining_query_time_seconds() const { |
111 | 0 | timespec now; |
112 | 0 | clock_gettime(CLOCK_MONOTONIC, &now); |
113 | 0 | if (is_timeout(now)) { |
114 | 0 | return -1; |
115 | 0 | } |
116 | 0 | int64_t elapsed_seconds = _query_watcher.elapsed_time_seconds(now); |
117 | 0 | return _timeout_second - elapsed_seconds; |
118 | 0 | } |
119 | | |
120 | | void set_ready_to_execute(Status reason); |
121 | | |
122 | 3.24M | [[nodiscard]] bool is_cancelled() const { return !_exec_status.ok(); } |
123 | | |
124 | | void cancel_all_pipeline_context(const Status& reason, int fragment_id = -1); |
125 | | std::string print_all_pipeline_context(); |
126 | | void set_pipeline_context(const int fragment_id, |
127 | | std::shared_ptr<pipeline::PipelineFragmentContext> pip_ctx); |
128 | | void cancel(Status new_status, int fragment_id = -1); |
129 | | |
130 | 20 | [[nodiscard]] Status exec_status() { return _exec_status.status(); } |
131 | | |
132 | | void set_execution_dependency_ready(); |
133 | | |
134 | | void set_memory_sufficient(bool sufficient); |
135 | | |
136 | | void set_ready_to_execute_only(); |
137 | | |
138 | 22 | bool has_runtime_predicate(int source_node_id) { |
139 | 22 | return _runtime_predicates.contains(source_node_id); |
140 | 22 | } |
141 | | |
142 | 0 | vectorized::RuntimePredicate& get_runtime_predicate(int source_node_id) { |
143 | 0 | DCHECK(has_runtime_predicate(source_node_id)); |
144 | 0 | return _runtime_predicates.find(source_node_id)->second; |
145 | 0 | } |
146 | | |
147 | 0 | void init_runtime_predicates(const std::vector<TTopnFilterDesc>& topn_filter_descs) { |
148 | 0 | for (auto desc : topn_filter_descs) { |
149 | 0 | _runtime_predicates.try_emplace(desc.source_node_id, desc); |
150 | 0 | } |
151 | 0 | } |
152 | | |
153 | | Status set_workload_group(WorkloadGroupPtr& wg); |
154 | | |
155 | 6 | int execution_timeout() const { |
156 | 6 | return _query_options.__isset.execution_timeout ? _query_options.execution_timeout |
157 | 6 | : _query_options.query_timeout; |
158 | 6 | } |
159 | | |
160 | 115 | int32_t runtime_filter_wait_time_ms() const { |
161 | 115 | return _query_options.runtime_filter_wait_time_ms; |
162 | 115 | } |
163 | | |
164 | 121 | bool runtime_filter_wait_infinitely() const { |
165 | 121 | return _query_options.__isset.runtime_filter_wait_infinitely && |
166 | 121 | _query_options.runtime_filter_wait_infinitely; |
167 | 121 | } |
168 | | |
169 | 0 | int be_exec_version() const { |
170 | 0 | if (!_query_options.__isset.be_exec_version) { |
171 | 0 | return 0; |
172 | 0 | } |
173 | 0 | return _query_options.be_exec_version; |
174 | 0 | } |
175 | | |
176 | 0 | [[nodiscard]] int64_t get_fe_process_uuid() const { |
177 | 0 | return _query_options.__isset.fe_process_uuid ? _query_options.fe_process_uuid : 0; |
178 | 0 | } |
179 | | |
180 | 0 | bool ignore_runtime_filter_error() const { |
181 | 0 | return _query_options.__isset.ignore_runtime_filter_error |
182 | 0 | ? _query_options.ignore_runtime_filter_error |
183 | 0 | : false; |
184 | 0 | } |
185 | | |
186 | 0 | bool enable_force_spill() const { |
187 | 0 | return _query_options.__isset.enable_force_spill && _query_options.enable_force_spill; |
188 | 0 | } |
189 | 7.49k | const TQueryOptions& query_options() const { return _query_options; } |
190 | | |
191 | | // global runtime filter mgr, the runtime filter have remote target or |
192 | | // need local merge should regist here. before publish() or push_to_remote() |
193 | | // the runtime filter should do the local merge work |
194 | 107 | RuntimeFilterMgr* runtime_filter_mgr() { return _runtime_filter_mgr.get(); } |
195 | | |
196 | 72.1k | TUniqueId query_id() const { return _query_id; } |
197 | | |
198 | 0 | vectorized::SimplifiedScanScheduler* get_scan_scheduler() { return _scan_task_scheduler; } |
199 | | |
200 | 0 | vectorized::SimplifiedScanScheduler* get_remote_scan_scheduler() { |
201 | 0 | return _remote_scan_task_scheduler; |
202 | 0 | } |
203 | | |
204 | 72.1k | pipeline::Dependency* get_execution_dependency() { return _execution_dependency.get(); } |
205 | 72.1k | pipeline::Dependency* get_memory_sufficient_dependency() { |
206 | 72.1k | return _memory_sufficient_dependency.get(); |
207 | 72.1k | } |
208 | | |
209 | | doris::pipeline::TaskScheduler* get_pipe_exec_scheduler(); |
210 | | |
211 | | void set_merge_controller_handler( |
212 | 0 | std::shared_ptr<RuntimeFilterMergeControllerEntity>& handler) { |
213 | 0 | _merge_controller_handler = handler; |
214 | 0 | } |
215 | 0 | std::shared_ptr<RuntimeFilterMergeControllerEntity> get_merge_controller_handler() const { |
216 | 0 | return _merge_controller_handler; |
217 | 0 | } |
218 | | |
219 | 60 | bool is_nereids() const { return _is_nereids; } |
220 | | |
221 | 122k | WorkloadGroupPtr workload_group() const { return _resource_ctx->workload_group(); } |
222 | 447k | std::shared_ptr<MemTrackerLimiter> query_mem_tracker() const { |
223 | 447k | DCHECK(_resource_ctx->memory_context()->mem_tracker() != nullptr); |
224 | 447k | return _resource_ctx->memory_context()->mem_tracker(); |
225 | 447k | } |
226 | | |
227 | 11 | int32_t get_slot_count() const { |
228 | 11 | return _query_options.__isset.query_slot_count ? _query_options.query_slot_count : 1; |
229 | 11 | } |
230 | | |
231 | | DescriptorTbl* desc_tbl = nullptr; |
232 | | bool set_rsc_info = false; |
233 | | std::string user; |
234 | | std::string group; |
235 | | TNetworkAddress coord_addr; |
236 | | TNetworkAddress current_connect_fe; |
237 | | TQueryGlobals query_globals; |
238 | 0 | const TQueryGlobals get_query_globals() const { return query_globals; } |
239 | | |
240 | | ObjectPool obj_pool; |
241 | | |
242 | 6.19k | std::shared_ptr<ResourceContext> resource_ctx() { return _resource_ctx; } |
243 | | |
244 | | // plan node id -> TFileScanRangeParams |
245 | | // only for file scan node |
246 | | std::map<int, TFileScanRangeParams> file_scan_range_params_map; |
247 | | |
248 | | void add_using_brpc_stub(const TNetworkAddress& network_address, |
249 | 0 | std::shared_ptr<PBackendService_Stub> brpc_stub) { |
250 | 0 | if (network_address.port == 0) { |
251 | 0 | return; |
252 | 0 | } |
253 | 0 | std::lock_guard<std::mutex> lock(_brpc_stubs_mutex); |
254 | 0 | if (!_using_brpc_stubs.contains(network_address)) { |
255 | 0 | _using_brpc_stubs.emplace(network_address, brpc_stub); |
256 | 0 | } |
257 | |
|
258 | 0 | DCHECK_EQ(_using_brpc_stubs[network_address].get(), brpc_stub.get()); |
259 | 0 | } |
260 | | |
261 | 72.5k | void set_ai_resources(std::map<std::string, TAIResource> ai_resources) { |
262 | 72.5k | _ai_resources = |
263 | 72.5k | std::make_unique<std::map<std::string, TAIResource>>(std::move(ai_resources)); |
264 | 72.5k | } |
265 | | |
266 | 15 | const std::map<std::string, TAIResource>& get_ai_resources() const { |
267 | 15 | if (_ai_resources == nullptr) { |
268 | 0 | throw Status::InternalError("AI resources not found"); |
269 | 0 | } |
270 | 15 | return *_ai_resources; |
271 | 15 | } |
272 | | |
273 | | std::unordered_map<TNetworkAddress, std::shared_ptr<PBackendService_Stub>> |
274 | 0 | get_using_brpc_stubs() { |
275 | 0 | std::lock_guard<std::mutex> lock(_brpc_stubs_mutex); |
276 | 0 | return _using_brpc_stubs; |
277 | 0 | } |
278 | | |
279 | 1.51k | void set_low_memory_mode() { |
280 | | // will not return from low memory mode to non-low memory mode. |
281 | 1.51k | _resource_ctx->task_controller()->set_low_memory_mode(true); |
282 | 1.51k | } |
283 | 1.07M | bool low_memory_mode() { return _resource_ctx->task_controller()->low_memory_mode(); } |
284 | | |
285 | 122k | bool is_pure_load_task() { |
286 | 122k | return _query_source == QuerySource::STREAM_LOAD || |
287 | 122k | _query_source == QuerySource::ROUTINE_LOAD || |
288 | 122k | _query_source == QuerySource::GROUP_COMMIT_LOAD; |
289 | 122k | } |
290 | | |
291 | | void set_load_error_url(std::string error_url); |
292 | | std::string get_load_error_url(); |
293 | | void set_first_error_msg(std::string error_msg); |
294 | | std::string get_first_error_msg(); |
295 | | |
296 | | private: |
297 | | friend class QueryTaskController; |
298 | | |
299 | | int _timeout_second; |
300 | | TUniqueId _query_id; |
301 | | ExecEnv* _exec_env = nullptr; |
302 | | MonotonicStopWatch _query_watcher; |
303 | | bool _is_nereids = false; |
304 | | |
305 | | std::shared_ptr<ResourceContext> _resource_ctx; |
306 | | |
307 | | void _init_resource_context(); |
308 | | void _init_query_mem_tracker(); |
309 | | |
310 | | std::unordered_map<int, vectorized::RuntimePredicate> _runtime_predicates; |
311 | | |
312 | | std::unique_ptr<RuntimeFilterMgr> _runtime_filter_mgr; |
313 | | const TQueryOptions _query_options; |
314 | | |
315 | | // All pipeline tasks use the same query context to report status. So we need a `_exec_status` |
316 | | // to report the real message if failed. |
317 | | AtomicStatus _exec_status; |
318 | | |
319 | | doris::pipeline::TaskScheduler* _task_scheduler = nullptr; |
320 | | vectorized::SimplifiedScanScheduler* _scan_task_scheduler = nullptr; |
321 | | vectorized::SimplifiedScanScheduler* _remote_scan_task_scheduler = nullptr; |
322 | | // This dependency indicates if the 2nd phase RPC received from FE. |
323 | | std::unique_ptr<pipeline::Dependency> _execution_dependency; |
324 | | // This dependency indicates if memory is sufficient to execute. |
325 | | std::unique_ptr<pipeline::Dependency> _memory_sufficient_dependency; |
326 | | |
327 | | // This shared ptr is never used. It is just a reference to hold the object. |
328 | | // There is a weak ptr in runtime filter manager to reference this object. |
329 | | std::shared_ptr<RuntimeFilterMergeControllerEntity> _merge_controller_handler; |
330 | | |
331 | | std::map<int, std::weak_ptr<pipeline::PipelineFragmentContext>> _fragment_id_to_pipeline_ctx; |
332 | | std::mutex _pipeline_map_write_lock; |
333 | | |
334 | | std::mutex _profile_mutex; |
335 | | timespec _query_arrival_timestamp; |
336 | | // Distinguish the query source, for query that comes from fe, we will have some memory structure on FE to |
337 | | // help us manage the query. |
338 | | QuerySource _query_source; |
339 | | |
340 | | std::mutex _brpc_stubs_mutex; |
341 | | std::unordered_map<TNetworkAddress, std::shared_ptr<PBackendService_Stub>> _using_brpc_stubs; |
342 | | |
343 | | // when fragment of pipeline is closed, it will register its profile to this map by using add_fragment_profile |
344 | | // flatten profile of one fragment: |
345 | | // Pipeline 0 |
346 | | // PipelineTask 0 |
347 | | // Operator 1 |
348 | | // Operator 2 |
349 | | // Scanner |
350 | | // PipelineTask 1 |
351 | | // Operator 1 |
352 | | // Operator 2 |
353 | | // Scanner |
354 | | // Pipeline 1 |
355 | | // PipelineTask 2 |
356 | | // Operator 3 |
357 | | // PipelineTask 3 |
358 | | // Operator 3 |
359 | | // fragment_id -> list<profile> |
360 | | std::unordered_map<int, std::vector<std::shared_ptr<TRuntimeProfileTree>>> _profile_map; |
361 | | std::unordered_map<int, std::shared_ptr<TRuntimeProfileTree>> _load_channel_profile_map; |
362 | | |
363 | | std::unique_ptr<std::map<std::string, TAIResource>> _ai_resources; |
364 | | |
365 | | void _report_query_profile(); |
366 | | |
367 | | std::unordered_map<int, std::vector<std::shared_ptr<TRuntimeProfileTree>>> |
368 | | _collect_realtime_query_profile(); |
369 | | |
370 | | std::mutex _error_url_lock; |
371 | | std::string _load_error_url; |
372 | | std::string _first_error_msg; |
373 | | |
374 | | public: |
375 | | // when fragment of pipeline is closed, it will register its profile to this map by using add_fragment_profile |
376 | | void add_fragment_profile( |
377 | | int fragment_id, |
378 | | const std::vector<std::shared_ptr<TRuntimeProfileTree>>& pipeline_profile, |
379 | | std::shared_ptr<TRuntimeProfileTree> load_channel_profile); |
380 | | |
381 | | TReportExecStatusParams get_realtime_exec_status(); |
382 | | |
383 | 122k | bool enable_profile() const { |
384 | 122k | return _query_options.__isset.enable_profile && _query_options.enable_profile; |
385 | 122k | } |
386 | | |
387 | 0 | timespec get_query_arrival_timestamp() const { return this->_query_arrival_timestamp; } |
388 | 0 | QuerySource get_query_source() const { return this->_query_source; } |
389 | | |
390 | 0 | const TQueryOptions get_query_options() const { return _query_options; } |
391 | | }; |
392 | | |
393 | | } // namespace doris |