Coverage Report

Created: 2025-09-05 19:41

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
/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::function<void(const Status&)> cancel_fn;
65
};
66
67
enum class QuerySource {
68
    INTERNAL_FRONTEND,
69
    STREAM_LOAD,
70
    GROUP_COMMIT_LOAD,
71
    ROUTINE_LOAD,
72
    EXTERNAL_CONNECTOR
73
};
74
75
const std::string toString(QuerySource query_source);
76
77
// Save the common components of fragments in a query.
78
// Some components like DescriptorTbl may be very large
79
// that will slow down each execution of fragments when DeSer them every time.
80
class DescriptorTbl;
81
class QueryContext : public std::enable_shared_from_this<QueryContext> {
82
    ENABLE_FACTORY_CREATOR(QueryContext);
83
84
public:
85
    static std::shared_ptr<QueryContext> create(TUniqueId query_id, ExecEnv* exec_env,
86
                                                const TQueryOptions& query_options,
87
                                                TNetworkAddress coord_addr, bool is_nereids,
88
                                                TNetworkAddress current_connect_fe,
89
                                                QuerySource query_type);
90
91
    // use QueryContext::create, cannot be made private because of ENABLE_FACTORY_CREATOR::create_shared.
92
    QueryContext(TUniqueId query_id, ExecEnv* exec_env, const TQueryOptions& query_options,
93
                 TNetworkAddress coord_addr, bool is_nereids, TNetworkAddress current_connect_fe,
94
                 QuerySource query_type);
95
96
    ~QueryContext();
97
98
    void init_query_task_controller();
99
100
0
    ExecEnv* exec_env() const { return _exec_env; }
101
102
0
    bool is_timeout(timespec now) const {
103
0
        if (_timeout_second <= 0) {
104
0
            return false;
105
0
        }
106
0
        return _query_watcher.elapsed_time_seconds(now) > _timeout_second;
107
0
    }
108
109
0
    int64_t get_remaining_query_time_seconds() const {
110
0
        timespec now;
111
0
        clock_gettime(CLOCK_MONOTONIC, &now);
112
0
        if (is_timeout(now)) {
113
0
            return -1;
114
0
        }
115
0
        int64_t elapsed_seconds = _query_watcher.elapsed_time_seconds(now);
116
0
        return _timeout_second - elapsed_seconds;
117
0
    }
118
119
    void set_ready_to_execute(Status reason);
120
121
3.15M
    [[nodiscard]] bool is_cancelled() const { return !_exec_status.ok(); }
122
123
    void cancel_all_pipeline_context(const Status& reason, int fragment_id = -1);
124
    std::string print_all_pipeline_context();
125
    void set_pipeline_context(const int fragment_id,
126
                              std::shared_ptr<pipeline::PipelineFragmentContext> pip_ctx);
127
    void cancel(Status new_status, int fragment_id = -1);
128
129
20
    [[nodiscard]] Status exec_status() { return _exec_status.status(); }
130
131
    void set_execution_dependency_ready();
132
133
    void set_memory_sufficient(bool sufficient);
134
135
    void set_ready_to_execute_only();
136
137
22
    bool has_runtime_predicate(int source_node_id) {
138
22
        return _runtime_predicates.contains(source_node_id);
139
22
    }
140
141
0
    vectorized::RuntimePredicate& get_runtime_predicate(int source_node_id) {
142
0
        DCHECK(has_runtime_predicate(source_node_id));
143
0
        return _runtime_predicates.find(source_node_id)->second;
144
0
    }
145
146
0
    void init_runtime_predicates(const std::vector<TTopnFilterDesc>& topn_filter_descs) {
147
0
        for (auto desc : topn_filter_descs) {
148
0
            _runtime_predicates.try_emplace(desc.source_node_id, desc);
149
0
        }
150
0
    }
151
152
    Status set_workload_group(WorkloadGroupPtr& wg);
153
154
6
    int execution_timeout() const {
155
6
        return _query_options.__isset.execution_timeout ? _query_options.execution_timeout
156
6
                                                        : _query_options.query_timeout;
157
6
    }
158
159
115
    int32_t runtime_filter_wait_time_ms() const {
160
115
        return _query_options.runtime_filter_wait_time_ms;
161
115
    }
162
163
121
    bool runtime_filter_wait_infinitely() const {
164
121
        return _query_options.__isset.runtime_filter_wait_infinitely &&
165
121
               _query_options.runtime_filter_wait_infinitely;
166
121
    }
167
168
0
    int be_exec_version() const {
169
0
        if (!_query_options.__isset.be_exec_version) {
170
0
            return 0;
171
0
        }
172
0
        return _query_options.be_exec_version;
173
0
    }
174
175
0
    [[nodiscard]] int64_t get_fe_process_uuid() const {
176
0
        return _query_options.__isset.fe_process_uuid ? _query_options.fe_process_uuid : 0;
177
0
    }
178
179
0
    bool ignore_runtime_filter_error() const {
180
0
        return _query_options.__isset.ignore_runtime_filter_error
181
0
                       ? _query_options.ignore_runtime_filter_error
182
0
                       : false;
183
0
    }
184
185
0
    bool enable_force_spill() const {
186
0
        return _query_options.__isset.enable_force_spill && _query_options.enable_force_spill;
187
0
    }
188
6.90k
    const TQueryOptions& query_options() const { return _query_options; }
189
190
    // global runtime filter mgr, the runtime filter have remote target or
191
    // need local merge should regist here. before publish() or push_to_remote()
192
    // the runtime filter should do the local merge work
193
106
    RuntimeFilterMgr* runtime_filter_mgr() { return _runtime_filter_mgr.get(); }
194
195
72.1k
    TUniqueId query_id() const { return _query_id; }
196
197
0
    vectorized::SimplifiedScanScheduler* get_scan_scheduler() { return _scan_task_scheduler; }
198
199
0
    vectorized::SimplifiedScanScheduler* get_remote_scan_scheduler() {
200
0
        return _remote_scan_task_scheduler;
201
0
    }
202
203
72.1k
    pipeline::Dependency* get_execution_dependency() { return _execution_dependency.get(); }
204
72.1k
    pipeline::Dependency* get_memory_sufficient_dependency() {
205
72.1k
        return _memory_sufficient_dependency.get();
206
72.1k
    }
207
208
    doris::pipeline::TaskScheduler* get_pipe_exec_scheduler();
209
210
    void set_merge_controller_handler(
211
0
            std::shared_ptr<RuntimeFilterMergeControllerEntity>& handler) {
212
0
        _merge_controller_handler = handler;
213
0
    }
214
0
    std::shared_ptr<RuntimeFilterMergeControllerEntity> get_merge_controller_handler() const {
215
0
        return _merge_controller_handler;
216
0
    }
217
218
60
    bool is_nereids() const { return _is_nereids; }
219
220
122k
    WorkloadGroupPtr workload_group() const { return _resource_ctx->workload_group(); }
221
442k
    std::shared_ptr<MemTrackerLimiter> query_mem_tracker() const {
222
442k
        DCHECK(_resource_ctx->memory_context()->mem_tracker() != nullptr);
223
442k
        return _resource_ctx->memory_context()->mem_tracker();
224
442k
    }
225
226
11
    int32_t get_slot_count() const {
227
11
        return _query_options.__isset.query_slot_count ? _query_options.query_slot_count : 1;
228
11
    }
229
230
    DescriptorTbl* desc_tbl = nullptr;
231
    bool set_rsc_info = false;
232
    std::string user;
233
    std::string group;
234
    TNetworkAddress coord_addr;
235
    TNetworkAddress current_connect_fe;
236
    TQueryGlobals query_globals;
237
0
    const TQueryGlobals get_query_globals() const { return query_globals; }
238
239
    ObjectPool obj_pool;
240
241
5.86k
    std::shared_ptr<ResourceContext> resource_ctx() { return _resource_ctx; }
242
243
    // plan node id -> TFileScanRangeParams
244
    // only for file scan node
245
    std::map<int, TFileScanRangeParams> file_scan_range_params_map;
246
247
    void add_using_brpc_stub(const TNetworkAddress& network_address,
248
0
                             std::shared_ptr<PBackendService_Stub> brpc_stub) {
249
0
        if (network_address.port == 0) {
250
0
            return;
251
0
        }
252
0
        std::lock_guard<std::mutex> lock(_brpc_stubs_mutex);
253
0
        if (!_using_brpc_stubs.contains(network_address)) {
254
0
            _using_brpc_stubs.emplace(network_address, brpc_stub);
255
0
        }
256
257
0
        DCHECK_EQ(_using_brpc_stubs[network_address].get(), brpc_stub.get());
258
0
    }
259
260
72.4k
    void set_ai_resources(std::map<std::string, TAIResource> ai_resources) {
261
72.4k
        _ai_resources =
262
72.4k
                std::make_unique<std::map<std::string, TAIResource>>(std::move(ai_resources));
263
72.4k
    }
264
265
12
    const std::map<std::string, TAIResource>& get_ai_resources() const {
266
12
        if (_ai_resources == nullptr) {
267
0
            throw Status::InternalError("AI resources not found");
268
0
        }
269
12
        return *_ai_resources;
270
12
    }
271
272
    std::unordered_map<TNetworkAddress, std::shared_ptr<PBackendService_Stub>>
273
0
    get_using_brpc_stubs() {
274
0
        std::lock_guard<std::mutex> lock(_brpc_stubs_mutex);
275
0
        return _using_brpc_stubs;
276
0
    }
277
278
1.26k
    void set_low_memory_mode() {
279
        // will not return from low memory mode to non-low memory mode.
280
1.26k
        _resource_ctx->task_controller()->set_low_memory_mode(true);
281
1.26k
    }
282
1.02M
    bool low_memory_mode() { return _resource_ctx->task_controller()->low_memory_mode(); }
283
284
122k
    bool is_pure_load_task() {
285
122k
        return _query_source == QuerySource::STREAM_LOAD ||
286
122k
               _query_source == QuerySource::ROUTINE_LOAD ||
287
122k
               _query_source == QuerySource::GROUP_COMMIT_LOAD;
288
122k
    }
289
290
    void set_load_error_url(std::string error_url);
291
    std::string get_load_error_url();
292
293
private:
294
    friend class QueryTaskController;
295
296
    int _timeout_second;
297
    TUniqueId _query_id;
298
    ExecEnv* _exec_env = nullptr;
299
    MonotonicStopWatch _query_watcher;
300
    bool _is_nereids = false;
301
302
    std::shared_ptr<ResourceContext> _resource_ctx;
303
304
    void _init_resource_context();
305
    void _init_query_mem_tracker();
306
307
    std::unordered_map<int, vectorized::RuntimePredicate> _runtime_predicates;
308
309
    std::unique_ptr<RuntimeFilterMgr> _runtime_filter_mgr;
310
    const TQueryOptions _query_options;
311
312
    // All pipeline tasks use the same query context to report status. So we need a `_exec_status`
313
    // to report the real message if failed.
314
    AtomicStatus _exec_status;
315
316
    doris::pipeline::TaskScheduler* _task_scheduler = nullptr;
317
    vectorized::SimplifiedScanScheduler* _scan_task_scheduler = nullptr;
318
    vectorized::SimplifiedScanScheduler* _remote_scan_task_scheduler = nullptr;
319
    // This dependency indicates if the 2nd phase RPC received from FE.
320
    std::unique_ptr<pipeline::Dependency> _execution_dependency;
321
    // This dependency indicates if memory is sufficient to execute.
322
    std::unique_ptr<pipeline::Dependency> _memory_sufficient_dependency;
323
324
    // This shared ptr is never used. It is just a reference to hold the object.
325
    // There is a weak ptr in runtime filter manager to reference this object.
326
    std::shared_ptr<RuntimeFilterMergeControllerEntity> _merge_controller_handler;
327
328
    std::map<int, std::weak_ptr<pipeline::PipelineFragmentContext>> _fragment_id_to_pipeline_ctx;
329
    std::mutex _pipeline_map_write_lock;
330
331
    std::mutex _profile_mutex;
332
    timespec _query_arrival_timestamp;
333
    // Distinguish the query source, for query that comes from fe, we will have some memory structure on FE to
334
    // help us manage the query.
335
    QuerySource _query_source;
336
337
    std::mutex _brpc_stubs_mutex;
338
    std::unordered_map<TNetworkAddress, std::shared_ptr<PBackendService_Stub>> _using_brpc_stubs;
339
340
    // when fragment of pipeline is closed, it will register its profile to this map by using add_fragment_profile
341
    // flatten profile of one fragment:
342
    // Pipeline 0
343
    //      PipelineTask 0
344
    //              Operator 1
345
    //              Operator 2
346
    //              Scanner
347
    //      PipelineTask 1
348
    //              Operator 1
349
    //              Operator 2
350
    //              Scanner
351
    // Pipeline 1
352
    //      PipelineTask 2
353
    //              Operator 3
354
    //      PipelineTask 3
355
    //              Operator 3
356
    // fragment_id -> list<profile>
357
    std::unordered_map<int, std::vector<std::shared_ptr<TRuntimeProfileTree>>> _profile_map;
358
    std::unordered_map<int, std::shared_ptr<TRuntimeProfileTree>> _load_channel_profile_map;
359
360
    std::unique_ptr<std::map<std::string, TAIResource>> _ai_resources;
361
362
    void _report_query_profile();
363
364
    std::unordered_map<int, std::vector<std::shared_ptr<TRuntimeProfileTree>>>
365
    _collect_realtime_query_profile();
366
367
    std::mutex _error_url_lock;
368
    std::string _load_error_url;
369
370
public:
371
    // when fragment of pipeline is closed, it will register its profile to this map by using add_fragment_profile
372
    void add_fragment_profile(
373
            int fragment_id,
374
            const std::vector<std::shared_ptr<TRuntimeProfileTree>>& pipeline_profile,
375
            std::shared_ptr<TRuntimeProfileTree> load_channel_profile);
376
377
    TReportExecStatusParams get_realtime_exec_status();
378
379
122k
    bool enable_profile() const {
380
122k
        return _query_options.__isset.enable_profile && _query_options.enable_profile;
381
122k
    }
382
383
0
    timespec get_query_arrival_timestamp() const { return this->_query_arrival_timestamp; }
384
0
    QuerySource get_query_source() const { return this->_query_source; }
385
386
0
    const TQueryOptions get_query_options() const { return _query_options; }
387
};
388
389
} // namespace doris