Coverage Report

Created: 2026-03-22 11:38

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 "common/status.h"
36
#include "exec/runtime_filter/runtime_filter_mgr.h"
37
#include "exec/scan/scanner_scheduler.h"
38
#include "runtime/exec_env.h"
39
#include "runtime/memory/mem_tracker_limiter.h"
40
#include "runtime/runtime_predicate.h"
41
#include "runtime/workload_group/workload_group.h"
42
#include "runtime/workload_management/resource_context.h"
43
#include "util/hash_util.hpp"
44
#include "util/threadpool.h"
45
46
namespace doris {
47
48
class PipelineFragmentContext;
49
class PipelineTask;
50
class Dependency;
51
52
struct ReportStatusRequest {
53
    const Status status;
54
    std::vector<RuntimeState*> runtime_states;
55
    bool done;
56
    TNetworkAddress coord_addr;
57
    TUniqueId query_id;
58
    int fragment_id;
59
    TUniqueId fragment_instance_id;
60
    int backend_num;
61
    RuntimeState* runtime_state;
62
    std::string load_error_url;
63
    std::string first_error_msg;
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
7
    bool is_single_backend_query() const { return _is_single_backend_query; }
110
111
0
    void set_single_backend_query(bool is_single_backend_query) {
112
0
        _is_single_backend_query = is_single_backend_query;
113
0
    }
114
115
0
    int64_t get_remaining_query_time_seconds() const {
116
0
        timespec now;
117
0
        clock_gettime(CLOCK_MONOTONIC, &now);
118
0
        if (is_timeout(now)) {
119
0
            return -1;
120
0
        }
121
0
        int64_t elapsed_seconds = _query_watcher.elapsed_time_seconds(now);
122
0
        return _timeout_second - elapsed_seconds;
123
0
    }
124
125
    void set_ready_to_execute(Status reason);
126
127
2.43M
    [[nodiscard]] bool is_cancelled() const { return !_exec_status.ok(); }
128
129
    void cancel_all_pipeline_context(const Status& reason, int fragment_id = -1);
130
    std::string print_all_pipeline_context();
131
    void set_pipeline_context(const int fragment_id,
132
                              std::shared_ptr<PipelineFragmentContext> pip_ctx);
133
    void cancel(Status new_status, int fragment_id = -1);
134
135
22
    [[nodiscard]] Status exec_status() { return _exec_status.status(); }
136
137
    void set_execution_dependency_ready();
138
139
    void set_memory_sufficient(bool sufficient);
140
141
    void set_ready_to_execute_only();
142
143
22
    bool has_runtime_predicate(int source_node_id) {
144
22
        return _runtime_predicates.contains(source_node_id);
145
22
    }
146
147
0
    RuntimePredicate& get_runtime_predicate(int source_node_id) {
148
0
        DCHECK(has_runtime_predicate(source_node_id));
149
0
        return _runtime_predicates.find(source_node_id)->second;
150
0
    }
151
152
0
    void init_runtime_predicates(const std::vector<TTopnFilterDesc>& topn_filter_descs) {
153
0
        for (auto desc : topn_filter_descs) {
154
0
            _runtime_predicates.try_emplace(desc.source_node_id, desc);
155
0
        }
156
0
    }
157
158
    Status set_workload_group(WorkloadGroupPtr& wg);
159
160
121
    int execution_timeout() const {
161
121
        return _query_options.__isset.execution_timeout ? _query_options.execution_timeout
162
121
                                                        : _query_options.query_timeout;
163
121
    }
164
165
0
    int32_t runtime_filter_wait_time_ms() const {
166
0
        return _query_options.runtime_filter_wait_time_ms;
167
0
    }
168
169
121
    bool runtime_filter_wait_infinitely() const {
170
121
        return _query_options.__isset.runtime_filter_wait_infinitely &&
171
121
               _query_options.runtime_filter_wait_infinitely;
172
121
    }
173
174
0
    int be_exec_version() const {
175
0
        if (!_query_options.__isset.be_exec_version) {
176
0
            return 0;
177
0
        }
178
0
        return _query_options.be_exec_version;
179
0
    }
180
181
0
    [[nodiscard]] int64_t get_fe_process_uuid() const {
182
0
        return _query_options.__isset.fe_process_uuid ? _query_options.fe_process_uuid : 0;
183
0
    }
184
185
0
    bool ignore_runtime_filter_error() const {
186
0
        return _query_options.__isset.ignore_runtime_filter_error
187
0
                       ? _query_options.ignore_runtime_filter_error
188
0
                       : false;
189
0
    }
190
191
0
    bool enable_force_spill() const {
192
0
        return _query_options.__isset.enable_force_spill && _query_options.enable_force_spill;
193
0
    }
194
92.9k
    const TQueryOptions& query_options() const { return _query_options; }
195
0
    bool should_be_shuffled_agg(int node_id) const {
196
0
        return _query_options.__isset.shuffled_agg_ids &&
197
0
               std::any_of(_query_options.shuffled_agg_ids.begin(),
198
0
                           _query_options.shuffled_agg_ids.end(),
199
0
                           [&](const int id) -> bool { return id == node_id; });
200
0
    }
201
202
    // global runtime filter mgr, the runtime filter have remote target or
203
    // need local merge should regist here. before publish() or push_to_remote()
204
    // the runtime filter should do the local merge work
205
107
    RuntimeFilterMgr* runtime_filter_mgr() { return _runtime_filter_mgr.get(); }
206
207
72.1k
    TUniqueId query_id() const { return _query_id; }
208
209
14
    ScannerScheduler* get_scan_scheduler() { return _scan_task_scheduler; }
210
211
0
    ScannerScheduler* get_remote_scan_scheduler() {
212
0
        return _remote_scan_task_scheduler;
213
0
    }
214
215
72.1k
    Dependency* get_execution_dependency() { return _execution_dependency.get(); }
216
72.1k
    Dependency* get_memory_sufficient_dependency() {
217
72.1k
        return _memory_sufficient_dependency.get();
218
72.1k
    }
219
220
    doris::TaskScheduler* get_pipe_exec_scheduler();
221
222
    void set_merge_controller_handler(
223
0
            std::shared_ptr<RuntimeFilterMergeControllerEntity>& handler) {
224
0
        _merge_controller_handler = handler;
225
0
    }
226
0
    std::shared_ptr<RuntimeFilterMergeControllerEntity> get_merge_controller_handler() const {
227
0
        return _merge_controller_handler;
228
0
    }
229
230
64
    bool is_nereids() const { return _is_nereids; }
231
232
121k
    WorkloadGroupPtr workload_group() const { return _resource_ctx->workload_group(); }
233
683k
    std::shared_ptr<MemTrackerLimiter> query_mem_tracker() const {
234
683k
        DCHECK(_resource_ctx->memory_context()->mem_tracker() != nullptr);
235
683k
        return _resource_ctx->memory_context()->mem_tracker();
236
683k
    }
237
238
11
    int32_t get_slot_count() const {
239
11
        return _query_options.__isset.query_slot_count ? _query_options.query_slot_count : 1;
240
11
    }
241
242
    DescriptorTbl* desc_tbl = nullptr;
243
    bool set_rsc_info = false;
244
    std::string user;
245
    std::string group;
246
    TNetworkAddress coord_addr;
247
    TNetworkAddress current_connect_fe;
248
    TQueryGlobals query_globals;
249
0
    const TQueryGlobals get_query_globals() const { return query_globals; }
250
251
    ObjectPool obj_pool;
252
253
48.7k
    std::shared_ptr<ResourceContext> resource_ctx() { return _resource_ctx; }
254
255
    // plan node id -> TFileScanRangeParams
256
    // only for file scan node
257
    std::map<int, TFileScanRangeParams> file_scan_range_params_map;
258
259
    void add_using_brpc_stub(const TNetworkAddress& network_address,
260
0
                             std::shared_ptr<PBackendService_Stub> brpc_stub) {
261
0
        if (network_address.port == 0) {
262
0
            return;
263
0
        }
264
0
        std::lock_guard<std::mutex> lock(_brpc_stubs_mutex);
265
0
        if (!_using_brpc_stubs.contains(network_address)) {
266
0
            _using_brpc_stubs.emplace(network_address, brpc_stub);
267
0
        }
268
269
0
        DCHECK_EQ(_using_brpc_stubs[network_address].get(), brpc_stub.get());
270
0
    }
271
272
72.5k
    void set_ai_resources(std::map<std::string, TAIResource> ai_resources) {
273
72.5k
        _ai_resources =
274
72.5k
                std::make_shared<std::map<std::string, TAIResource>>(std::move(ai_resources));
275
72.5k
    }
276
277
63
    const std::shared_ptr<std::map<std::string, TAIResource>>& get_ai_resources() const {
278
63
        return _ai_resources;
279
63
    }
280
281
    std::unordered_map<TNetworkAddress, std::shared_ptr<PBackendService_Stub>>
282
0
    get_using_brpc_stubs() {
283
0
        std::lock_guard<std::mutex> lock(_brpc_stubs_mutex);
284
0
        return _using_brpc_stubs;
285
0
    }
286
287
0
    void set_low_memory_mode() {
288
        // will not return from low memory mode to non-low memory mode.
289
0
        _resource_ctx->task_controller()->set_low_memory_mode(true);
290
0
    }
291
667k
    bool low_memory_mode() { return _resource_ctx->task_controller()->low_memory_mode(); }
292
293
121k
    bool is_pure_load_task() {
294
121k
        return _query_source == QuerySource::STREAM_LOAD ||
295
121k
               _query_source == QuerySource::ROUTINE_LOAD ||
296
121k
               _query_source == QuerySource::GROUP_COMMIT_LOAD;
297
121k
    }
298
299
    void set_load_error_url(std::string error_url);
300
    std::string get_load_error_url();
301
    void set_first_error_msg(std::string error_msg);
302
    std::string get_first_error_msg();
303
304
private:
305
    friend class QueryTaskController;
306
307
    int _timeout_second;
308
    TUniqueId _query_id;
309
    ExecEnv* _exec_env = nullptr;
310
    MonotonicStopWatch _query_watcher;
311
    bool _is_nereids = false;
312
313
    std::shared_ptr<ResourceContext> _resource_ctx;
314
315
    void _init_resource_context();
316
    void _init_query_mem_tracker();
317
318
    std::unordered_map<int, RuntimePredicate> _runtime_predicates;
319
320
    std::unique_ptr<RuntimeFilterMgr> _runtime_filter_mgr;
321
    const TQueryOptions _query_options;
322
323
    // All pipeline tasks use the same query context to report status. So we need a `_exec_status`
324
    // to report the real message if failed.
325
    AtomicStatus _exec_status;
326
327
    doris::TaskScheduler* _task_scheduler = nullptr;
328
    ScannerScheduler* _scan_task_scheduler = nullptr;
329
    ScannerScheduler* _remote_scan_task_scheduler = nullptr;
330
    // This dependency indicates if the 2nd phase RPC received from FE.
331
    std::unique_ptr<Dependency> _execution_dependency;
332
    // This dependency indicates if memory is sufficient to execute.
333
    std::unique_ptr<Dependency> _memory_sufficient_dependency;
334
335
    // This shared ptr is never used. It is just a reference to hold the object.
336
    // There is a weak ptr in runtime filter manager to reference this object.
337
    std::shared_ptr<RuntimeFilterMergeControllerEntity> _merge_controller_handler;
338
339
    std::map<int, std::weak_ptr<PipelineFragmentContext>> _fragment_id_to_pipeline_ctx;
340
    std::mutex _pipeline_map_write_lock;
341
342
    std::mutex _profile_mutex;
343
    timespec _query_arrival_timestamp;
344
    // Distinguish the query source, for query that comes from fe, we will have some memory structure on FE to
345
    // help us manage the query.
346
    QuerySource _query_source;
347
348
    std::mutex _brpc_stubs_mutex;
349
    std::unordered_map<TNetworkAddress, std::shared_ptr<PBackendService_Stub>> _using_brpc_stubs;
350
351
    // when fragment of pipeline is closed, it will register its profile to this map by using add_fragment_profile
352
    // flatten profile of one fragment:
353
    // Pipeline 0
354
    //      PipelineTask 0
355
    //              Operator 1
356
    //              Operator 2
357
    //              Scanner
358
    //      PipelineTask 1
359
    //              Operator 1
360
    //              Operator 2
361
    //              Scanner
362
    // Pipeline 1
363
    //      PipelineTask 2
364
    //              Operator 3
365
    //      PipelineTask 3
366
    //              Operator 3
367
    // fragment_id -> list<profile>
368
    std::unordered_map<int, std::vector<std::shared_ptr<TRuntimeProfileTree>>> _profile_map;
369
    std::unordered_map<int, std::shared_ptr<TRuntimeProfileTree>> _load_channel_profile_map;
370
371
    std::shared_ptr<std::map<std::string, TAIResource>> _ai_resources;
372
373
    void _report_query_profile();
374
375
    std::unordered_map<int, std::vector<std::shared_ptr<TRuntimeProfileTree>>>
376
    _collect_realtime_query_profile();
377
378
    std::mutex _error_url_lock;
379
    std::string _load_error_url;
380
    std::string _first_error_msg;
381
382
    bool _is_single_backend_query = false;
383
384
    // file cache context holders
385
    std::vector<io::BlockFileCache::QueryFileCacheContextHolderPtr> _query_context_holders;
386
387
public:
388
    // when fragment of pipeline is closed, it will register its profile to this map by using add_fragment_profile
389
    void add_fragment_profile(
390
            int fragment_id,
391
            const std::vector<std::shared_ptr<TRuntimeProfileTree>>& pipeline_profile,
392
            std::shared_ptr<TRuntimeProfileTree> load_channel_profile);
393
394
    TReportExecStatusParams get_realtime_exec_status();
395
396
121k
    bool enable_profile() const {
397
121k
        return _query_options.__isset.enable_profile && _query_options.enable_profile;
398
121k
    }
399
400
0
    timespec get_query_arrival_timestamp() const { return this->_query_arrival_timestamp; }
401
0
    QuerySource get_query_source() const { return this->_query_source; }
402
403
0
    const TQueryOptions get_query_options() const { return _query_options; }
404
};
405
406
} // namespace doris