Coverage Report

Created: 2026-03-16 17:20

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::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
7
    bool is_single_backend_query() const { return _is_single_backend_query; }
111
112
0
    void set_single_backend_query(bool is_single_backend_query) {
113
0
        _is_single_backend_query = is_single_backend_query;
114
0
    }
115
116
0
    int64_t get_remaining_query_time_seconds() const {
117
0
        timespec now;
118
0
        clock_gettime(CLOCK_MONOTONIC, &now);
119
0
        if (is_timeout(now)) {
120
0
            return -1;
121
0
        }
122
0
        int64_t elapsed_seconds = _query_watcher.elapsed_time_seconds(now);
123
0
        return _timeout_second - elapsed_seconds;
124
0
    }
125
126
    void set_ready_to_execute(Status reason);
127
128
2.07M
    [[nodiscard]] bool is_cancelled() const { return !_exec_status.ok(); }
129
130
    void cancel_all_pipeline_context(const Status& reason, int fragment_id = -1);
131
    std::string print_all_pipeline_context();
132
    void set_pipeline_context(const int fragment_id,
133
                              std::shared_ptr<pipeline::PipelineFragmentContext> pip_ctx);
134
    void cancel(Status new_status, int fragment_id = -1);
135
136
22
    [[nodiscard]] Status exec_status() { return _exec_status.status(); }
137
138
    void set_execution_dependency_ready();
139
140
    void set_memory_sufficient(bool sufficient);
141
142
    void set_ready_to_execute_only();
143
144
22
    bool has_runtime_predicate(int source_node_id) {
145
22
        return _runtime_predicates.contains(source_node_id);
146
22
    }
147
148
0
    vectorized::RuntimePredicate& get_runtime_predicate(int source_node_id) {
149
0
        DCHECK(has_runtime_predicate(source_node_id));
150
0
        return _runtime_predicates.find(source_node_id)->second;
151
0
    }
152
153
0
    void init_runtime_predicates(const std::vector<TTopnFilterDesc>& topn_filter_descs) {
154
0
        for (auto desc : topn_filter_descs) {
155
0
            _runtime_predicates.try_emplace(desc.source_node_id, desc);
156
0
        }
157
0
    }
158
159
    Status set_workload_group(WorkloadGroupPtr& wg);
160
161
121
    int execution_timeout() const {
162
121
        return _query_options.__isset.execution_timeout ? _query_options.execution_timeout
163
121
                                                        : _query_options.query_timeout;
164
121
    }
165
166
0
    int32_t runtime_filter_wait_time_ms() const {
167
0
        return _query_options.runtime_filter_wait_time_ms;
168
0
    }
169
170
121
    bool runtime_filter_wait_infinitely() const {
171
121
        return _query_options.__isset.runtime_filter_wait_infinitely &&
172
121
               _query_options.runtime_filter_wait_infinitely;
173
121
    }
174
175
0
    int be_exec_version() const {
176
0
        if (!_query_options.__isset.be_exec_version) {
177
0
            return 0;
178
0
        }
179
0
        return _query_options.be_exec_version;
180
0
    }
181
182
0
    [[nodiscard]] int64_t get_fe_process_uuid() const {
183
0
        return _query_options.__isset.fe_process_uuid ? _query_options.fe_process_uuid : 0;
184
0
    }
185
186
0
    bool ignore_runtime_filter_error() const {
187
0
        return _query_options.__isset.ignore_runtime_filter_error
188
0
                       ? _query_options.ignore_runtime_filter_error
189
0
                       : false;
190
0
    }
191
192
0
    bool enable_force_spill() const {
193
0
        return _query_options.__isset.enable_force_spill && _query_options.enable_force_spill;
194
0
    }
195
95.9k
    const TQueryOptions& query_options() const { return _query_options; }
196
0
    bool should_be_shuffled_agg(int node_id) const {
197
0
        return _query_options.__isset.shuffled_agg_ids &&
198
0
               std::any_of(_query_options.shuffled_agg_ids.begin(),
199
0
                           _query_options.shuffled_agg_ids.end(),
200
0
                           [&](const int id) -> bool { return id == node_id; });
201
0
    }
202
203
    // global runtime filter mgr, the runtime filter have remote target or
204
    // need local merge should regist here. before publish() or push_to_remote()
205
    // the runtime filter should do the local merge work
206
107
    RuntimeFilterMgr* runtime_filter_mgr() { return _runtime_filter_mgr.get(); }
207
208
72.1k
    TUniqueId query_id() const { return _query_id; }
209
210
14
    vectorized::ScannerScheduler* get_scan_scheduler() { return _scan_task_scheduler; }
211
212
0
    vectorized::ScannerScheduler* get_remote_scan_scheduler() {
213
0
        return _remote_scan_task_scheduler;
214
0
    }
215
216
72.1k
    pipeline::Dependency* get_execution_dependency() { return _execution_dependency.get(); }
217
72.1k
    pipeline::Dependency* get_memory_sufficient_dependency() {
218
72.1k
        return _memory_sufficient_dependency.get();
219
72.1k
    }
220
221
    doris::pipeline::TaskScheduler* get_pipe_exec_scheduler();
222
223
    void set_merge_controller_handler(
224
0
            std::shared_ptr<RuntimeFilterMergeControllerEntity>& handler) {
225
0
        _merge_controller_handler = handler;
226
0
    }
227
0
    std::shared_ptr<RuntimeFilterMergeControllerEntity> get_merge_controller_handler() const {
228
0
        return _merge_controller_handler;
229
0
    }
230
231
64
    bool is_nereids() const { return _is_nereids; }
232
233
121k
    WorkloadGroupPtr workload_group() const { return _resource_ctx->workload_group(); }
234
583k
    std::shared_ptr<MemTrackerLimiter> query_mem_tracker() const {
235
583k
        DCHECK(_resource_ctx->memory_context()->mem_tracker() != nullptr);
236
583k
        return _resource_ctx->memory_context()->mem_tracker();
237
583k
    }
238
239
11
    int32_t get_slot_count() const {
240
11
        return _query_options.__isset.query_slot_count ? _query_options.query_slot_count : 1;
241
11
    }
242
243
    DescriptorTbl* desc_tbl = nullptr;
244
    bool set_rsc_info = false;
245
    std::string user;
246
    std::string group;
247
    TNetworkAddress coord_addr;
248
    TNetworkAddress current_connect_fe;
249
    TQueryGlobals query_globals;
250
0
    const TQueryGlobals get_query_globals() const { return query_globals; }
251
252
    ObjectPool obj_pool;
253
254
50.1k
    std::shared_ptr<ResourceContext> resource_ctx() { return _resource_ctx; }
255
256
    // plan node id -> TFileScanRangeParams
257
    // only for file scan node
258
    std::map<int, TFileScanRangeParams> file_scan_range_params_map;
259
260
    void add_using_brpc_stub(const TNetworkAddress& network_address,
261
0
                             std::shared_ptr<PBackendService_Stub> brpc_stub) {
262
0
        if (network_address.port == 0) {
263
0
            return;
264
0
        }
265
0
        std::lock_guard<std::mutex> lock(_brpc_stubs_mutex);
266
0
        if (!_using_brpc_stubs.contains(network_address)) {
267
0
            _using_brpc_stubs.emplace(network_address, brpc_stub);
268
0
        }
269
270
0
        DCHECK_EQ(_using_brpc_stubs[network_address].get(), brpc_stub.get());
271
0
    }
272
273
72.5k
    void set_ai_resources(std::map<std::string, TAIResource> ai_resources) {
274
72.5k
        _ai_resources =
275
72.5k
                std::make_shared<std::map<std::string, TAIResource>>(std::move(ai_resources));
276
72.5k
    }
277
278
63
    const std::shared_ptr<std::map<std::string, TAIResource>>& get_ai_resources() const {
279
63
        return _ai_resources;
280
63
    }
281
282
    std::unordered_map<TNetworkAddress, std::shared_ptr<PBackendService_Stub>>
283
0
    get_using_brpc_stubs() {
284
0
        std::lock_guard<std::mutex> lock(_brpc_stubs_mutex);
285
0
        return _using_brpc_stubs;
286
0
    }
287
288
0
    void set_low_memory_mode() {
289
        // will not return from low memory mode to non-low memory mode.
290
0
        _resource_ctx->task_controller()->set_low_memory_mode(true);
291
0
    }
292
486k
    bool low_memory_mode() { return _resource_ctx->task_controller()->low_memory_mode(); }
293
294
121k
    bool is_pure_load_task() {
295
121k
        return _query_source == QuerySource::STREAM_LOAD ||
296
121k
               _query_source == QuerySource::ROUTINE_LOAD ||
297
121k
               _query_source == QuerySource::GROUP_COMMIT_LOAD;
298
121k
    }
299
300
    void set_load_error_url(std::string error_url);
301
    std::string get_load_error_url();
302
    void set_first_error_msg(std::string error_msg);
303
    std::string get_first_error_msg();
304
305
private:
306
    friend class QueryTaskController;
307
308
    int _timeout_second;
309
    TUniqueId _query_id;
310
    ExecEnv* _exec_env = nullptr;
311
    MonotonicStopWatch _query_watcher;
312
    bool _is_nereids = false;
313
314
    std::shared_ptr<ResourceContext> _resource_ctx;
315
316
    void _init_resource_context();
317
    void _init_query_mem_tracker();
318
319
    std::unordered_map<int, vectorized::RuntimePredicate> _runtime_predicates;
320
321
    std::unique_ptr<RuntimeFilterMgr> _runtime_filter_mgr;
322
    const TQueryOptions _query_options;
323
324
    // All pipeline tasks use the same query context to report status. So we need a `_exec_status`
325
    // to report the real message if failed.
326
    AtomicStatus _exec_status;
327
328
    doris::pipeline::TaskScheduler* _task_scheduler = nullptr;
329
    vectorized::ScannerScheduler* _scan_task_scheduler = nullptr;
330
    vectorized::ScannerScheduler* _remote_scan_task_scheduler = nullptr;
331
    // This dependency indicates if the 2nd phase RPC received from FE.
332
    std::unique_ptr<pipeline::Dependency> _execution_dependency;
333
    // This dependency indicates if memory is sufficient to execute.
334
    std::unique_ptr<pipeline::Dependency> _memory_sufficient_dependency;
335
336
    // This shared ptr is never used. It is just a reference to hold the object.
337
    // There is a weak ptr in runtime filter manager to reference this object.
338
    std::shared_ptr<RuntimeFilterMergeControllerEntity> _merge_controller_handler;
339
340
    std::map<int, std::weak_ptr<pipeline::PipelineFragmentContext>> _fragment_id_to_pipeline_ctx;
341
    std::mutex _pipeline_map_write_lock;
342
343
    std::mutex _profile_mutex;
344
    timespec _query_arrival_timestamp;
345
    // Distinguish the query source, for query that comes from fe, we will have some memory structure on FE to
346
    // help us manage the query.
347
    QuerySource _query_source;
348
349
    std::mutex _brpc_stubs_mutex;
350
    std::unordered_map<TNetworkAddress, std::shared_ptr<PBackendService_Stub>> _using_brpc_stubs;
351
352
    // when fragment of pipeline is closed, it will register its profile to this map by using add_fragment_profile
353
    // flatten profile of one fragment:
354
    // Pipeline 0
355
    //      PipelineTask 0
356
    //              Operator 1
357
    //              Operator 2
358
    //              Scanner
359
    //      PipelineTask 1
360
    //              Operator 1
361
    //              Operator 2
362
    //              Scanner
363
    // Pipeline 1
364
    //      PipelineTask 2
365
    //              Operator 3
366
    //      PipelineTask 3
367
    //              Operator 3
368
    // fragment_id -> list<profile>
369
    std::unordered_map<int, std::vector<std::shared_ptr<TRuntimeProfileTree>>> _profile_map;
370
    std::unordered_map<int, std::shared_ptr<TRuntimeProfileTree>> _load_channel_profile_map;
371
372
    std::shared_ptr<std::map<std::string, TAIResource>> _ai_resources;
373
374
    void _report_query_profile();
375
376
    std::unordered_map<int, std::vector<std::shared_ptr<TRuntimeProfileTree>>>
377
    _collect_realtime_query_profile();
378
379
    std::mutex _error_url_lock;
380
    std::string _load_error_url;
381
    std::string _first_error_msg;
382
383
    bool _is_single_backend_query = false;
384
385
    // file cache context holders
386
    std::vector<io::BlockFileCache::QueryFileCacheContextHolderPtr> _query_context_holders;
387
388
public:
389
    // when fragment of pipeline is closed, it will register its profile to this map by using add_fragment_profile
390
    void add_fragment_profile(
391
            int fragment_id,
392
            const std::vector<std::shared_ptr<TRuntimeProfileTree>>& pipeline_profile,
393
            std::shared_ptr<TRuntimeProfileTree> load_channel_profile);
394
395
    TReportExecStatusParams get_realtime_exec_status();
396
397
121k
    bool enable_profile() const {
398
121k
        return _query_options.__isset.enable_profile && _query_options.enable_profile;
399
121k
    }
400
401
0
    timespec get_query_arrival_timestamp() const { return this->_query_arrival_timestamp; }
402
0
    QuerySource get_query_source() const { return this->_query_source; }
403
404
0
    const TQueryOptions get_query_options() const { return _query_options; }
405
};
406
407
} // namespace doris