Coverage Report

Created: 2026-05-09 05:26

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
be/src/exec/pipeline/pipeline_fragment_context.cpp
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
#include "exec/pipeline/pipeline_fragment_context.h"
19
20
#include <gen_cpp/DataSinks_types.h>
21
#include <gen_cpp/FrontendService.h>
22
#include <gen_cpp/FrontendService_types.h>
23
#include <gen_cpp/PaloInternalService_types.h>
24
#include <gen_cpp/PlanNodes_types.h>
25
#include <pthread.h>
26
27
#include <algorithm>
28
#include <cstdlib>
29
// IWYU pragma: no_include <bits/chrono.h>
30
#include <fmt/format.h>
31
#include <thrift/Thrift.h>
32
#include <thrift/protocol/TDebugProtocol.h>
33
#include <thrift/transport/TTransportException.h>
34
35
#include <chrono> // IWYU pragma: keep
36
#include <map>
37
#include <memory>
38
#include <ostream>
39
#include <utility>
40
41
#include "cloud/config.h"
42
#include "common/cast_set.h"
43
#include "common/config.h"
44
#include "common/exception.h"
45
#include "common/logging.h"
46
#include "common/status.h"
47
#include "exec/exchange/local_exchange_sink_operator.h"
48
#include "exec/exchange/local_exchange_source_operator.h"
49
#include "exec/exchange/local_exchanger.h"
50
#include "exec/exchange/vdata_stream_mgr.h"
51
#include "exec/operator/aggregation_sink_operator.h"
52
#include "exec/operator/aggregation_source_operator.h"
53
#include "exec/operator/analytic_sink_operator.h"
54
#include "exec/operator/analytic_source_operator.h"
55
#include "exec/operator/assert_num_rows_operator.h"
56
#include "exec/operator/blackhole_sink_operator.h"
57
#include "exec/operator/bucketed_aggregation_sink_operator.h"
58
#include "exec/operator/bucketed_aggregation_source_operator.h"
59
#include "exec/operator/cache_sink_operator.h"
60
#include "exec/operator/cache_source_operator.h"
61
#include "exec/operator/datagen_operator.h"
62
#include "exec/operator/dict_sink_operator.h"
63
#include "exec/operator/distinct_streaming_aggregation_operator.h"
64
#include "exec/operator/empty_set_operator.h"
65
#include "exec/operator/exchange_sink_operator.h"
66
#include "exec/operator/exchange_source_operator.h"
67
#include "exec/operator/file_scan_operator.h"
68
#include "exec/operator/group_commit_block_sink_operator.h"
69
#include "exec/operator/group_commit_scan_operator.h"
70
#include "exec/operator/hashjoin_build_sink.h"
71
#include "exec/operator/hashjoin_probe_operator.h"
72
#include "exec/operator/hive_table_sink_operator.h"
73
#include "exec/operator/iceberg_delete_sink_operator.h"
74
#include "exec/operator/iceberg_merge_sink_operator.h"
75
#include "exec/operator/iceberg_table_sink_operator.h"
76
#include "exec/operator/jdbc_scan_operator.h"
77
#include "exec/operator/jdbc_table_sink_operator.h"
78
#include "exec/operator/local_merge_sort_source_operator.h"
79
#include "exec/operator/materialization_opertor.h"
80
#include "exec/operator/maxcompute_table_sink_operator.h"
81
#include "exec/operator/memory_scratch_sink_operator.h"
82
#include "exec/operator/meta_scan_operator.h"
83
#include "exec/operator/multi_cast_data_stream_sink.h"
84
#include "exec/operator/multi_cast_data_stream_source.h"
85
#include "exec/operator/nested_loop_join_build_operator.h"
86
#include "exec/operator/nested_loop_join_probe_operator.h"
87
#include "exec/operator/olap_scan_operator.h"
88
#include "exec/operator/olap_table_sink_operator.h"
89
#include "exec/operator/olap_table_sink_v2_operator.h"
90
#include "exec/operator/partition_sort_sink_operator.h"
91
#include "exec/operator/partition_sort_source_operator.h"
92
#include "exec/operator/partitioned_aggregation_sink_operator.h"
93
#include "exec/operator/partitioned_aggregation_source_operator.h"
94
#include "exec/operator/partitioned_hash_join_probe_operator.h"
95
#include "exec/operator/partitioned_hash_join_sink_operator.h"
96
#include "exec/operator/rec_cte_anchor_sink_operator.h"
97
#include "exec/operator/rec_cte_scan_operator.h"
98
#include "exec/operator/rec_cte_sink_operator.h"
99
#include "exec/operator/rec_cte_source_operator.h"
100
#include "exec/operator/repeat_operator.h"
101
#include "exec/operator/result_file_sink_operator.h"
102
#include "exec/operator/result_sink_operator.h"
103
#include "exec/operator/schema_scan_operator.h"
104
#include "exec/operator/select_operator.h"
105
#include "exec/operator/set_probe_sink_operator.h"
106
#include "exec/operator/set_sink_operator.h"
107
#include "exec/operator/set_source_operator.h"
108
#include "exec/operator/sort_sink_operator.h"
109
#include "exec/operator/sort_source_operator.h"
110
#include "exec/operator/spill_iceberg_table_sink_operator.h"
111
#include "exec/operator/spill_sort_sink_operator.h"
112
#include "exec/operator/spill_sort_source_operator.h"
113
#include "exec/operator/streaming_aggregation_operator.h"
114
#include "exec/operator/table_function_operator.h"
115
#include "exec/operator/tvf_table_sink_operator.h"
116
#include "exec/operator/union_sink_operator.h"
117
#include "exec/operator/union_source_operator.h"
118
#include "exec/pipeline/dependency.h"
119
#include "exec/pipeline/pipeline_task.h"
120
#include "exec/pipeline/task_scheduler.h"
121
#include "exec/runtime_filter/runtime_filter_mgr.h"
122
#include "exec/sort/topn_sorter.h"
123
#include "exec/spill/spill_file.h"
124
#include "io/fs/stream_load_pipe.h"
125
#include "load/stream_load/new_load_stream_mgr.h"
126
#include "runtime/exec_env.h"
127
#include "runtime/fragment_mgr.h"
128
#include "runtime/result_buffer_mgr.h"
129
#include "runtime/runtime_state.h"
130
#include "runtime/thread_context.h"
131
#include "service/backend_options.h"
132
#include "util/client_cache.h"
133
#include "util/countdown_latch.h"
134
#include "util/debug_util.h"
135
#include "util/network_util.h"
136
#include "util/uid_util.h"
137
138
namespace doris {
139
PipelineFragmentContext::PipelineFragmentContext(
140
        TUniqueId query_id, const TPipelineFragmentParams& request,
141
        std::shared_ptr<QueryContext> query_ctx, ExecEnv* exec_env,
142
        const std::function<void(RuntimeState*, Status*)>& call_back)
143
433k
        : _query_id(std::move(query_id)),
144
433k
          _fragment_id(request.fragment_id),
145
433k
          _exec_env(exec_env),
146
433k
          _query_ctx(std::move(query_ctx)),
147
433k
          _call_back(call_back),
148
433k
          _is_report_on_cancel(true),
149
433k
          _params(request),
150
433k
          _parallel_instances(_params.__isset.parallel_instances ? _params.parallel_instances : 0),
151
433k
          _need_notify_close(request.__isset.need_notify_close ? request.need_notify_close
152
433k
                                                               : false) {
153
433k
    _fragment_watcher.start();
154
433k
}
155
156
434k
PipelineFragmentContext::~PipelineFragmentContext() {
157
434k
    LOG_INFO("PipelineFragmentContext::~PipelineFragmentContext")
158
434k
            .tag("query_id", print_id(_query_id))
159
434k
            .tag("fragment_id", _fragment_id);
160
434k
    _release_resource();
161
434k
    {
162
        // The memory released by the query end is recorded in the query mem tracker.
163
434k
        SCOPED_SWITCH_THREAD_MEM_TRACKER_LIMITER(_query_ctx->query_mem_tracker());
164
434k
        _runtime_state.reset();
165
434k
        _query_ctx.reset();
166
434k
    }
167
434k
}
168
169
72
bool PipelineFragmentContext::is_timeout(timespec now) const {
170
72
    if (_timeout <= 0) {
171
0
        return false;
172
0
    }
173
72
    return _fragment_watcher.elapsed_time_seconds(now) > _timeout;
174
72
}
175
176
// notify_close() transitions the PFC from "waiting for external close notification" to
177
// "self-managed close". For recursive CTE fragments, the old PFC is kept alive until
178
// the rerun_fragment(wait_for_destroy) RPC calls this to trigger shutdown.
179
// Returns true if all tasks have already closed (i.e., the PFC can be safely destroyed).
180
9.65k
bool PipelineFragmentContext::notify_close() {
181
9.65k
    bool all_closed = false;
182
9.65k
    bool need_remove = false;
183
9.65k
    {
184
9.65k
        std::lock_guard<std::mutex> l(_task_mutex);
185
9.65k
        if (_closed_tasks >= _total_tasks) {
186
3.48k
            if (_need_notify_close) {
187
                // Fragment was cancelled and waiting for notify to close.
188
                // Record that we need to remove from fragment mgr, but do it
189
                // after releasing _task_mutex to avoid ABBA deadlock with
190
                // dump_pipeline_tasks() (which acquires _pipeline_map lock
191
                // first, then _task_mutex via debug_string()).
192
3.41k
                need_remove = true;
193
3.41k
            }
194
3.48k
            all_closed = true;
195
3.48k
        }
196
        // make fragment release by self after cancel
197
9.65k
        _need_notify_close = false;
198
9.65k
    }
199
9.65k
    if (need_remove) {
200
3.41k
        _exec_env->fragment_mgr()->remove_pipeline_context({_query_id, _fragment_id});
201
3.41k
    }
202
9.65k
    return all_closed;
203
9.65k
}
204
205
// Must not add lock in this method. Because it will call query ctx cancel. And
206
// QueryCtx cancel will call fragment ctx cancel. And Also Fragment ctx's running
207
// Method like exchange sink buffer will call query ctx cancel. If we add lock here
208
// There maybe dead lock.
209
6.17k
void PipelineFragmentContext::cancel(const Status reason) {
210
6.17k
    LOG_INFO("PipelineFragmentContext::cancel")
211
6.17k
            .tag("query_id", print_id(_query_id))
212
6.17k
            .tag("fragment_id", _fragment_id)
213
6.17k
            .tag("reason", reason.to_string());
214
6.17k
    if (notify_close()) {
215
88
        return;
216
88
    }
217
    // Timeout is a special error code, we need print current stack to debug timeout issue.
218
6.09k
    if (reason.is<ErrorCode::TIMEOUT>()) {
219
1
        auto dbg_str = fmt::format("PipelineFragmentContext is cancelled due to timeout:\n{}",
220
1
                                   debug_string());
221
1
        LOG_LONG_STRING(WARNING, dbg_str);
222
1
    }
223
224
    // `ILLEGAL_STATE` means queries this fragment belongs to was not found in FE (maybe finished)
225
6.09k
    if (reason.is<ErrorCode::ILLEGAL_STATE>()) {
226
0
        LOG_WARNING("PipelineFragmentContext is cancelled due to illegal state : {}",
227
0
                    debug_string());
228
0
    }
229
230
6.09k
    if (reason.is<ErrorCode::MEM_LIMIT_EXCEEDED>() || reason.is<ErrorCode::MEM_ALLOC_FAILED>()) {
231
0
        print_profile("cancel pipeline, reason: " + reason.to_string());
232
0
    }
233
234
6.09k
    if (auto error_url = get_load_error_url(); !error_url.empty()) {
235
21
        _query_ctx->set_load_error_url(error_url);
236
21
    }
237
238
6.09k
    if (auto first_error_msg = get_first_error_msg(); !first_error_msg.empty()) {
239
21
        _query_ctx->set_first_error_msg(first_error_msg);
240
21
    }
241
242
6.09k
    _query_ctx->cancel(reason, _fragment_id);
243
6.09k
    if (reason.is<ErrorCode::LIMIT_REACH>()) {
244
420
        _is_report_on_cancel = false;
245
5.67k
    } else {
246
24.8k
        for (auto& id : _fragment_instance_ids) {
247
24.8k
            LOG(WARNING) << "PipelineFragmentContext cancel instance: " << print_id(id);
248
24.8k
        }
249
5.67k
    }
250
    // Get pipe from new load stream manager and send cancel to it or the fragment may hang to wait read from pipe
251
    // For stream load the fragment's query_id == load id, it is set in FE.
252
6.09k
    auto stream_load_ctx = _exec_env->new_load_stream_mgr()->get(_query_id);
253
6.09k
    if (stream_load_ctx != nullptr) {
254
30
        stream_load_ctx->pipe->cancel(reason.to_string());
255
        // Set error URL here because after pipe is cancelled, stream load execution may return early.
256
        // We need to set the error URL at this point to ensure error information is properly
257
        // propagated to the client.
258
30
        stream_load_ctx->error_url = get_load_error_url();
259
30
        stream_load_ctx->first_error_msg = get_first_error_msg();
260
30
    }
261
262
25.9k
    for (auto& tasks : _tasks) {
263
57.8k
        for (auto& task : tasks) {
264
57.8k
            task.first->unblock_all_dependencies();
265
57.8k
        }
266
25.9k
    }
267
6.09k
}
268
269
674k
PipelinePtr PipelineFragmentContext::add_pipeline(PipelinePtr parent, int idx) {
270
674k
    PipelineId id = _next_pipeline_id++;
271
674k
    auto pipeline = std::make_shared<Pipeline>(
272
674k
            id, parent ? std::min(parent->num_tasks(), _num_instances) : _num_instances,
273
674k
            parent ? parent->num_tasks() : _num_instances);
274
674k
    if (idx >= 0) {
275
109k
        _pipelines.insert(_pipelines.begin() + idx, pipeline);
276
564k
    } else {
277
564k
        _pipelines.emplace_back(pipeline);
278
564k
    }
279
674k
    if (parent) {
280
234k
        parent->set_children(pipeline);
281
234k
    }
282
674k
    return pipeline;
283
674k
}
284
285
433k
Status PipelineFragmentContext::_build_and_prepare_full_pipeline(ThreadPool* thread_pool) {
286
433k
    {
287
433k
        SCOPED_TIMER(_build_pipelines_timer);
288
        // 2. Build pipelines with operators in this fragment.
289
433k
        auto root_pipeline = add_pipeline();
290
433k
        RETURN_IF_ERROR(_build_pipelines(_runtime_state->obj_pool(), *_query_ctx->desc_tbl,
291
433k
                                         &_root_op, root_pipeline));
292
293
        // 3. Create sink operator
294
433k
        if (!_params.fragment.__isset.output_sink) {
295
0
            return Status::InternalError("No output sink in this fragment!");
296
0
        }
297
433k
        RETURN_IF_ERROR(_create_data_sink(_runtime_state->obj_pool(), _params.fragment.output_sink,
298
433k
                                          _params.fragment.output_exprs, _params,
299
433k
                                          root_pipeline->output_row_desc(), _runtime_state.get(),
300
433k
                                          *_desc_tbl, root_pipeline->id()));
301
433k
        RETURN_IF_ERROR(_sink->init(_params.fragment.output_sink));
302
433k
        RETURN_IF_ERROR(root_pipeline->set_sink(_sink));
303
304
563k
        for (PipelinePtr& pipeline : _pipelines) {
305
563k
            DCHECK(pipeline->sink() != nullptr) << pipeline->operators().size();
306
563k
            RETURN_IF_ERROR(pipeline->sink()->set_child(pipeline->operators().back()));
307
563k
        }
308
433k
    }
309
    // 4. Build local exchanger
310
433k
    if (_runtime_state->enable_local_shuffle()) {
311
431k
        SCOPED_TIMER(_plan_local_exchanger_timer);
312
431k
        RETURN_IF_ERROR(_plan_local_exchange(_params.num_buckets,
313
431k
                                             _params.bucket_seq_to_instance_idx,
314
431k
                                             _params.shuffle_idx_to_instance_idx));
315
431k
    }
316
317
    // 5. Initialize global states in pipelines.
318
674k
    for (PipelinePtr& pipeline : _pipelines) {
319
674k
        SCOPED_TIMER(_prepare_all_pipelines_timer);
320
674k
        pipeline->children().clear();
321
674k
        RETURN_IF_ERROR(pipeline->prepare(_runtime_state.get()));
322
674k
    }
323
324
432k
    {
325
432k
        SCOPED_TIMER(_build_tasks_timer);
326
        // 6. Build pipeline tasks and initialize local state.
327
432k
        RETURN_IF_ERROR(_build_pipeline_tasks(thread_pool));
328
432k
    }
329
330
432k
    return Status::OK();
331
432k
}
332
333
433k
Status PipelineFragmentContext::prepare(ThreadPool* thread_pool) {
334
433k
    if (_prepared) {
335
0
        return Status::InternalError("Already prepared");
336
0
    }
337
433k
    if (_params.__isset.query_options && _params.query_options.__isset.execution_timeout) {
338
433k
        _timeout = _params.query_options.execution_timeout;
339
433k
    }
340
341
433k
    _fragment_level_profile = std::make_unique<RuntimeProfile>("PipelineContext");
342
433k
    _prepare_timer = ADD_TIMER(_fragment_level_profile, "PrepareTime");
343
433k
    SCOPED_TIMER(_prepare_timer);
344
433k
    _build_pipelines_timer = ADD_TIMER(_fragment_level_profile, "BuildPipelinesTime");
345
433k
    _init_context_timer = ADD_TIMER(_fragment_level_profile, "InitContextTime");
346
433k
    _plan_local_exchanger_timer = ADD_TIMER(_fragment_level_profile, "PlanLocalLocalExchangerTime");
347
433k
    _build_tasks_timer = ADD_TIMER(_fragment_level_profile, "BuildTasksTime");
348
433k
    _prepare_all_pipelines_timer = ADD_TIMER(_fragment_level_profile, "PrepareAllPipelinesTime");
349
433k
    {
350
433k
        SCOPED_TIMER(_init_context_timer);
351
433k
        cast_set(_num_instances, _params.local_params.size());
352
433k
        _total_instances =
353
433k
                _params.__isset.total_instances ? _params.total_instances : _num_instances;
354
355
433k
        auto* fragment_context = this;
356
357
433k
        if (_params.query_options.__isset.is_report_success) {
358
433k
            fragment_context->set_is_report_success(_params.query_options.is_report_success);
359
433k
        }
360
361
        // 1. Set up the global runtime state.
362
433k
        _runtime_state = RuntimeState::create_unique(
363
433k
                _params.query_id, _params.fragment_id, _params.query_options,
364
433k
                _query_ctx->query_globals, _exec_env, _query_ctx.get());
365
433k
        _runtime_state->set_task_execution_context(shared_from_this());
366
433k
        SCOPED_SWITCH_THREAD_MEM_TRACKER_LIMITER(_runtime_state->query_mem_tracker());
367
433k
        if (_params.__isset.backend_id) {
368
431k
            _runtime_state->set_backend_id(_params.backend_id);
369
431k
        }
370
433k
        if (_params.__isset.import_label) {
371
240
            _runtime_state->set_import_label(_params.import_label);
372
240
        }
373
433k
        if (_params.__isset.db_name) {
374
192
            _runtime_state->set_db_name(_params.db_name);
375
192
        }
376
433k
        if (_params.__isset.load_job_id) {
377
0
            _runtime_state->set_load_job_id(_params.load_job_id);
378
0
        }
379
380
433k
        if (_params.is_simplified_param) {
381
146k
            _desc_tbl = _query_ctx->desc_tbl;
382
287k
        } else {
383
287k
            DCHECK(_params.__isset.desc_tbl);
384
287k
            RETURN_IF_ERROR(DescriptorTbl::create(_runtime_state->obj_pool(), _params.desc_tbl,
385
287k
                                                  &_desc_tbl));
386
287k
        }
387
433k
        _runtime_state->set_desc_tbl(_desc_tbl);
388
433k
        _runtime_state->set_num_per_fragment_instances(_params.num_senders);
389
433k
        _runtime_state->set_load_stream_per_node(_params.load_stream_per_node);
390
433k
        _runtime_state->set_total_load_streams(_params.total_load_streams);
391
433k
        _runtime_state->set_num_local_sink(_params.num_local_sink);
392
393
        // init fragment_instance_ids
394
433k
        const auto target_size = _params.local_params.size();
395
433k
        _fragment_instance_ids.resize(target_size);
396
1.65M
        for (size_t i = 0; i < _params.local_params.size(); i++) {
397
1.21M
            auto fragment_instance_id = _params.local_params[i].fragment_instance_id;
398
1.21M
            _fragment_instance_ids[i] = fragment_instance_id;
399
1.21M
        }
400
433k
    }
401
402
433k
    RETURN_IF_ERROR(_build_and_prepare_full_pipeline(thread_pool));
403
404
432k
    _init_next_report_time();
405
406
432k
    _prepared = true;
407
432k
    return Status::OK();
408
433k
}
409
410
Status PipelineFragmentContext::_build_pipeline_tasks_for_instance(
411
        int instance_idx,
412
1.21M
        const std::vector<std::shared_ptr<RuntimeProfile>>& pipeline_id_to_profile) {
413
1.21M
    const auto& local_params = _params.local_params[instance_idx];
414
1.21M
    auto fragment_instance_id = local_params.fragment_instance_id;
415
1.21M
    auto runtime_filter_mgr = std::make_unique<RuntimeFilterMgr>(false);
416
1.21M
    std::map<PipelineId, PipelineTask*> pipeline_id_to_task;
417
1.21M
    auto get_shared_state = [&](PipelinePtr pipeline)
418
1.21M
            -> std::map<int, std::pair<std::shared_ptr<BasicSharedState>,
419
1.98M
                                       std::vector<std::shared_ptr<Dependency>>>> {
420
1.98M
        std::map<int, std::pair<std::shared_ptr<BasicSharedState>,
421
1.98M
                                std::vector<std::shared_ptr<Dependency>>>>
422
1.98M
                shared_state_map;
423
2.61M
        for (auto& op : pipeline->operators()) {
424
2.61M
            auto source_id = op->operator_id();
425
2.61M
            if (auto iter = _op_id_to_shared_state.find(source_id);
426
2.61M
                iter != _op_id_to_shared_state.end()) {
427
740k
                shared_state_map.insert({source_id, iter->second});
428
740k
            }
429
2.61M
        }
430
1.98M
        for (auto sink_to_source_id : pipeline->sink()->dests_id()) {
431
1.98M
            if (auto iter = _op_id_to_shared_state.find(sink_to_source_id);
432
1.98M
                iter != _op_id_to_shared_state.end()) {
433
278k
                shared_state_map.insert({sink_to_source_id, iter->second});
434
278k
            }
435
1.98M
        }
436
1.98M
        return shared_state_map;
437
1.98M
    };
438
439
3.65M
    for (size_t pip_idx = 0; pip_idx < _pipelines.size(); pip_idx++) {
440
2.43M
        auto& pipeline = _pipelines[pip_idx];
441
2.43M
        if (pipeline->num_tasks() > 1 || instance_idx == 0) {
442
1.97M
            auto task_runtime_state = RuntimeState::create_unique(
443
1.97M
                    local_params.fragment_instance_id, _params.query_id, _params.fragment_id,
444
1.97M
                    _params.query_options, _query_ctx->query_globals, _exec_env, _query_ctx.get());
445
1.97M
            {
446
                // Initialize runtime state for this task
447
1.97M
                task_runtime_state->set_query_mem_tracker(_query_ctx->query_mem_tracker());
448
449
1.97M
                task_runtime_state->set_task_execution_context(shared_from_this());
450
1.97M
                task_runtime_state->set_be_number(local_params.backend_num);
451
452
1.97M
                if (_params.__isset.backend_id) {
453
1.97M
                    task_runtime_state->set_backend_id(_params.backend_id);
454
1.97M
                }
455
1.97M
                if (_params.__isset.import_label) {
456
241
                    task_runtime_state->set_import_label(_params.import_label);
457
241
                }
458
1.97M
                if (_params.__isset.db_name) {
459
193
                    task_runtime_state->set_db_name(_params.db_name);
460
193
                }
461
1.97M
                if (_params.__isset.load_job_id) {
462
0
                    task_runtime_state->set_load_job_id(_params.load_job_id);
463
0
                }
464
1.97M
                if (_params.__isset.wal_id) {
465
114
                    task_runtime_state->set_wal_id(_params.wal_id);
466
114
                }
467
1.97M
                if (_params.__isset.content_length) {
468
31
                    task_runtime_state->set_content_length(_params.content_length);
469
31
                }
470
471
1.97M
                task_runtime_state->set_desc_tbl(_desc_tbl);
472
1.97M
                task_runtime_state->set_per_fragment_instance_idx(local_params.sender_id);
473
1.97M
                task_runtime_state->set_num_per_fragment_instances(_params.num_senders);
474
1.97M
                task_runtime_state->resize_op_id_to_local_state(max_operator_id());
475
1.97M
                task_runtime_state->set_max_operator_id(max_operator_id());
476
1.97M
                task_runtime_state->set_load_stream_per_node(_params.load_stream_per_node);
477
1.97M
                task_runtime_state->set_total_load_streams(_params.total_load_streams);
478
1.97M
                task_runtime_state->set_num_local_sink(_params.num_local_sink);
479
480
1.97M
                task_runtime_state->set_runtime_filter_mgr(runtime_filter_mgr.get());
481
1.97M
            }
482
1.97M
            auto cur_task_id = _total_tasks++;
483
1.97M
            task_runtime_state->set_task_id(cur_task_id);
484
1.97M
            task_runtime_state->set_task_num(pipeline->num_tasks());
485
1.97M
            auto task = std::make_shared<PipelineTask>(
486
1.97M
                    pipeline, cur_task_id, task_runtime_state.get(),
487
1.97M
                    std::dynamic_pointer_cast<PipelineFragmentContext>(shared_from_this()),
488
1.97M
                    pipeline_id_to_profile[pip_idx].get(), get_shared_state(pipeline),
489
1.97M
                    instance_idx);
490
1.97M
            pipeline->incr_created_tasks(instance_idx, task.get());
491
1.97M
            pipeline_id_to_task.insert({pipeline->id(), task.get()});
492
1.97M
            _tasks[instance_idx].emplace_back(
493
1.97M
                    std::pair<std::shared_ptr<PipelineTask>, std::unique_ptr<RuntimeState>> {
494
1.97M
                            std::move(task), std::move(task_runtime_state)});
495
1.97M
        }
496
2.43M
    }
497
498
    /**
499
         * Build DAG for pipeline tasks.
500
         * For example, we have
501
         *
502
         *   ExchangeSink (Pipeline1)     JoinBuildSink (Pipeline2)
503
         *            \                      /
504
         *          JoinProbeOperator1 (Pipeline1)    JoinBuildSink (Pipeline3)
505
         *                 \                          /
506
         *               JoinProbeOperator2 (Pipeline1)
507
         *
508
         * In this fragment, we have three pipelines and pipeline 1 depends on pipeline 2 and pipeline 3.
509
         * To build this DAG, `_dag` manage dependencies between pipelines by pipeline ID and
510
         * `pipeline_id_to_task` is used to find the task by a unique pipeline ID.
511
         *
512
         * Finally, we have two upstream dependencies in Pipeline1 corresponding to JoinProbeOperator1
513
         * and JoinProbeOperator2.
514
         */
515
2.43M
    for (auto& _pipeline : _pipelines) {
516
2.43M
        if (pipeline_id_to_task.contains(_pipeline->id())) {
517
1.97M
            auto* task = pipeline_id_to_task[_pipeline->id()];
518
1.97M
            DCHECK(task != nullptr);
519
520
            // If this task has upstream dependency, then inject it into this task.
521
1.97M
            if (_dag.contains(_pipeline->id())) {
522
1.22M
                auto& deps = _dag[_pipeline->id()];
523
1.94M
                for (auto& dep : deps) {
524
1.94M
                    if (pipeline_id_to_task.contains(dep)) {
525
1.02M
                        auto ss = pipeline_id_to_task[dep]->get_sink_shared_state();
526
1.02M
                        if (ss) {
527
471k
                            task->inject_shared_state(ss);
528
550k
                        } else {
529
550k
                            pipeline_id_to_task[dep]->inject_shared_state(
530
550k
                                    task->get_source_shared_state());
531
550k
                        }
532
1.02M
                    }
533
1.94M
                }
534
1.22M
            }
535
1.97M
        }
536
2.43M
    }
537
3.65M
    for (size_t pip_idx = 0; pip_idx < _pipelines.size(); pip_idx++) {
538
2.43M
        if (pipeline_id_to_task.contains(_pipelines[pip_idx]->id())) {
539
1.97M
            auto* task = pipeline_id_to_task[_pipelines[pip_idx]->id()];
540
1.97M
            DCHECK(pipeline_id_to_profile[pip_idx]);
541
1.97M
            std::vector<TScanRangeParams> scan_ranges;
542
1.97M
            auto node_id = _pipelines[pip_idx]->operators().front()->node_id();
543
1.97M
            if (local_params.per_node_scan_ranges.contains(node_id)) {
544
343k
                scan_ranges = local_params.per_node_scan_ranges.find(node_id)->second;
545
343k
            }
546
1.97M
            RETURN_IF_ERROR_OR_CATCH_EXCEPTION(task->prepare(scan_ranges, local_params.sender_id,
547
1.97M
                                                             _params.fragment.output_sink));
548
1.97M
        }
549
2.43M
    }
550
1.21M
    {
551
1.21M
        std::lock_guard<std::mutex> l(_state_map_lock);
552
1.21M
        _runtime_filter_mgr_map[instance_idx] = std::move(runtime_filter_mgr);
553
1.21M
    }
554
1.21M
    return Status::OK();
555
1.21M
}
556
557
432k
Status PipelineFragmentContext::_build_pipeline_tasks(ThreadPool* thread_pool) {
558
432k
    _total_tasks = 0;
559
432k
    _closed_tasks = 0;
560
432k
    const auto target_size = _params.local_params.size();
561
432k
    _tasks.resize(target_size);
562
432k
    _runtime_filter_mgr_map.resize(target_size);
563
1.10M
    for (size_t pip_idx = 0; pip_idx < _pipelines.size(); pip_idx++) {
564
673k
        _pip_id_to_pipeline[_pipelines[pip_idx]->id()] = _pipelines[pip_idx].get();
565
673k
    }
566
432k
    auto pipeline_id_to_profile = _runtime_state->build_pipeline_profile(_pipelines.size());
567
568
432k
    if (target_size > 1 &&
569
432k
        (_runtime_state->query_options().__isset.parallel_prepare_threshold &&
570
143k
         target_size > _runtime_state->query_options().parallel_prepare_threshold)) {
571
        // If instances parallelism is big enough ( > parallel_prepare_threshold), we will prepare all tasks by multi-threads
572
20.8k
        std::vector<Status> prepare_status(target_size);
573
20.8k
        int submitted_tasks = 0;
574
20.8k
        Status submit_status;
575
20.8k
        CountDownLatch latch((int)target_size);
576
288k
        for (int i = 0; i < target_size; i++) {
577
267k
            submit_status = thread_pool->submit_func([&, i]() {
578
267k
                SCOPED_ATTACH_TASK(_query_ctx.get());
579
267k
                prepare_status[i] = _build_pipeline_tasks_for_instance(i, pipeline_id_to_profile);
580
267k
                latch.count_down();
581
267k
            });
582
267k
            if (LIKELY(submit_status.ok())) {
583
267k
                submitted_tasks++;
584
18.4E
            } else {
585
18.4E
                break;
586
18.4E
            }
587
267k
        }
588
20.8k
        latch.arrive_and_wait(target_size - submitted_tasks);
589
20.8k
        if (UNLIKELY(!submit_status.ok())) {
590
0
            return submit_status;
591
0
        }
592
288k
        for (int i = 0; i < submitted_tasks; i++) {
593
267k
            if (!prepare_status[i].ok()) {
594
0
                return prepare_status[i];
595
0
            }
596
267k
        }
597
411k
    } else {
598
1.36M
        for (int i = 0; i < target_size; i++) {
599
950k
            RETURN_IF_ERROR(_build_pipeline_tasks_for_instance(i, pipeline_id_to_profile));
600
950k
        }
601
411k
    }
602
432k
    _pipeline_parent_map.clear();
603
432k
    _op_id_to_shared_state.clear();
604
605
432k
    return Status::OK();
606
432k
}
607
608
432k
void PipelineFragmentContext::_init_next_report_time() {
609
432k
    auto interval_s = config::pipeline_status_report_interval;
610
432k
    if (_is_report_success && interval_s > 0 && _timeout > interval_s) {
611
41.4k
        VLOG_FILE << "enable period report: fragment id=" << _fragment_id;
612
41.4k
        uint64_t report_fragment_offset = (uint64_t)(rand() % interval_s) * NANOS_PER_SEC;
613
        // We don't want to wait longer than it takes to run the entire fragment.
614
41.4k
        _previous_report_time =
615
41.4k
                MonotonicNanos() + report_fragment_offset - (uint64_t)(interval_s)*NANOS_PER_SEC;
616
41.4k
        _disable_period_report = false;
617
41.4k
    }
618
432k
}
619
620
4.83k
void PipelineFragmentContext::refresh_next_report_time() {
621
4.83k
    auto disable = _disable_period_report.load(std::memory_order_acquire);
622
4.83k
    DCHECK(disable == true);
623
4.83k
    _previous_report_time.store(MonotonicNanos(), std::memory_order_release);
624
4.83k
    _disable_period_report.compare_exchange_strong(disable, false);
625
4.83k
}
626
627
7.16M
void PipelineFragmentContext::trigger_report_if_necessary() {
628
7.16M
    if (!_is_report_success) {
629
6.64M
        return;
630
6.64M
    }
631
521k
    auto disable = _disable_period_report.load(std::memory_order_acquire);
632
521k
    if (disable) {
633
8.57k
        return;
634
8.57k
    }
635
512k
    int32_t interval_s = config::pipeline_status_report_interval;
636
512k
    if (interval_s <= 0) {
637
0
        LOG(WARNING) << "config::status_report_interval is equal to or less than zero, do not "
638
0
                        "trigger "
639
0
                        "report.";
640
0
    }
641
512k
    uint64_t next_report_time = _previous_report_time.load(std::memory_order_acquire) +
642
512k
                                (uint64_t)(interval_s)*NANOS_PER_SEC;
643
512k
    if (MonotonicNanos() > next_report_time) {
644
4.84k
        if (!_disable_period_report.compare_exchange_strong(disable, true,
645
4.84k
                                                            std::memory_order_acq_rel)) {
646
4
            return;
647
4
        }
648
4.83k
        if (VLOG_FILE_IS_ON) {
649
0
            VLOG_FILE << "Reporting "
650
0
                      << "profile for query_id " << print_id(_query_id)
651
0
                      << ", fragment id: " << _fragment_id;
652
653
0
            std::stringstream ss;
654
0
            _runtime_state->runtime_profile()->compute_time_in_profile();
655
0
            _runtime_state->runtime_profile()->pretty_print(&ss);
656
0
            if (_runtime_state->load_channel_profile()) {
657
0
                _runtime_state->load_channel_profile()->pretty_print(&ss);
658
0
            }
659
660
0
            VLOG_FILE << "Query " << print_id(get_query_id()) << " fragment " << get_fragment_id()
661
0
                      << " profile:\n"
662
0
                      << ss.str();
663
0
        }
664
4.83k
        auto st = send_report(false);
665
4.83k
        if (!st.ok()) {
666
0
            disable = true;
667
0
            _disable_period_report.compare_exchange_strong(disable, false,
668
0
                                                           std::memory_order_acq_rel);
669
0
        }
670
4.83k
    }
671
512k
}
672
673
Status PipelineFragmentContext::_build_pipelines(ObjectPool* pool, const DescriptorTbl& descs,
674
433k
                                                 OperatorPtr* root, PipelinePtr cur_pipe) {
675
433k
    if (_params.fragment.plan.nodes.empty()) {
676
0
        throw Exception(ErrorCode::INTERNAL_ERROR, "Invalid plan which has no plan node!");
677
0
    }
678
679
433k
    int node_idx = 0;
680
681
433k
    RETURN_IF_ERROR(_create_tree_helper(pool, _params.fragment.plan.nodes, descs, nullptr,
682
433k
                                        &node_idx, root, cur_pipe, 0, false, false));
683
684
433k
    if (node_idx + 1 != _params.fragment.plan.nodes.size()) {
685
0
        return Status::InternalError(
686
0
                "Plan tree only partially reconstructed. Not all thrift nodes were used.");
687
0
    }
688
433k
    return Status::OK();
689
433k
}
690
691
Status PipelineFragmentContext::_create_tree_helper(
692
        ObjectPool* pool, const std::vector<TPlanNode>& tnodes, const DescriptorTbl& descs,
693
        OperatorPtr parent, int* node_idx, OperatorPtr* root, PipelinePtr& cur_pipe, int child_idx,
694
672k
        const bool followed_by_shuffled_operator, const bool require_bucket_distribution) {
695
    // propagate error case
696
672k
    if (*node_idx >= tnodes.size()) {
697
0
        return Status::InternalError(
698
0
                "Failed to reconstruct plan tree from thrift. Node id: {}, number of nodes: {}",
699
0
                *node_idx, tnodes.size());
700
0
    }
701
672k
    const TPlanNode& tnode = tnodes[*node_idx];
702
703
672k
    int num_children = tnodes[*node_idx].num_children;
704
672k
    bool current_followed_by_shuffled_operator = followed_by_shuffled_operator;
705
672k
    bool current_require_bucket_distribution = require_bucket_distribution;
706
    // TODO: Create CacheOperator is confused now
707
672k
    OperatorPtr op = nullptr;
708
672k
    OperatorPtr cache_op = nullptr;
709
672k
    RETURN_IF_ERROR(_create_operator(pool, tnodes[*node_idx], descs, op, cur_pipe,
710
672k
                                     parent == nullptr ? -1 : parent->node_id(), child_idx,
711
672k
                                     followed_by_shuffled_operator,
712
672k
                                     current_require_bucket_distribution, cache_op));
713
    // Initialization must be done here. For example, group by expressions in agg will be used to
714
    // decide if a local shuffle should be planed, so it must be initialized here.
715
672k
    RETURN_IF_ERROR(op->init(tnode, _runtime_state.get()));
716
    // assert(parent != nullptr || (node_idx == 0 && root_expr != nullptr));
717
672k
    if (parent != nullptr) {
718
        // add to parent's child(s)
719
239k
        RETURN_IF_ERROR(parent->set_child(cache_op ? cache_op : op));
720
432k
    } else {
721
432k
        *root = op;
722
432k
    }
723
    /**
724
     * `ExchangeType::HASH_SHUFFLE` should be used if an operator is followed by a shuffled operator (shuffled hash join, union operator followed by co-located operators).
725
     *
726
     * For plan:
727
     * LocalExchange(id=0) -> Aggregation(id=1) -> ShuffledHashJoin(id=2)
728
     *                           Exchange(id=3) -> ShuffledHashJoinBuild(id=2)
729
     * We must ensure data distribution of `LocalExchange(id=0)` is same as Exchange(id=3).
730
     *
731
     * If an operator's is followed by a local exchange without shuffle (e.g. passthrough), a
732
     * shuffled local exchanger will be used before join so it is not followed by shuffle join.
733
     */
734
672k
    auto required_data_distribution =
735
672k
            cur_pipe->operators().empty()
736
672k
                    ? cur_pipe->sink()->required_data_distribution(_runtime_state.get())
737
672k
                    : op->required_data_distribution(_runtime_state.get());
738
672k
    current_followed_by_shuffled_operator =
739
672k
            ((followed_by_shuffled_operator ||
740
672k
              (cur_pipe->operators().empty() ? cur_pipe->sink()->is_shuffled_operator()
741
611k
                                             : op->is_shuffled_operator())) &&
742
672k
             Pipeline::is_hash_exchange(required_data_distribution.distribution_type)) ||
743
672k
            (followed_by_shuffled_operator &&
744
558k
             required_data_distribution.distribution_type == ExchangeType::NOOP);
745
746
672k
    current_require_bucket_distribution =
747
672k
            ((require_bucket_distribution ||
748
672k
              (cur_pipe->operators().empty() ? cur_pipe->sink()->is_colocated_operator()
749
616k
                                             : op->is_colocated_operator())) &&
750
672k
             Pipeline::is_hash_exchange(required_data_distribution.distribution_type)) ||
751
672k
            (require_bucket_distribution &&
752
564k
             required_data_distribution.distribution_type == ExchangeType::NOOP);
753
754
672k
    if (num_children == 0) {
755
449k
        _use_serial_source = op->is_serial_operator();
756
449k
    }
757
    // rely on that tnodes is preorder of the plan
758
912k
    for (int i = 0; i < num_children; i++) {
759
239k
        ++*node_idx;
760
239k
        RETURN_IF_ERROR(_create_tree_helper(pool, tnodes, descs, op, node_idx, nullptr, cur_pipe, i,
761
239k
                                            current_followed_by_shuffled_operator,
762
239k
                                            current_require_bucket_distribution));
763
764
        // we are expecting a child, but have used all nodes
765
        // this means we have been given a bad tree and must fail
766
239k
        if (*node_idx >= tnodes.size()) {
767
0
            return Status::InternalError(
768
0
                    "Failed to reconstruct plan tree from thrift. Node id: {}, number of "
769
0
                    "nodes: {}",
770
0
                    *node_idx, tnodes.size());
771
0
        }
772
239k
    }
773
774
672k
    return Status::OK();
775
672k
}
776
777
void PipelineFragmentContext::_inherit_pipeline_properties(
778
        const DataDistribution& data_distribution, PipelinePtr pipe_with_source,
779
109k
        PipelinePtr pipe_with_sink) {
780
109k
    pipe_with_sink->set_num_tasks(pipe_with_source->num_tasks());
781
109k
    pipe_with_source->set_num_tasks(_num_instances);
782
109k
    pipe_with_source->set_data_distribution(data_distribution);
783
109k
}
784
785
Status PipelineFragmentContext::_add_local_exchange_impl(
786
        int idx, ObjectPool* pool, PipelinePtr cur_pipe, PipelinePtr new_pip,
787
        DataDistribution data_distribution, bool* do_local_exchange, int num_buckets,
788
        const std::map<int, int>& bucket_seq_to_instance_idx,
789
109k
        const std::map<int, int>& shuffle_idx_to_instance_idx) {
790
109k
    auto& operators = cur_pipe->operators();
791
109k
    const auto downstream_pipeline_id = cur_pipe->id();
792
109k
    auto local_exchange_id = next_operator_id();
793
    // 1. Create a new pipeline with local exchange sink.
794
109k
    DataSinkOperatorPtr sink;
795
109k
    auto sink_id = next_sink_operator_id();
796
797
    /**
798
     * `bucket_seq_to_instance_idx` is empty if no scan operator is contained in this fragment.
799
     * So co-located operators(e.g. Agg, Analytic) should use `HASH_SHUFFLE` instead of `BUCKET_HASH_SHUFFLE`.
800
     */
801
109k
    const bool followed_by_shuffled_operator =
802
109k
            operators.size() > idx ? operators[idx]->followed_by_shuffled_operator()
803
109k
                                   : cur_pipe->sink()->followed_by_shuffled_operator();
804
109k
    const bool use_global_hash_shuffle = bucket_seq_to_instance_idx.empty() &&
805
109k
                                         !shuffle_idx_to_instance_idx.contains(-1) &&
806
109k
                                         followed_by_shuffled_operator && !_use_serial_source;
807
109k
    sink = std::make_shared<LocalExchangeSinkOperatorX>(
808
109k
            sink_id, local_exchange_id, use_global_hash_shuffle ? _total_instances : _num_instances,
809
109k
            data_distribution.partition_exprs, bucket_seq_to_instance_idx);
810
109k
    if (bucket_seq_to_instance_idx.empty() &&
811
109k
        data_distribution.distribution_type == ExchangeType::BUCKET_HASH_SHUFFLE) {
812
7
        data_distribution.distribution_type = ExchangeType::HASH_SHUFFLE;
813
7
    }
814
109k
    RETURN_IF_ERROR(new_pip->set_sink(sink));
815
109k
    RETURN_IF_ERROR(new_pip->sink()->init(_runtime_state.get(), data_distribution.distribution_type,
816
109k
                                          num_buckets, use_global_hash_shuffle,
817
109k
                                          shuffle_idx_to_instance_idx));
818
819
    // 2. Create and initialize LocalExchangeSharedState.
820
109k
    std::shared_ptr<LocalExchangeSharedState> shared_state =
821
109k
            LocalExchangeSharedState::create_shared(_num_instances);
822
109k
    switch (data_distribution.distribution_type) {
823
10.9k
    case ExchangeType::HASH_SHUFFLE:
824
10.9k
        shared_state->exchanger = ShuffleExchanger::create_unique(
825
10.9k
                std::max(cur_pipe->num_tasks(), _num_instances), _num_instances,
826
10.9k
                use_global_hash_shuffle ? _total_instances : _num_instances,
827
10.9k
                _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
828
10.9k
                        ? cast_set<int>(
829
10.9k
                                  _runtime_state->query_options().local_exchange_free_blocks_limit)
830
10.9k
                        : 0);
831
10.9k
        break;
832
654
    case ExchangeType::BUCKET_HASH_SHUFFLE:
833
654
        shared_state->exchanger = BucketShuffleExchanger::create_unique(
834
654
                std::max(cur_pipe->num_tasks(), _num_instances), _num_instances, num_buckets,
835
654
                _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
836
654
                        ? cast_set<int>(
837
654
                                  _runtime_state->query_options().local_exchange_free_blocks_limit)
838
654
                        : 0);
839
654
        break;
840
94.0k
    case ExchangeType::PASSTHROUGH:
841
94.0k
        shared_state->exchanger = PassthroughExchanger::create_unique(
842
94.0k
                cur_pipe->num_tasks(), _num_instances,
843
94.0k
                _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
844
94.0k
                        ? cast_set<int>(
845
93.9k
                                  _runtime_state->query_options().local_exchange_free_blocks_limit)
846
94.0k
                        : 0);
847
94.0k
        break;
848
474
    case ExchangeType::BROADCAST:
849
474
        shared_state->exchanger = BroadcastExchanger::create_unique(
850
474
                cur_pipe->num_tasks(), _num_instances,
851
474
                _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
852
474
                        ? cast_set<int>(
853
474
                                  _runtime_state->query_options().local_exchange_free_blocks_limit)
854
474
                        : 0);
855
474
        break;
856
2.94k
    case ExchangeType::PASS_TO_ONE:
857
2.94k
        if (_runtime_state->enable_share_hash_table_for_broadcast_join()) {
858
            // If shared hash table is enabled for BJ, hash table will be built by only one task
859
1.37k
            shared_state->exchanger = PassToOneExchanger::create_unique(
860
1.37k
                    cur_pipe->num_tasks(), _num_instances,
861
1.37k
                    _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
862
1.37k
                            ? cast_set<int>(_runtime_state->query_options()
863
1.37k
                                                    .local_exchange_free_blocks_limit)
864
1.37k
                            : 0);
865
1.56k
        } else {
866
1.56k
            shared_state->exchanger = BroadcastExchanger::create_unique(
867
1.56k
                    cur_pipe->num_tasks(), _num_instances,
868
1.56k
                    _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
869
1.56k
                            ? cast_set<int>(_runtime_state->query_options()
870
1.56k
                                                    .local_exchange_free_blocks_limit)
871
1.56k
                            : 0);
872
1.56k
        }
873
2.94k
        break;
874
929
    case ExchangeType::ADAPTIVE_PASSTHROUGH:
875
929
        shared_state->exchanger = AdaptivePassthroughExchanger::create_unique(
876
929
                std::max(cur_pipe->num_tasks(), _num_instances), _num_instances,
877
929
                _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
878
929
                        ? cast_set<int>(
879
929
                                  _runtime_state->query_options().local_exchange_free_blocks_limit)
880
929
                        : 0);
881
929
        break;
882
0
    default:
883
0
        return Status::InternalError("Unsupported local exchange type : " +
884
0
                                     std::to_string((int)data_distribution.distribution_type));
885
109k
    }
886
110k
    shared_state->create_source_dependencies(_num_instances, local_exchange_id, local_exchange_id,
887
110k
                                             "LOCAL_EXCHANGE_OPERATOR");
888
110k
    shared_state->create_sink_dependency(sink_id, local_exchange_id, "LOCAL_EXCHANGE_SINK");
889
110k
    _op_id_to_shared_state.insert({local_exchange_id, {shared_state, shared_state->sink_deps}});
890
891
    // 3. Set two pipelines' operator list. For example, split pipeline [Scan - AggSink] to
892
    // pipeline1 [Scan - LocalExchangeSink] and pipeline2 [LocalExchangeSource - AggSink].
893
894
    // 3.1 Initialize new pipeline's operator list.
895
110k
    std::copy(operators.begin(), operators.begin() + idx,
896
110k
              std::inserter(new_pip->operators(), new_pip->operators().end()));
897
898
    // 3.2 Erase unused operators in previous pipeline.
899
110k
    operators.erase(operators.begin(), operators.begin() + idx);
900
901
    // 4. Initialize LocalExchangeSource and insert it into this pipeline.
902
110k
    OperatorPtr source_op;
903
110k
    source_op = std::make_shared<LocalExchangeSourceOperatorX>(pool, local_exchange_id);
904
110k
    RETURN_IF_ERROR(source_op->set_child(new_pip->operators().back()));
905
110k
    RETURN_IF_ERROR(source_op->init(data_distribution.distribution_type));
906
110k
    if (!operators.empty()) {
907
48.5k
        RETURN_IF_ERROR(operators.front()->set_child(nullptr));
908
48.5k
        RETURN_IF_ERROR(operators.front()->set_child(source_op));
909
48.5k
    }
910
110k
    operators.insert(operators.begin(), source_op);
911
912
    // 5. Set children for two pipelines separately.
913
110k
    std::vector<std::shared_ptr<Pipeline>> new_children;
914
110k
    std::vector<PipelineId> edges_with_source;
915
128k
    for (auto child : cur_pipe->children()) {
916
128k
        bool found = false;
917
143k
        for (auto op : new_pip->operators()) {
918
143k
            if (child->sink()->node_id() == op->node_id()) {
919
12.7k
                new_pip->set_children(child);
920
12.7k
                found = true;
921
12.7k
            };
922
143k
        }
923
128k
        if (!found) {
924
115k
            new_children.push_back(child);
925
115k
            edges_with_source.push_back(child->id());
926
115k
        }
927
128k
    }
928
110k
    new_children.push_back(new_pip);
929
110k
    edges_with_source.push_back(new_pip->id());
930
931
    // 6. Set DAG for new pipelines.
932
110k
    if (!new_pip->children().empty()) {
933
7.10k
        std::vector<PipelineId> edges_with_sink;
934
12.7k
        for (auto child : new_pip->children()) {
935
12.7k
            edges_with_sink.push_back(child->id());
936
12.7k
        }
937
7.10k
        _dag.insert({new_pip->id(), edges_with_sink});
938
7.10k
    }
939
110k
    cur_pipe->set_children(new_children);
940
110k
    _dag[downstream_pipeline_id] = edges_with_source;
941
110k
    RETURN_IF_ERROR(new_pip->sink()->set_child(new_pip->operators().back()));
942
110k
    RETURN_IF_ERROR(cur_pipe->sink()->set_child(nullptr));
943
110k
    RETURN_IF_ERROR(cur_pipe->sink()->set_child(cur_pipe->operators().back()));
944
945
    // 7. Inherit properties from current pipeline.
946
110k
    _inherit_pipeline_properties(data_distribution, cur_pipe, new_pip);
947
110k
    return Status::OK();
948
110k
}
949
950
Status PipelineFragmentContext::_add_local_exchange(
951
        int pip_idx, int idx, int node_id, ObjectPool* pool, PipelinePtr cur_pipe,
952
        DataDistribution data_distribution, bool* do_local_exchange, int num_buckets,
953
        const std::map<int, int>& bucket_seq_to_instance_idx,
954
191k
        const std::map<int, int>& shuffle_idx_to_instance_idx) {
955
191k
    if (_num_instances <= 1 || cur_pipe->num_tasks_of_parent() <= 1) {
956
40.2k
        return Status::OK();
957
40.2k
    }
958
959
151k
    if (!cur_pipe->need_to_local_exchange(data_distribution, idx)) {
960
52.5k
        return Status::OK();
961
52.5k
    }
962
98.6k
    *do_local_exchange = true;
963
964
98.6k
    auto& operators = cur_pipe->operators();
965
98.6k
    auto total_op_num = operators.size();
966
98.6k
    auto new_pip = add_pipeline(cur_pipe, pip_idx + 1);
967
98.6k
    RETURN_IF_ERROR(_add_local_exchange_impl(
968
98.6k
            idx, pool, cur_pipe, new_pip, data_distribution, do_local_exchange, num_buckets,
969
98.6k
            bucket_seq_to_instance_idx, shuffle_idx_to_instance_idx));
970
971
98.6k
    CHECK(total_op_num + 1 == cur_pipe->operators().size() + new_pip->operators().size())
972
124
            << "total_op_num: " << total_op_num
973
124
            << " cur_pipe->operators().size(): " << cur_pipe->operators().size()
974
124
            << " new_pip->operators().size(): " << new_pip->operators().size();
975
976
    // There are some local shuffles with relatively heavy operations on the sink.
977
    // If the local sink concurrency is 1 and the local source concurrency is n, the sink becomes a bottleneck.
978
    // Therefore, local passthrough is used to increase the concurrency of the sink.
979
    // op -> local sink(1) -> local source (n)
980
    // op -> local passthrough(1) -> local passthrough(n) ->  local sink(n) -> local source (n)
981
98.6k
    if (cur_pipe->num_tasks() > 1 && new_pip->num_tasks() == 1 &&
982
98.6k
        Pipeline::heavy_operations_on_the_sink(data_distribution.distribution_type)) {
983
11.4k
        RETURN_IF_ERROR(_add_local_exchange_impl(
984
11.4k
                cast_set<int>(new_pip->operators().size()), pool, new_pip,
985
11.4k
                add_pipeline(new_pip, pip_idx + 2), DataDistribution(ExchangeType::PASSTHROUGH),
986
11.4k
                do_local_exchange, num_buckets, bucket_seq_to_instance_idx,
987
11.4k
                shuffle_idx_to_instance_idx));
988
11.4k
    }
989
98.6k
    return Status::OK();
990
98.6k
}
991
992
Status PipelineFragmentContext::_plan_local_exchange(
993
        int num_buckets, const std::map<int, int>& bucket_seq_to_instance_idx,
994
430k
        const std::map<int, int>& shuffle_idx_to_instance_idx) {
995
992k
    for (int pip_idx = cast_set<int>(_pipelines.size()) - 1; pip_idx >= 0; pip_idx--) {
996
561k
        _pipelines[pip_idx]->init_data_distribution(_runtime_state.get());
997
        // Set property if child pipeline is not join operator's child.
998
561k
        if (!_pipelines[pip_idx]->children().empty()) {
999
124k
            for (auto& child : _pipelines[pip_idx]->children()) {
1000
124k
                if (child->sink()->node_id() ==
1001
124k
                    _pipelines[pip_idx]->operators().front()->node_id()) {
1002
110k
                    _pipelines[pip_idx]->set_data_distribution(child->data_distribution());
1003
110k
                }
1004
124k
            }
1005
119k
        }
1006
1007
        // if 'num_buckets == 0' means the fragment is colocated by exchange node not the
1008
        // scan node. so here use `_num_instance` to replace the `num_buckets` to prevent dividing 0
1009
        // still keep colocate plan after local shuffle
1010
561k
        RETURN_IF_ERROR(_plan_local_exchange(num_buckets, pip_idx, _pipelines[pip_idx],
1011
561k
                                             bucket_seq_to_instance_idx,
1012
561k
                                             shuffle_idx_to_instance_idx));
1013
561k
    }
1014
430k
    return Status::OK();
1015
430k
}
1016
1017
Status PipelineFragmentContext::_plan_local_exchange(
1018
        int num_buckets, int pip_idx, PipelinePtr pip,
1019
        const std::map<int, int>& bucket_seq_to_instance_idx,
1020
561k
        const std::map<int, int>& shuffle_idx_to_instance_idx) {
1021
561k
    int idx = 1;
1022
561k
    bool do_local_exchange = false;
1023
609k
    do {
1024
609k
        auto& ops = pip->operators();
1025
609k
        do_local_exchange = false;
1026
        // Plan local exchange for each operator.
1027
676k
        for (; idx < ops.size();) {
1028
115k
            if (ops[idx]->required_data_distribution(_runtime_state.get()).need_local_exchange()) {
1029
105k
                RETURN_IF_ERROR(_add_local_exchange(
1030
105k
                        pip_idx, idx, ops[idx]->node_id(), _runtime_state->obj_pool(), pip,
1031
105k
                        ops[idx]->required_data_distribution(_runtime_state.get()),
1032
105k
                        &do_local_exchange, num_buckets, bucket_seq_to_instance_idx,
1033
105k
                        shuffle_idx_to_instance_idx));
1034
105k
            }
1035
115k
            if (do_local_exchange) {
1036
                // If local exchange is needed for current operator, we will split this pipeline to
1037
                // two pipelines by local exchange sink/source. And then we need to process remaining
1038
                // operators in this pipeline so we set idx to 2 (0 is local exchange source and 1
1039
                // is current operator was already processed) and continue to plan local exchange.
1040
48.6k
                idx = 2;
1041
48.6k
                break;
1042
48.6k
            }
1043
66.3k
            idx++;
1044
66.3k
        }
1045
609k
    } while (do_local_exchange);
1046
561k
    if (pip->sink()->required_data_distribution(_runtime_state.get()).need_local_exchange()) {
1047
86.2k
        RETURN_IF_ERROR(_add_local_exchange(
1048
86.2k
                pip_idx, idx, pip->sink()->node_id(), _runtime_state->obj_pool(), pip,
1049
86.2k
                pip->sink()->required_data_distribution(_runtime_state.get()), &do_local_exchange,
1050
86.2k
                num_buckets, bucket_seq_to_instance_idx, shuffle_idx_to_instance_idx));
1051
86.2k
    }
1052
561k
    return Status::OK();
1053
561k
}
1054
1055
Status PipelineFragmentContext::_create_data_sink(ObjectPool* pool, const TDataSink& thrift_sink,
1056
                                                  const std::vector<TExpr>& output_exprs,
1057
                                                  const TPipelineFragmentParams& params,
1058
                                                  const RowDescriptor& row_desc,
1059
                                                  RuntimeState* state, DescriptorTbl& desc_tbl,
1060
433k
                                                  PipelineId cur_pipeline_id) {
1061
433k
    switch (thrift_sink.type) {
1062
144k
    case TDataSinkType::DATA_STREAM_SINK: {
1063
144k
        if (!thrift_sink.__isset.stream_sink) {
1064
0
            return Status::InternalError("Missing data stream sink.");
1065
0
        }
1066
144k
        _sink = std::make_shared<ExchangeSinkOperatorX>(
1067
144k
                state, row_desc, next_sink_operator_id(), thrift_sink.stream_sink,
1068
144k
                params.destinations, _fragment_instance_ids);
1069
144k
        break;
1070
144k
    }
1071
251k
    case TDataSinkType::RESULT_SINK: {
1072
251k
        if (!thrift_sink.__isset.result_sink) {
1073
0
            return Status::InternalError("Missing data buffer sink.");
1074
0
        }
1075
1076
251k
        auto& pipeline = _pipelines[cur_pipeline_id];
1077
251k
        int child_node_id = pipeline->operators().back()->node_id();
1078
251k
        _sink = std::make_shared<ResultSinkOperatorX>(next_sink_operator_id(), child_node_id + 1,
1079
251k
                                                      row_desc, output_exprs,
1080
251k
                                                      thrift_sink.result_sink);
1081
251k
        break;
1082
251k
    }
1083
105
    case TDataSinkType::DICTIONARY_SINK: {
1084
105
        if (!thrift_sink.__isset.dictionary_sink) {
1085
0
            return Status::InternalError("Missing dict sink.");
1086
0
        }
1087
1088
105
        _sink = std::make_shared<DictSinkOperatorX>(next_sink_operator_id(), row_desc, output_exprs,
1089
105
                                                    thrift_sink.dictionary_sink);
1090
105
        break;
1091
105
    }
1092
0
    case TDataSinkType::GROUP_COMMIT_OLAP_TABLE_SINK:
1093
31.3k
    case TDataSinkType::OLAP_TABLE_SINK: {
1094
31.3k
        auto& pipeline = _pipelines[cur_pipeline_id];
1095
31.3k
        int child_node_id = pipeline->operators().back()->node_id();
1096
31.3k
        if (state->query_options().enable_memtable_on_sink_node &&
1097
31.3k
            !_has_inverted_index_v1_or_partial_update(thrift_sink.olap_table_sink) &&
1098
31.3k
            !config::is_cloud_mode()) {
1099
2.15k
            _sink = std::make_shared<OlapTableSinkV2OperatorX>(
1100
2.15k
                    pool, next_sink_operator_id(), child_node_id + 1, row_desc, output_exprs);
1101
29.1k
        } else {
1102
29.1k
            _sink = std::make_shared<OlapTableSinkOperatorX>(
1103
29.1k
                    pool, next_sink_operator_id(), child_node_id + 1, row_desc, output_exprs);
1104
29.1k
        }
1105
31.3k
        break;
1106
0
    }
1107
165
    case TDataSinkType::GROUP_COMMIT_BLOCK_SINK: {
1108
165
        DCHECK(thrift_sink.__isset.olap_table_sink);
1109
165
        DCHECK(state->get_query_ctx() != nullptr);
1110
165
        state->get_query_ctx()->query_mem_tracker()->is_group_commit_load = true;
1111
165
        _sink = std::make_shared<GroupCommitBlockSinkOperatorX>(next_sink_operator_id(), row_desc,
1112
165
                                                                output_exprs);
1113
165
        break;
1114
0
    }
1115
1.46k
    case TDataSinkType::HIVE_TABLE_SINK: {
1116
1.46k
        if (!thrift_sink.__isset.hive_table_sink) {
1117
0
            return Status::InternalError("Missing hive table sink.");
1118
0
        }
1119
1.46k
        _sink = std::make_shared<HiveTableSinkOperatorX>(pool, next_sink_operator_id(), row_desc,
1120
1.46k
                                                         output_exprs);
1121
1.46k
        break;
1122
1.46k
    }
1123
1.73k
    case TDataSinkType::ICEBERG_TABLE_SINK: {
1124
1.73k
        if (!thrift_sink.__isset.iceberg_table_sink) {
1125
0
            return Status::InternalError("Missing iceberg table sink.");
1126
0
        }
1127
1.73k
        if (thrift_sink.iceberg_table_sink.__isset.sort_info) {
1128
0
            _sink = std::make_shared<SpillIcebergTableSinkOperatorX>(pool, next_sink_operator_id(),
1129
0
                                                                     row_desc, output_exprs);
1130
1.73k
        } else {
1131
1.73k
            _sink = std::make_shared<IcebergTableSinkOperatorX>(pool, next_sink_operator_id(),
1132
1.73k
                                                                row_desc, output_exprs);
1133
1.73k
        }
1134
1.73k
        break;
1135
1.73k
    }
1136
20
    case TDataSinkType::ICEBERG_DELETE_SINK: {
1137
20
        if (!thrift_sink.__isset.iceberg_delete_sink) {
1138
0
            return Status::InternalError("Missing iceberg delete sink.");
1139
0
        }
1140
20
        _sink = std::make_shared<IcebergDeleteSinkOperatorX>(pool, next_sink_operator_id(),
1141
20
                                                             row_desc, output_exprs);
1142
20
        break;
1143
20
    }
1144
80
    case TDataSinkType::ICEBERG_MERGE_SINK: {
1145
80
        if (!thrift_sink.__isset.iceberg_merge_sink) {
1146
0
            return Status::InternalError("Missing iceberg merge sink.");
1147
0
        }
1148
80
        _sink = std::make_shared<IcebergMergeSinkOperatorX>(pool, next_sink_operator_id(), row_desc,
1149
80
                                                            output_exprs);
1150
80
        break;
1151
80
    }
1152
0
    case TDataSinkType::MAXCOMPUTE_TABLE_SINK: {
1153
0
        if (!thrift_sink.__isset.max_compute_table_sink) {
1154
0
            return Status::InternalError("Missing max compute table sink.");
1155
0
        }
1156
0
        _sink = std::make_shared<MCTableSinkOperatorX>(pool, next_sink_operator_id(), row_desc,
1157
0
                                                       output_exprs);
1158
0
        break;
1159
0
    }
1160
88
    case TDataSinkType::JDBC_TABLE_SINK: {
1161
88
        if (!thrift_sink.__isset.jdbc_table_sink) {
1162
0
            return Status::InternalError("Missing data jdbc sink.");
1163
0
        }
1164
88
        if (config::enable_java_support) {
1165
88
            _sink = std::make_shared<JdbcTableSinkOperatorX>(row_desc, next_sink_operator_id(),
1166
88
                                                             output_exprs);
1167
88
        } else {
1168
0
            return Status::InternalError(
1169
0
                    "Jdbc table sink is not enabled, you can change be config "
1170
0
                    "enable_java_support to true and restart be.");
1171
0
        }
1172
88
        break;
1173
88
    }
1174
88
    case TDataSinkType::MEMORY_SCRATCH_SINK: {
1175
3
        if (!thrift_sink.__isset.memory_scratch_sink) {
1176
0
            return Status::InternalError("Missing data buffer sink.");
1177
0
        }
1178
1179
3
        _sink = std::make_shared<MemoryScratchSinkOperatorX>(row_desc, next_sink_operator_id(),
1180
3
                                                             output_exprs);
1181
3
        break;
1182
3
    }
1183
502
    case TDataSinkType::RESULT_FILE_SINK: {
1184
502
        if (!thrift_sink.__isset.result_file_sink) {
1185
0
            return Status::InternalError("Missing result file sink.");
1186
0
        }
1187
1188
        // Result file sink is not the top sink
1189
502
        if (params.__isset.destinations && !params.destinations.empty()) {
1190
0
            _sink = std::make_shared<ResultFileSinkOperatorX>(
1191
0
                    next_sink_operator_id(), row_desc, thrift_sink.result_file_sink,
1192
0
                    params.destinations, output_exprs, desc_tbl);
1193
502
        } else {
1194
502
            _sink = std::make_shared<ResultFileSinkOperatorX>(next_sink_operator_id(), row_desc,
1195
502
                                                              output_exprs);
1196
502
        }
1197
502
        break;
1198
502
    }
1199
1.99k
    case TDataSinkType::MULTI_CAST_DATA_STREAM_SINK: {
1200
1.99k
        DCHECK(thrift_sink.__isset.multi_cast_stream_sink);
1201
1.99k
        DCHECK_GT(thrift_sink.multi_cast_stream_sink.sinks.size(), 0);
1202
1.99k
        auto sink_id = next_sink_operator_id();
1203
1.99k
        const int multi_cast_node_id = sink_id;
1204
1.99k
        auto sender_size = thrift_sink.multi_cast_stream_sink.sinks.size();
1205
        // one sink has multiple sources.
1206
1.99k
        std::vector<int> sources;
1207
7.77k
        for (int i = 0; i < sender_size; ++i) {
1208
5.77k
            auto source_id = next_operator_id();
1209
5.77k
            sources.push_back(source_id);
1210
5.77k
        }
1211
1212
1.99k
        _sink = std::make_shared<MultiCastDataStreamSinkOperatorX>(
1213
1.99k
                sink_id, multi_cast_node_id, sources, pool, thrift_sink.multi_cast_stream_sink);
1214
7.77k
        for (int i = 0; i < sender_size; ++i) {
1215
5.77k
            auto new_pipeline = add_pipeline();
1216
            // use to exchange sink
1217
5.77k
            RowDescriptor* exchange_row_desc = nullptr;
1218
5.77k
            {
1219
5.77k
                const auto& tmp_row_desc =
1220
5.77k
                        !thrift_sink.multi_cast_stream_sink.sinks[i].output_exprs.empty()
1221
5.77k
                                ? RowDescriptor(state->desc_tbl(),
1222
5.77k
                                                {thrift_sink.multi_cast_stream_sink.sinks[i]
1223
5.77k
                                                         .output_tuple_id})
1224
5.77k
                                : row_desc;
1225
5.77k
                exchange_row_desc = pool->add(new RowDescriptor(tmp_row_desc));
1226
5.77k
            }
1227
5.77k
            auto source_id = sources[i];
1228
5.77k
            OperatorPtr source_op;
1229
            // 1. create and set the source operator of multi_cast_data_stream_source for new pipeline
1230
5.77k
            source_op = std::make_shared<MultiCastDataStreamerSourceOperatorX>(
1231
5.77k
                    /*node_id*/ source_id, /*consumer_id*/ i, pool,
1232
5.77k
                    thrift_sink.multi_cast_stream_sink.sinks[i], row_desc,
1233
5.77k
                    /*operator_id=*/source_id);
1234
5.77k
            RETURN_IF_ERROR(new_pipeline->add_operator(
1235
5.77k
                    source_op, params.__isset.parallel_instances ? params.parallel_instances : 0));
1236
            // 2. create and set sink operator of data stream sender for new pipeline
1237
1238
5.77k
            DataSinkOperatorPtr sink_op;
1239
5.77k
            sink_op = std::make_shared<ExchangeSinkOperatorX>(
1240
5.77k
                    state, *exchange_row_desc, next_sink_operator_id(),
1241
5.77k
                    thrift_sink.multi_cast_stream_sink.sinks[i],
1242
5.77k
                    thrift_sink.multi_cast_stream_sink.destinations[i], _fragment_instance_ids);
1243
1244
5.77k
            RETURN_IF_ERROR(new_pipeline->set_sink(sink_op));
1245
5.77k
            {
1246
5.77k
                TDataSink* t = pool->add(new TDataSink());
1247
5.77k
                t->stream_sink = thrift_sink.multi_cast_stream_sink.sinks[i];
1248
5.77k
                RETURN_IF_ERROR(sink_op->init(*t));
1249
5.77k
            }
1250
1251
            // 3. set dependency dag
1252
5.77k
            _dag[new_pipeline->id()].push_back(cur_pipeline_id);
1253
5.77k
        }
1254
1.99k
        if (sources.empty()) {
1255
0
            return Status::InternalError("size of sources must be greater than 0");
1256
0
        }
1257
1.99k
        break;
1258
1.99k
    }
1259
1.99k
    case TDataSinkType::BLACKHOLE_SINK: {
1260
13
        if (!thrift_sink.__isset.blackhole_sink) {
1261
0
            return Status::InternalError("Missing blackhole sink.");
1262
0
        }
1263
1264
13
        _sink.reset(new BlackholeSinkOperatorX(next_sink_operator_id()));
1265
13
        break;
1266
13
    }
1267
156
    case TDataSinkType::TVF_TABLE_SINK: {
1268
156
        if (!thrift_sink.__isset.tvf_table_sink) {
1269
0
            return Status::InternalError("Missing TVF table sink.");
1270
0
        }
1271
156
        _sink = std::make_shared<TVFTableSinkOperatorX>(pool, next_sink_operator_id(), row_desc,
1272
156
                                                        output_exprs);
1273
156
        break;
1274
156
    }
1275
0
    default:
1276
0
        return Status::InternalError("Unsuported sink type in pipeline: {}", thrift_sink.type);
1277
433k
    }
1278
433k
    return Status::OK();
1279
433k
}
1280
1281
// NOLINTBEGIN(readability-function-size)
1282
// NOLINTBEGIN(readability-function-cognitive-complexity)
1283
Status PipelineFragmentContext::_create_operator(ObjectPool* pool, const TPlanNode& tnode,
1284
                                                 const DescriptorTbl& descs, OperatorPtr& op,
1285
                                                 PipelinePtr& cur_pipe, int parent_idx,
1286
                                                 int child_idx,
1287
                                                 const bool followed_by_shuffled_operator,
1288
                                                 const bool require_bucket_distribution,
1289
673k
                                                 OperatorPtr& cache_op) {
1290
673k
    std::vector<DataSinkOperatorPtr> sink_ops;
1291
673k
    Defer defer = Defer([&]() {
1292
673k
        if (op) {
1293
673k
            op->update_operator(tnode, followed_by_shuffled_operator, require_bucket_distribution);
1294
673k
        }
1295
672k
        for (auto& s : sink_ops) {
1296
124k
            s->update_operator(tnode, followed_by_shuffled_operator, require_bucket_distribution);
1297
124k
        }
1298
672k
    });
1299
    // We directly construct the operator from Thrift because the given array is in the order of preorder traversal.
1300
    // Therefore, here we need to use a stack-like structure.
1301
673k
    _pipeline_parent_map.pop(cur_pipe, parent_idx, child_idx);
1302
673k
    std::stringstream error_msg;
1303
673k
    bool enable_query_cache = _params.fragment.__isset.query_cache_param;
1304
1305
673k
    bool fe_with_old_version = false;
1306
673k
    switch (tnode.node_type) {
1307
212k
    case TPlanNodeType::OLAP_SCAN_NODE: {
1308
212k
        op = std::make_shared<OlapScanOperatorX>(
1309
212k
                pool, tnode, next_operator_id(), descs, _num_instances,
1310
212k
                enable_query_cache ? _params.fragment.query_cache_param : TQueryCacheParam {});
1311
212k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1312
212k
        fe_with_old_version = !tnode.__isset.is_serial_operator;
1313
212k
        break;
1314
212k
    }
1315
79
    case TPlanNodeType::GROUP_COMMIT_SCAN_NODE: {
1316
79
        DCHECK(_query_ctx != nullptr);
1317
79
        _query_ctx->query_mem_tracker()->is_group_commit_load = true;
1318
79
        op = std::make_shared<GroupCommitOperatorX>(pool, tnode, next_operator_id(), descs,
1319
79
                                                    _num_instances);
1320
79
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1321
79
        fe_with_old_version = !tnode.__isset.is_serial_operator;
1322
79
        break;
1323
79
    }
1324
0
    case TPlanNodeType::JDBC_SCAN_NODE: {
1325
0
        if (config::enable_java_support) {
1326
0
            op = std::make_shared<JDBCScanOperatorX>(pool, tnode, next_operator_id(), descs,
1327
0
                                                     _num_instances);
1328
0
            RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1329
0
        } else {
1330
0
            return Status::InternalError(
1331
0
                    "Jdbc scan node is disabled, you can change be config enable_java_support "
1332
0
                    "to true and restart be.");
1333
0
        }
1334
0
        fe_with_old_version = !tnode.__isset.is_serial_operator;
1335
0
        break;
1336
0
    }
1337
23.2k
    case TPlanNodeType::FILE_SCAN_NODE: {
1338
23.2k
        op = std::make_shared<FileScanOperatorX>(pool, tnode, next_operator_id(), descs,
1339
23.2k
                                                 _num_instances);
1340
23.2k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1341
23.2k
        fe_with_old_version = !tnode.__isset.is_serial_operator;
1342
23.2k
        break;
1343
23.2k
    }
1344
148k
    case TPlanNodeType::EXCHANGE_NODE: {
1345
148k
        int num_senders = _params.per_exch_num_senders.contains(tnode.node_id)
1346
148k
                                  ? _params.per_exch_num_senders.find(tnode.node_id)->second
1347
18.4E
                                  : 0;
1348
148k
        DCHECK_GT(num_senders, 0);
1349
148k
        op = std::make_shared<ExchangeSourceOperatorX>(pool, tnode, next_operator_id(), descs,
1350
148k
                                                       num_senders);
1351
148k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1352
148k
        fe_with_old_version = !tnode.__isset.is_serial_operator;
1353
148k
        break;
1354
148k
    }
1355
158k
    case TPlanNodeType::AGGREGATION_NODE: {
1356
158k
        if (tnode.agg_node.grouping_exprs.empty() &&
1357
158k
            descs.get_tuple_descriptor(tnode.agg_node.output_tuple_id)->slots().empty()) {
1358
0
            return Status::InternalError("Illegal aggregate node " + std::to_string(tnode.node_id) +
1359
0
                                         ": group by and output is empty");
1360
0
        }
1361
158k
        bool need_create_cache_op =
1362
158k
                enable_query_cache && tnode.node_id == _params.fragment.query_cache_param.node_id;
1363
158k
        auto create_query_cache_operator = [&](PipelinePtr& new_pipe) {
1364
10
            auto cache_node_id = _params.local_params[0].per_node_scan_ranges.begin()->first;
1365
10
            auto cache_source_id = next_operator_id();
1366
10
            op = std::make_shared<CacheSourceOperatorX>(pool, cache_node_id, cache_source_id,
1367
10
                                                        _params.fragment.query_cache_param);
1368
10
            RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1369
1370
10
            const auto downstream_pipeline_id = cur_pipe->id();
1371
10
            if (!_dag.contains(downstream_pipeline_id)) {
1372
10
                _dag.insert({downstream_pipeline_id, {}});
1373
10
            }
1374
10
            new_pipe = add_pipeline(cur_pipe);
1375
10
            _dag[downstream_pipeline_id].push_back(new_pipe->id());
1376
1377
10
            DataSinkOperatorPtr cache_sink(new CacheSinkOperatorX(
1378
10
                    next_sink_operator_id(), op->node_id(), op->operator_id()));
1379
10
            RETURN_IF_ERROR(new_pipe->set_sink(cache_sink));
1380
10
            return Status::OK();
1381
10
        };
1382
158k
        const bool group_by_limit_opt =
1383
158k
                tnode.agg_node.__isset.agg_sort_info_by_group_key && tnode.limit > 0;
1384
1385
        /// PartitionedAggSourceOperatorX does not support "group by limit opt(#29641)" yet.
1386
        /// If `group_by_limit_opt` is true, then it might not need to spill at all.
1387
158k
        const bool enable_spill = _runtime_state->enable_spill() &&
1388
158k
                                  !tnode.agg_node.grouping_exprs.empty() && !group_by_limit_opt;
1389
158k
        const bool is_streaming_agg = tnode.agg_node.__isset.use_streaming_preaggregation &&
1390
158k
                                      tnode.agg_node.use_streaming_preaggregation &&
1391
158k
                                      !tnode.agg_node.grouping_exprs.empty();
1392
        // TODO: distinct streaming agg does not support spill.
1393
158k
        const bool can_use_distinct_streaming_agg =
1394
158k
                (!enable_spill || is_streaming_agg) && tnode.agg_node.aggregate_functions.empty() &&
1395
158k
                !tnode.agg_node.__isset.agg_sort_info_by_group_key &&
1396
158k
                _params.query_options.__isset.enable_distinct_streaming_aggregation &&
1397
158k
                _params.query_options.enable_distinct_streaming_aggregation;
1398
1399
158k
        if (can_use_distinct_streaming_agg) {
1400
92.4k
            if (need_create_cache_op) {
1401
8
                PipelinePtr new_pipe;
1402
8
                RETURN_IF_ERROR(create_query_cache_operator(new_pipe));
1403
1404
8
                cache_op = op;
1405
8
                op = std::make_shared<DistinctStreamingAggOperatorX>(pool, next_operator_id(),
1406
8
                                                                     tnode, descs);
1407
8
                RETURN_IF_ERROR(new_pipe->add_operator(op, _parallel_instances));
1408
8
                RETURN_IF_ERROR(cur_pipe->operators().front()->set_child(op));
1409
8
                cur_pipe = new_pipe;
1410
92.4k
            } else {
1411
92.4k
                op = std::make_shared<DistinctStreamingAggOperatorX>(pool, next_operator_id(),
1412
92.4k
                                                                     tnode, descs);
1413
92.4k
                RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1414
92.4k
            }
1415
92.4k
        } else if (is_streaming_agg) {
1416
3.20k
            if (need_create_cache_op) {
1417
0
                PipelinePtr new_pipe;
1418
0
                RETURN_IF_ERROR(create_query_cache_operator(new_pipe));
1419
0
                cache_op = op;
1420
0
                op = std::make_shared<StreamingAggOperatorX>(pool, next_operator_id(), tnode,
1421
0
                                                             descs);
1422
0
                RETURN_IF_ERROR(cur_pipe->operators().front()->set_child(op));
1423
0
                RETURN_IF_ERROR(new_pipe->add_operator(op, _parallel_instances));
1424
0
                cur_pipe = new_pipe;
1425
3.20k
            } else {
1426
3.20k
                op = std::make_shared<StreamingAggOperatorX>(pool, next_operator_id(), tnode,
1427
3.20k
                                                             descs);
1428
3.20k
                RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1429
3.20k
            }
1430
62.6k
        } else {
1431
            // create new pipeline to add query cache operator
1432
62.6k
            PipelinePtr new_pipe;
1433
62.6k
            if (need_create_cache_op) {
1434
2
                RETURN_IF_ERROR(create_query_cache_operator(new_pipe));
1435
2
                cache_op = op;
1436
2
            }
1437
1438
62.6k
            if (enable_spill) {
1439
177
                op = std::make_shared<PartitionedAggSourceOperatorX>(pool, tnode,
1440
177
                                                                     next_operator_id(), descs);
1441
62.4k
            } else {
1442
62.4k
                op = std::make_shared<AggSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1443
62.4k
            }
1444
62.6k
            if (need_create_cache_op) {
1445
2
                RETURN_IF_ERROR(cur_pipe->operators().front()->set_child(op));
1446
2
                RETURN_IF_ERROR(new_pipe->add_operator(op, _parallel_instances));
1447
2
                cur_pipe = new_pipe;
1448
62.6k
            } else {
1449
62.6k
                RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1450
62.6k
            }
1451
1452
62.6k
            const auto downstream_pipeline_id = cur_pipe->id();
1453
62.6k
            if (!_dag.contains(downstream_pipeline_id)) {
1454
60.6k
                _dag.insert({downstream_pipeline_id, {}});
1455
60.6k
            }
1456
62.6k
            cur_pipe = add_pipeline(cur_pipe);
1457
62.6k
            _dag[downstream_pipeline_id].push_back(cur_pipe->id());
1458
1459
62.6k
            if (enable_spill) {
1460
177
                sink_ops.push_back(std::make_shared<PartitionedAggSinkOperatorX>(
1461
177
                        pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1462
62.4k
            } else {
1463
62.4k
                sink_ops.push_back(std::make_shared<AggSinkOperatorX>(
1464
62.4k
                        pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1465
62.4k
            }
1466
62.6k
            RETURN_IF_ERROR(cur_pipe->set_sink(sink_ops.back()));
1467
62.6k
            RETURN_IF_ERROR(cur_pipe->sink()->init(tnode, _runtime_state.get()));
1468
62.6k
        }
1469
158k
        break;
1470
158k
    }
1471
158k
    case TPlanNodeType::BUCKETED_AGGREGATION_NODE: {
1472
75
        if (tnode.bucketed_agg_node.grouping_exprs.empty()) {
1473
0
            return Status::InternalError(
1474
0
                    "Bucketed aggregation node {} should not be used without group by keys",
1475
0
                    tnode.node_id);
1476
0
        }
1477
1478
        // Create source operator (goes on the current / downstream pipeline).
1479
75
        op = std::make_shared<BucketedAggSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1480
75
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1481
1482
        // Create a new pipeline for the sink side.
1483
75
        const auto downstream_pipeline_id = cur_pipe->id();
1484
75
        if (!_dag.contains(downstream_pipeline_id)) {
1485
75
            _dag.insert({downstream_pipeline_id, {}});
1486
75
        }
1487
75
        cur_pipe = add_pipeline(cur_pipe);
1488
75
        _dag[downstream_pipeline_id].push_back(cur_pipe->id());
1489
1490
        // Create sink operator.
1491
75
        sink_ops.push_back(std::make_shared<BucketedAggSinkOperatorX>(
1492
75
                pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1493
75
        RETURN_IF_ERROR(cur_pipe->set_sink(sink_ops.back()));
1494
75
        RETURN_IF_ERROR(cur_pipe->sink()->init(tnode, _runtime_state.get()));
1495
1496
        // Pre-register a single shared state for ALL instances so that every
1497
        // sink instance writes its per-instance hash table into the same
1498
        // BucketedAggSharedState and every source instance can merge across
1499
        // all of them.
1500
75
        {
1501
75
            auto shared_state = BucketedAggSharedState::create_shared();
1502
75
            shared_state->id = op->operator_id();
1503
75
            shared_state->related_op_ids.insert(op->operator_id());
1504
1505
489
            for (int i = 0; i < _num_instances; i++) {
1506
414
                auto sink_dep = std::make_shared<Dependency>(op->operator_id(), op->node_id(),
1507
414
                                                             "BUCKETED_AGG_SINK_DEPENDENCY");
1508
414
                sink_dep->set_shared_state(shared_state.get());
1509
414
                shared_state->sink_deps.push_back(sink_dep);
1510
414
            }
1511
75
            shared_state->create_source_dependencies(_num_instances, op->operator_id(),
1512
75
                                                     op->node_id(), "BUCKETED_AGG_SOURCE");
1513
75
            _op_id_to_shared_state.insert(
1514
75
                    {op->operator_id(), {shared_state, shared_state->sink_deps}});
1515
75
        }
1516
75
        break;
1517
75
    }
1518
9.63k
    case TPlanNodeType::HASH_JOIN_NODE: {
1519
9.63k
        const auto is_broadcast_join = tnode.hash_join_node.__isset.is_broadcast_join &&
1520
9.63k
                                       tnode.hash_join_node.is_broadcast_join;
1521
9.63k
        const auto enable_spill = _runtime_state->enable_spill();
1522
9.63k
        if (enable_spill && !is_broadcast_join) {
1523
0
            auto tnode_ = tnode;
1524
0
            tnode_.runtime_filters.clear();
1525
0
            auto inner_probe_operator =
1526
0
                    std::make_shared<HashJoinProbeOperatorX>(pool, tnode_, 0, descs);
1527
1528
            // probe side inner sink operator is used to build hash table on probe side when data is spilled.
1529
            // So here use `tnode_` which has no runtime filters.
1530
0
            auto probe_side_inner_sink_operator =
1531
0
                    std::make_shared<HashJoinBuildSinkOperatorX>(pool, 0, 0, tnode_, descs);
1532
1533
0
            RETURN_IF_ERROR(inner_probe_operator->init(tnode_, _runtime_state.get()));
1534
0
            RETURN_IF_ERROR(probe_side_inner_sink_operator->init(tnode_, _runtime_state.get()));
1535
1536
0
            auto probe_operator = std::make_shared<PartitionedHashJoinProbeOperatorX>(
1537
0
                    pool, tnode_, next_operator_id(), descs);
1538
0
            probe_operator->set_inner_operators(probe_side_inner_sink_operator,
1539
0
                                                inner_probe_operator);
1540
0
            op = std::move(probe_operator);
1541
0
            RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1542
1543
0
            const auto downstream_pipeline_id = cur_pipe->id();
1544
0
            if (!_dag.contains(downstream_pipeline_id)) {
1545
0
                _dag.insert({downstream_pipeline_id, {}});
1546
0
            }
1547
0
            PipelinePtr build_side_pipe = add_pipeline(cur_pipe);
1548
0
            _dag[downstream_pipeline_id].push_back(build_side_pipe->id());
1549
1550
0
            auto inner_sink_operator =
1551
0
                    std::make_shared<HashJoinBuildSinkOperatorX>(pool, 0, 0, tnode, descs);
1552
0
            auto sink_operator = std::make_shared<PartitionedHashJoinSinkOperatorX>(
1553
0
                    pool, next_sink_operator_id(), op->operator_id(), tnode_, descs);
1554
0
            RETURN_IF_ERROR(inner_sink_operator->init(tnode, _runtime_state.get()));
1555
1556
0
            sink_operator->set_inner_operators(inner_sink_operator, inner_probe_operator);
1557
0
            sink_ops.push_back(std::move(sink_operator));
1558
0
            RETURN_IF_ERROR(build_side_pipe->set_sink(sink_ops.back()));
1559
0
            RETURN_IF_ERROR(build_side_pipe->sink()->init(tnode_, _runtime_state.get()));
1560
1561
0
            _pipeline_parent_map.push(op->node_id(), cur_pipe);
1562
0
            _pipeline_parent_map.push(op->node_id(), build_side_pipe);
1563
9.63k
        } else {
1564
9.63k
            op = std::make_shared<HashJoinProbeOperatorX>(pool, tnode, next_operator_id(), descs);
1565
9.63k
            RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1566
1567
9.63k
            const auto downstream_pipeline_id = cur_pipe->id();
1568
9.63k
            if (!_dag.contains(downstream_pipeline_id)) {
1569
7.92k
                _dag.insert({downstream_pipeline_id, {}});
1570
7.92k
            }
1571
9.63k
            PipelinePtr build_side_pipe = add_pipeline(cur_pipe);
1572
9.63k
            _dag[downstream_pipeline_id].push_back(build_side_pipe->id());
1573
1574
9.63k
            sink_ops.push_back(std::make_shared<HashJoinBuildSinkOperatorX>(
1575
9.63k
                    pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1576
9.63k
            RETURN_IF_ERROR(build_side_pipe->set_sink(sink_ops.back()));
1577
9.63k
            RETURN_IF_ERROR(build_side_pipe->sink()->init(tnode, _runtime_state.get()));
1578
1579
9.63k
            _pipeline_parent_map.push(op->node_id(), cur_pipe);
1580
9.63k
            _pipeline_parent_map.push(op->node_id(), build_side_pipe);
1581
9.63k
        }
1582
9.63k
        if (is_broadcast_join && _runtime_state->enable_share_hash_table_for_broadcast_join()) {
1583
2.79k
            std::shared_ptr<HashJoinSharedState> shared_state =
1584
2.79k
                    HashJoinSharedState::create_shared(_num_instances);
1585
19.3k
            for (int i = 0; i < _num_instances; i++) {
1586
16.5k
                auto sink_dep = std::make_shared<Dependency>(op->operator_id(), op->node_id(),
1587
16.5k
                                                             "HASH_JOIN_BUILD_DEPENDENCY");
1588
16.5k
                sink_dep->set_shared_state(shared_state.get());
1589
16.5k
                shared_state->sink_deps.push_back(sink_dep);
1590
16.5k
            }
1591
2.79k
            shared_state->create_source_dependencies(_num_instances, op->operator_id(),
1592
2.79k
                                                     op->node_id(), "HASH_JOIN_PROBE");
1593
2.79k
            _op_id_to_shared_state.insert(
1594
2.79k
                    {op->operator_id(), {shared_state, shared_state->sink_deps}});
1595
2.79k
        }
1596
9.63k
        break;
1597
9.63k
    }
1598
4.91k
    case TPlanNodeType::CROSS_JOIN_NODE: {
1599
4.91k
        op = std::make_shared<NestedLoopJoinProbeOperatorX>(pool, tnode, next_operator_id(), descs);
1600
4.91k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1601
1602
4.91k
        const auto downstream_pipeline_id = cur_pipe->id();
1603
4.91k
        if (!_dag.contains(downstream_pipeline_id)) {
1604
4.68k
            _dag.insert({downstream_pipeline_id, {}});
1605
4.68k
        }
1606
4.91k
        PipelinePtr build_side_pipe = add_pipeline(cur_pipe);
1607
4.91k
        _dag[downstream_pipeline_id].push_back(build_side_pipe->id());
1608
1609
4.91k
        sink_ops.push_back(std::make_shared<NestedLoopJoinBuildSinkOperatorX>(
1610
4.91k
                pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1611
4.91k
        RETURN_IF_ERROR(build_side_pipe->set_sink(sink_ops.back()));
1612
4.91k
        RETURN_IF_ERROR(build_side_pipe->sink()->init(tnode, _runtime_state.get()));
1613
4.91k
        _pipeline_parent_map.push(op->node_id(), cur_pipe);
1614
4.91k
        _pipeline_parent_map.push(op->node_id(), build_side_pipe);
1615
4.91k
        break;
1616
4.91k
    }
1617
53.5k
    case TPlanNodeType::UNION_NODE: {
1618
53.5k
        int child_count = tnode.num_children;
1619
53.5k
        op = std::make_shared<UnionSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1620
53.5k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1621
1622
53.5k
        const auto downstream_pipeline_id = cur_pipe->id();
1623
53.5k
        if (!_dag.contains(downstream_pipeline_id)) {
1624
53.3k
            _dag.insert({downstream_pipeline_id, {}});
1625
53.3k
        }
1626
54.9k
        for (int i = 0; i < child_count; i++) {
1627
1.40k
            PipelinePtr build_side_pipe = add_pipeline(cur_pipe);
1628
1.40k
            _dag[downstream_pipeline_id].push_back(build_side_pipe->id());
1629
1.40k
            sink_ops.push_back(std::make_shared<UnionSinkOperatorX>(
1630
1.40k
                    i, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
1631
1.40k
            RETURN_IF_ERROR(build_side_pipe->set_sink(sink_ops.back()));
1632
1.40k
            RETURN_IF_ERROR(build_side_pipe->sink()->init(tnode, _runtime_state.get()));
1633
            // preset children pipelines. if any pipeline found this as its father, will use the prepared pipeline to build.
1634
1.40k
            _pipeline_parent_map.push(op->node_id(), build_side_pipe);
1635
1.40k
        }
1636
53.5k
        break;
1637
53.5k
    }
1638
53.5k
    case TPlanNodeType::SORT_NODE: {
1639
43.5k
        const auto should_spill = _runtime_state->enable_spill() &&
1640
43.5k
                                  tnode.sort_node.algorithm == TSortAlgorithm::FULL_SORT;
1641
43.5k
        const bool use_local_merge =
1642
43.5k
                tnode.sort_node.__isset.use_local_merge && tnode.sort_node.use_local_merge;
1643
43.5k
        if (should_spill) {
1644
7
            op = std::make_shared<SpillSortSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1645
43.5k
        } else if (use_local_merge) {
1646
41.1k
            op = std::make_shared<LocalMergeSortSourceOperatorX>(pool, tnode, next_operator_id(),
1647
41.1k
                                                                 descs);
1648
41.1k
        } else {
1649
2.31k
            op = std::make_shared<SortSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1650
2.31k
        }
1651
43.5k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1652
1653
43.5k
        const auto downstream_pipeline_id = cur_pipe->id();
1654
43.5k
        if (!_dag.contains(downstream_pipeline_id)) {
1655
43.4k
            _dag.insert({downstream_pipeline_id, {}});
1656
43.4k
        }
1657
43.5k
        cur_pipe = add_pipeline(cur_pipe);
1658
43.5k
        _dag[downstream_pipeline_id].push_back(cur_pipe->id());
1659
1660
43.5k
        if (should_spill) {
1661
7
            sink_ops.push_back(std::make_shared<SpillSortSinkOperatorX>(
1662
7
                    pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1663
43.5k
        } else {
1664
43.5k
            sink_ops.push_back(std::make_shared<SortSinkOperatorX>(
1665
43.5k
                    pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1666
43.5k
        }
1667
43.5k
        RETURN_IF_ERROR(cur_pipe->set_sink(sink_ops.back()));
1668
43.5k
        RETURN_IF_ERROR(cur_pipe->sink()->init(tnode, _runtime_state.get()));
1669
43.5k
        break;
1670
43.5k
    }
1671
43.5k
    case TPlanNodeType::PARTITION_SORT_NODE: {
1672
62
        op = std::make_shared<PartitionSortSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1673
62
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1674
1675
62
        const auto downstream_pipeline_id = cur_pipe->id();
1676
62
        if (!_dag.contains(downstream_pipeline_id)) {
1677
62
            _dag.insert({downstream_pipeline_id, {}});
1678
62
        }
1679
62
        cur_pipe = add_pipeline(cur_pipe);
1680
62
        _dag[downstream_pipeline_id].push_back(cur_pipe->id());
1681
1682
62
        sink_ops.push_back(std::make_shared<PartitionSortSinkOperatorX>(
1683
62
                pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1684
62
        RETURN_IF_ERROR(cur_pipe->set_sink(sink_ops.back()));
1685
62
        RETURN_IF_ERROR(cur_pipe->sink()->init(tnode, _runtime_state.get()));
1686
62
        break;
1687
62
    }
1688
1.64k
    case TPlanNodeType::ANALYTIC_EVAL_NODE: {
1689
1.64k
        op = std::make_shared<AnalyticSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1690
1.64k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1691
1692
1.64k
        const auto downstream_pipeline_id = cur_pipe->id();
1693
1.64k
        if (!_dag.contains(downstream_pipeline_id)) {
1694
1.63k
            _dag.insert({downstream_pipeline_id, {}});
1695
1.63k
        }
1696
1.64k
        cur_pipe = add_pipeline(cur_pipe);
1697
1.64k
        _dag[downstream_pipeline_id].push_back(cur_pipe->id());
1698
1699
1.64k
        sink_ops.push_back(std::make_shared<AnalyticSinkOperatorX>(
1700
1.64k
                pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1701
1.64k
        RETURN_IF_ERROR(cur_pipe->set_sink(sink_ops.back()));
1702
1.64k
        RETURN_IF_ERROR(cur_pipe->sink()->init(tnode, _runtime_state.get()));
1703
1.64k
        break;
1704
1.64k
    }
1705
1.64k
    case TPlanNodeType::MATERIALIZATION_NODE: {
1706
1.60k
        op = std::make_shared<MaterializationOperator>(pool, tnode, next_operator_id(), descs);
1707
1.60k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1708
1.60k
        break;
1709
1.60k
    }
1710
1.60k
    case TPlanNodeType::INTERSECT_NODE: {
1711
134
        RETURN_IF_ERROR(_build_operators_for_set_operation_node<true>(pool, tnode, descs, op,
1712
134
                                                                      cur_pipe, sink_ops));
1713
134
        break;
1714
134
    }
1715
134
    case TPlanNodeType::EXCEPT_NODE: {
1716
131
        RETURN_IF_ERROR(_build_operators_for_set_operation_node<false>(pool, tnode, descs, op,
1717
131
                                                                       cur_pipe, sink_ops));
1718
131
        break;
1719
131
    }
1720
296
    case TPlanNodeType::REPEAT_NODE: {
1721
296
        op = std::make_shared<RepeatOperatorX>(pool, tnode, next_operator_id(), descs);
1722
296
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1723
296
        break;
1724
296
    }
1725
911
    case TPlanNodeType::TABLE_FUNCTION_NODE: {
1726
911
        op = std::make_shared<TableFunctionOperatorX>(pool, tnode, next_operator_id(), descs);
1727
911
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1728
911
        break;
1729
911
    }
1730
911
    case TPlanNodeType::ASSERT_NUM_ROWS_NODE: {
1731
218
        op = std::make_shared<AssertNumRowsOperatorX>(pool, tnode, next_operator_id(), descs);
1732
218
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1733
218
        break;
1734
218
    }
1735
1.60k
    case TPlanNodeType::EMPTY_SET_NODE: {
1736
1.60k
        op = std::make_shared<EmptySetSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1737
1.60k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1738
1.60k
        break;
1739
1.60k
    }
1740
1.60k
    case TPlanNodeType::DATA_GEN_SCAN_NODE: {
1741
457
        op = std::make_shared<DataGenSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1742
457
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1743
457
        fe_with_old_version = !tnode.__isset.is_serial_operator;
1744
457
        break;
1745
457
    }
1746
2.02k
    case TPlanNodeType::SCHEMA_SCAN_NODE: {
1747
2.02k
        op = std::make_shared<SchemaScanOperatorX>(pool, tnode, next_operator_id(), descs);
1748
2.02k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1749
2.02k
        break;
1750
2.02k
    }
1751
6.06k
    case TPlanNodeType::META_SCAN_NODE: {
1752
6.06k
        op = std::make_shared<MetaScanOperatorX>(pool, tnode, next_operator_id(), descs);
1753
6.06k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1754
6.06k
        break;
1755
6.06k
    }
1756
6.06k
    case TPlanNodeType::SELECT_NODE: {
1757
1.97k
        op = std::make_shared<SelectOperatorX>(pool, tnode, next_operator_id(), descs);
1758
1.97k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1759
1.97k
        break;
1760
1.97k
    }
1761
1.97k
    case TPlanNodeType::REC_CTE_NODE: {
1762
151
        op = std::make_shared<RecCTESourceOperatorX>(pool, tnode, next_operator_id(), descs);
1763
151
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1764
1765
151
        const auto downstream_pipeline_id = cur_pipe->id();
1766
151
        if (!_dag.contains(downstream_pipeline_id)) {
1767
148
            _dag.insert({downstream_pipeline_id, {}});
1768
148
        }
1769
1770
151
        PipelinePtr anchor_side_pipe = add_pipeline(cur_pipe);
1771
151
        _dag[downstream_pipeline_id].push_back(anchor_side_pipe->id());
1772
1773
151
        DataSinkOperatorPtr anchor_sink;
1774
151
        anchor_sink = std::make_shared<RecCTEAnchorSinkOperatorX>(next_sink_operator_id(),
1775
151
                                                                  op->operator_id(), tnode, descs);
1776
151
        RETURN_IF_ERROR(anchor_side_pipe->set_sink(anchor_sink));
1777
151
        RETURN_IF_ERROR(anchor_side_pipe->sink()->init(tnode, _runtime_state.get()));
1778
151
        _pipeline_parent_map.push(op->node_id(), anchor_side_pipe);
1779
1780
151
        PipelinePtr rec_side_pipe = add_pipeline(cur_pipe);
1781
151
        _dag[downstream_pipeline_id].push_back(rec_side_pipe->id());
1782
1783
151
        DataSinkOperatorPtr rec_sink;
1784
151
        rec_sink = std::make_shared<RecCTESinkOperatorX>(next_sink_operator_id(), op->operator_id(),
1785
151
                                                         tnode, descs);
1786
151
        RETURN_IF_ERROR(rec_side_pipe->set_sink(rec_sink));
1787
151
        RETURN_IF_ERROR(rec_side_pipe->sink()->init(tnode, _runtime_state.get()));
1788
151
        _pipeline_parent_map.push(op->node_id(), rec_side_pipe);
1789
1790
151
        break;
1791
151
    }
1792
1.95k
    case TPlanNodeType::REC_CTE_SCAN_NODE: {
1793
1.95k
        op = std::make_shared<RecCTEScanOperatorX>(pool, tnode, next_operator_id(), descs);
1794
1.95k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1795
1.95k
        break;
1796
1.95k
    }
1797
1.95k
    default:
1798
0
        return Status::InternalError("Unsupported exec type in pipeline: {}",
1799
0
                                     print_plan_node_type(tnode.node_type));
1800
673k
    }
1801
672k
    if (_params.__isset.parallel_instances && fe_with_old_version) {
1802
0
        cur_pipe->set_num_tasks(_params.parallel_instances);
1803
0
        op->set_serial_operator();
1804
0
    }
1805
1806
672k
    return Status::OK();
1807
673k
}
1808
// NOLINTEND(readability-function-cognitive-complexity)
1809
// NOLINTEND(readability-function-size)
1810
1811
template <bool is_intersect>
1812
Status PipelineFragmentContext::_build_operators_for_set_operation_node(
1813
        ObjectPool* pool, const TPlanNode& tnode, const DescriptorTbl& descs, OperatorPtr& op,
1814
265
        PipelinePtr& cur_pipe, std::vector<DataSinkOperatorPtr>& sink_ops) {
1815
265
    op.reset(new SetSourceOperatorX<is_intersect>(pool, tnode, next_operator_id(), descs));
1816
265
    RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1817
1818
265
    const auto downstream_pipeline_id = cur_pipe->id();
1819
265
    if (!_dag.contains(downstream_pipeline_id)) {
1820
240
        _dag.insert({downstream_pipeline_id, {}});
1821
240
    }
1822
1823
889
    for (int child_id = 0; child_id < tnode.num_children; child_id++) {
1824
624
        PipelinePtr probe_side_pipe = add_pipeline(cur_pipe);
1825
624
        _dag[downstream_pipeline_id].push_back(probe_side_pipe->id());
1826
1827
624
        if (child_id == 0) {
1828
265
            sink_ops.push_back(std::make_shared<SetSinkOperatorX<is_intersect>>(
1829
265
                    child_id, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
1830
359
        } else {
1831
359
            sink_ops.push_back(std::make_shared<SetProbeSinkOperatorX<is_intersect>>(
1832
359
                    child_id, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
1833
359
        }
1834
624
        RETURN_IF_ERROR(probe_side_pipe->set_sink(sink_ops.back()));
1835
624
        RETURN_IF_ERROR(probe_side_pipe->sink()->init(tnode, _runtime_state.get()));
1836
        // prepare children pipelines. if any pipeline found this as its father, will use the prepared pipeline to build.
1837
624
        _pipeline_parent_map.push(op->node_id(), probe_side_pipe);
1838
624
    }
1839
1840
265
    return Status::OK();
1841
265
}
_ZN5doris23PipelineFragmentContext39_build_operators_for_set_operation_nodeILb1EEENS_6StatusEPNS_10ObjectPoolERKNS_9TPlanNodeERKNS_13DescriptorTblERSt10shared_ptrINS_13OperatorXBaseEERSB_INS_8PipelineEERSt6vectorISB_INS_21DataSinkOperatorXBaseEESaISK_EE
Line
Count
Source
1814
134
        PipelinePtr& cur_pipe, std::vector<DataSinkOperatorPtr>& sink_ops) {
1815
134
    op.reset(new SetSourceOperatorX<is_intersect>(pool, tnode, next_operator_id(), descs));
1816
134
    RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1817
1818
134
    const auto downstream_pipeline_id = cur_pipe->id();
1819
134
    if (!_dag.contains(downstream_pipeline_id)) {
1820
118
        _dag.insert({downstream_pipeline_id, {}});
1821
118
    }
1822
1823
481
    for (int child_id = 0; child_id < tnode.num_children; child_id++) {
1824
347
        PipelinePtr probe_side_pipe = add_pipeline(cur_pipe);
1825
347
        _dag[downstream_pipeline_id].push_back(probe_side_pipe->id());
1826
1827
347
        if (child_id == 0) {
1828
134
            sink_ops.push_back(std::make_shared<SetSinkOperatorX<is_intersect>>(
1829
134
                    child_id, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
1830
213
        } else {
1831
213
            sink_ops.push_back(std::make_shared<SetProbeSinkOperatorX<is_intersect>>(
1832
213
                    child_id, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
1833
213
        }
1834
347
        RETURN_IF_ERROR(probe_side_pipe->set_sink(sink_ops.back()));
1835
347
        RETURN_IF_ERROR(probe_side_pipe->sink()->init(tnode, _runtime_state.get()));
1836
        // prepare children pipelines. if any pipeline found this as its father, will use the prepared pipeline to build.
1837
347
        _pipeline_parent_map.push(op->node_id(), probe_side_pipe);
1838
347
    }
1839
1840
134
    return Status::OK();
1841
134
}
_ZN5doris23PipelineFragmentContext39_build_operators_for_set_operation_nodeILb0EEENS_6StatusEPNS_10ObjectPoolERKNS_9TPlanNodeERKNS_13DescriptorTblERSt10shared_ptrINS_13OperatorXBaseEERSB_INS_8PipelineEERSt6vectorISB_INS_21DataSinkOperatorXBaseEESaISK_EE
Line
Count
Source
1814
131
        PipelinePtr& cur_pipe, std::vector<DataSinkOperatorPtr>& sink_ops) {
1815
131
    op.reset(new SetSourceOperatorX<is_intersect>(pool, tnode, next_operator_id(), descs));
1816
131
    RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1817
1818
131
    const auto downstream_pipeline_id = cur_pipe->id();
1819
131
    if (!_dag.contains(downstream_pipeline_id)) {
1820
122
        _dag.insert({downstream_pipeline_id, {}});
1821
122
    }
1822
1823
408
    for (int child_id = 0; child_id < tnode.num_children; child_id++) {
1824
277
        PipelinePtr probe_side_pipe = add_pipeline(cur_pipe);
1825
277
        _dag[downstream_pipeline_id].push_back(probe_side_pipe->id());
1826
1827
277
        if (child_id == 0) {
1828
131
            sink_ops.push_back(std::make_shared<SetSinkOperatorX<is_intersect>>(
1829
131
                    child_id, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
1830
146
        } else {
1831
146
            sink_ops.push_back(std::make_shared<SetProbeSinkOperatorX<is_intersect>>(
1832
146
                    child_id, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
1833
146
        }
1834
277
        RETURN_IF_ERROR(probe_side_pipe->set_sink(sink_ops.back()));
1835
277
        RETURN_IF_ERROR(probe_side_pipe->sink()->init(tnode, _runtime_state.get()));
1836
        // prepare children pipelines. if any pipeline found this as its father, will use the prepared pipeline to build.
1837
277
        _pipeline_parent_map.push(op->node_id(), probe_side_pipe);
1838
277
    }
1839
1840
131
    return Status::OK();
1841
131
}
1842
1843
431k
Status PipelineFragmentContext::submit() {
1844
431k
    if (_submitted) {
1845
0
        return Status::InternalError("submitted");
1846
0
    }
1847
431k
    _submitted = true;
1848
1849
431k
    int submit_tasks = 0;
1850
431k
    Status st;
1851
431k
    auto* scheduler = _query_ctx->get_pipe_exec_scheduler();
1852
1.21M
    for (auto& task : _tasks) {
1853
1.98M
        for (auto& t : task) {
1854
1.98M
            st = scheduler->submit(t.first);
1855
1.98M
            DBUG_EXECUTE_IF("PipelineFragmentContext.submit.failed",
1856
1.98M
                            { st = Status::Aborted("PipelineFragmentContext.submit.failed"); });
1857
1.98M
            if (!st) {
1858
0
                cancel(Status::InternalError("submit context to executor fail"));
1859
0
                std::lock_guard<std::mutex> l(_task_mutex);
1860
0
                _total_tasks = submit_tasks;
1861
0
                break;
1862
0
            }
1863
1.98M
            submit_tasks++;
1864
1.98M
        }
1865
1.21M
    }
1866
431k
    if (!st.ok()) {
1867
0
        bool need_remove = false;
1868
0
        {
1869
0
            std::lock_guard<std::mutex> l(_task_mutex);
1870
0
            if (_closed_tasks >= _total_tasks) {
1871
0
                need_remove = _close_fragment_instance();
1872
0
            }
1873
0
        }
1874
        // Call remove_pipeline_context() outside _task_mutex to avoid ABBA deadlock.
1875
0
        if (need_remove) {
1876
0
            _exec_env->fragment_mgr()->remove_pipeline_context({_query_id, _fragment_id});
1877
0
        }
1878
0
        return Status::InternalError("Submit pipeline failed. err = {}, BE: {}", st.to_string(),
1879
0
                                     BackendOptions::get_localhost());
1880
431k
    } else {
1881
431k
        return st;
1882
431k
    }
1883
431k
}
1884
1885
0
void PipelineFragmentContext::print_profile(const std::string& extra_info) {
1886
0
    if (_runtime_state->enable_profile()) {
1887
0
        std::stringstream ss;
1888
0
        for (auto runtime_profile_ptr : _runtime_state->pipeline_id_to_profile()) {
1889
0
            runtime_profile_ptr->pretty_print(&ss);
1890
0
        }
1891
1892
0
        if (_runtime_state->load_channel_profile()) {
1893
0
            _runtime_state->load_channel_profile()->pretty_print(&ss);
1894
0
        }
1895
1896
0
        auto profile_str =
1897
0
                fmt::format("Query {} fragment {} {}, profile, {}", print_id(this->_query_id),
1898
0
                            this->_fragment_id, extra_info, ss.str());
1899
0
        LOG_LONG_STRING(INFO, profile_str);
1900
0
    }
1901
0
}
1902
// If all pipeline tasks binded to the fragment instance are finished, then we could
1903
// close the fragment instance.
1904
// Returns true if the caller should call remove_pipeline_context() **after** releasing
1905
// _task_mutex. We must not call remove_pipeline_context() here because it acquires
1906
// _pipeline_map's shard lock, and this function is called while _task_mutex is held.
1907
// Acquiring _pipeline_map while holding _task_mutex creates an ABBA deadlock with
1908
// dump_pipeline_tasks(), which acquires _pipeline_map first and then _task_mutex
1909
// (via debug_string()).
1910
432k
bool PipelineFragmentContext::_close_fragment_instance() {
1911
432k
    if (_is_fragment_instance_closed) {
1912
0
        return false;
1913
0
    }
1914
432k
    Defer defer_op {[&]() { _is_fragment_instance_closed = true; }};
1915
432k
    _fragment_level_profile->total_time_counter()->update(_fragment_watcher.elapsed_time());
1916
432k
    if (!_need_notify_close) {
1917
429k
        auto st = send_report(true);
1918
429k
        if (!st) {
1919
0
            LOG(WARNING) << fmt::format("Failed to send report for query {}, fragment {}: {}",
1920
0
                                        print_id(_query_id), _fragment_id, st.to_string());
1921
0
        }
1922
429k
    }
1923
    // Print profile content in info log is a tempoeray solution for stream load and external_connector.
1924
    // Since stream load does not have someting like coordinator on FE, so
1925
    // backend can not report profile to FE, ant its profile can not be shown
1926
    // in the same way with other query. So we print the profile content to info log.
1927
1928
432k
    if (_runtime_state->enable_profile() &&
1929
432k
        (_query_ctx->get_query_source() == QuerySource::STREAM_LOAD ||
1930
2.23k
         _query_ctx->get_query_source() == QuerySource::EXTERNAL_CONNECTOR ||
1931
2.23k
         _query_ctx->get_query_source() == QuerySource::GROUP_COMMIT_LOAD)) {
1932
0
        std::stringstream ss;
1933
        // Compute the _local_time_percent before pretty_print the runtime_profile
1934
        // Before add this operation, the print out like that:
1935
        // UNION_NODE (id=0):(Active: 56.720us, non-child: 00.00%)
1936
        // After add the operation, the print out like that:
1937
        // UNION_NODE (id=0):(Active: 56.720us, non-child: 82.53%)
1938
        // We can easily know the exec node execute time without child time consumed.
1939
0
        for (auto runtime_profile_ptr : _runtime_state->pipeline_id_to_profile()) {
1940
0
            runtime_profile_ptr->pretty_print(&ss);
1941
0
        }
1942
1943
0
        if (_runtime_state->load_channel_profile()) {
1944
0
            _runtime_state->load_channel_profile()->pretty_print(&ss);
1945
0
        }
1946
1947
0
        LOG_INFO("Query {} fragment {} profile:\n {}", print_id(_query_id), _fragment_id, ss.str());
1948
0
    }
1949
1950
432k
    if (_query_ctx->enable_profile()) {
1951
2.23k
        _query_ctx->add_fragment_profile(_fragment_id, collect_realtime_profile(),
1952
2.23k
                                         collect_realtime_load_channel_profile());
1953
2.23k
    }
1954
1955
    // Return whether the caller needs to remove from the pipeline map.
1956
    // The caller must do this after releasing _task_mutex.
1957
432k
    return !_need_notify_close;
1958
432k
}
1959
1960
1.97M
void PipelineFragmentContext::decrement_running_task(PipelineId pipeline_id) {
1961
    // If all tasks of this pipeline has been closed, upstream tasks is never needed, and we just make those runnable here
1962
1.97M
    DCHECK(_pip_id_to_pipeline.contains(pipeline_id));
1963
1.97M
    if (_pip_id_to_pipeline[pipeline_id]->close_task()) {
1964
673k
        if (_dag.contains(pipeline_id)) {
1965
350k
            for (auto dep : _dag[pipeline_id]) {
1966
350k
                _pip_id_to_pipeline[dep]->make_all_runnable(pipeline_id);
1967
350k
            }
1968
282k
        }
1969
673k
    }
1970
1.97M
    bool need_remove = false;
1971
1.97M
    {
1972
1.97M
        std::lock_guard<std::mutex> l(_task_mutex);
1973
1.97M
        ++_closed_tasks;
1974
1.97M
        if (_closed_tasks >= _total_tasks) {
1975
432k
            need_remove = _close_fragment_instance();
1976
432k
        }
1977
1.97M
    }
1978
    // Call remove_pipeline_context() outside _task_mutex to avoid ABBA deadlock.
1979
1.97M
    if (need_remove) {
1980
429k
        _exec_env->fragment_mgr()->remove_pipeline_context({_query_id, _fragment_id});
1981
429k
    }
1982
1.97M
}
1983
1984
53.8k
std::string PipelineFragmentContext::get_load_error_url() {
1985
53.8k
    if (const auto& str = _runtime_state->get_error_log_file_path(); !str.empty()) {
1986
0
        return to_load_error_http_path(str);
1987
0
    }
1988
144k
    for (auto& tasks : _tasks) {
1989
231k
        for (auto& task : tasks) {
1990
231k
            if (const auto& str = task.second->get_error_log_file_path(); !str.empty()) {
1991
172
                return to_load_error_http_path(str);
1992
172
            }
1993
231k
        }
1994
144k
    }
1995
53.6k
    return "";
1996
53.8k
}
1997
1998
53.8k
std::string PipelineFragmentContext::get_first_error_msg() {
1999
53.8k
    if (const auto& str = _runtime_state->get_first_error_msg(); !str.empty()) {
2000
0
        return str;
2001
0
    }
2002
144k
    for (auto& tasks : _tasks) {
2003
231k
        for (auto& task : tasks) {
2004
231k
            if (const auto& str = task.second->get_first_error_msg(); !str.empty()) {
2005
172
                return str;
2006
172
            }
2007
231k
        }
2008
144k
    }
2009
53.6k
    return "";
2010
53.8k
}
2011
2012
0
std::string PipelineFragmentContext::_to_http_path(const std::string& file_name) const {
2013
0
    std::stringstream url;
2014
0
    url << "http://" << BackendOptions::get_localhost() << ":" << config::webserver_port
2015
0
        << "/api/_download_load?"
2016
0
        << "token=" << _exec_env->token() << "&file=" << file_name;
2017
0
    return url.str();
2018
0
}
2019
2020
47.7k
void PipelineFragmentContext::_coordinator_callback(const ReportStatusRequest& req) {
2021
47.7k
    DBUG_EXECUTE_IF("FragmentMgr::coordinator_callback.report_delay", {
2022
47.7k
        int random_seconds = req.status.is<ErrorCode::DATA_QUALITY_ERROR>() ? 8 : 2;
2023
47.7k
        LOG_INFO("sleep : ").tag("time", random_seconds).tag("query_id", print_id(req.query_id));
2024
47.7k
        std::this_thread::sleep_for(std::chrono::seconds(random_seconds));
2025
47.7k
        LOG_INFO("sleep done").tag("query_id", print_id(req.query_id));
2026
47.7k
    });
2027
2028
47.7k
    DCHECK(req.status.ok() || req.done); // if !status.ok() => done
2029
47.7k
    if (req.coord_addr.hostname == "external") {
2030
        // External query (flink/spark read tablets) not need to report to FE.
2031
0
        return;
2032
0
    }
2033
47.7k
    int callback_retries = 10;
2034
47.7k
    const int sleep_ms = 1000;
2035
47.7k
    Status exec_status = req.status;
2036
47.7k
    Status coord_status;
2037
47.7k
    std::unique_ptr<FrontendServiceConnection> coord = nullptr;
2038
47.7k
    do {
2039
47.7k
        coord = std::make_unique<FrontendServiceConnection>(_exec_env->frontend_client_cache(),
2040
47.7k
                                                            req.coord_addr, &coord_status);
2041
47.7k
        if (!coord_status.ok()) {
2042
0
            std::this_thread::sleep_for(std::chrono::milliseconds(sleep_ms));
2043
0
        }
2044
47.7k
    } while (!coord_status.ok() && callback_retries-- > 0);
2045
2046
47.7k
    if (!coord_status.ok()) {
2047
0
        UniqueId uid(req.query_id.hi, req.query_id.lo);
2048
0
        static_cast<void>(req.cancel_fn(Status::InternalError(
2049
0
                "query_id: {}, couldn't get a client for {}, reason is {}", uid.to_string(),
2050
0
                PrintThriftNetworkAddress(req.coord_addr), coord_status.to_string())));
2051
0
        return;
2052
0
    }
2053
2054
47.7k
    TReportExecStatusParams params;
2055
47.7k
    params.protocol_version = FrontendServiceVersion::V1;
2056
47.7k
    params.__set_query_id(req.query_id);
2057
47.7k
    params.__set_backend_num(req.backend_num);
2058
47.7k
    params.__set_fragment_instance_id(req.fragment_instance_id);
2059
47.7k
    params.__set_fragment_id(req.fragment_id);
2060
47.7k
    params.__set_status(exec_status.to_thrift());
2061
47.7k
    params.__set_done(req.done);
2062
47.7k
    params.__set_query_type(req.runtime_state->query_type());
2063
47.7k
    params.__isset.profile = false;
2064
2065
47.7k
    DCHECK(req.runtime_state != nullptr);
2066
2067
47.7k
    if (req.runtime_state->query_type() == TQueryType::LOAD) {
2068
43.4k
        params.__set_loaded_rows(req.runtime_state->num_rows_load_total());
2069
43.4k
        params.__set_loaded_bytes(req.runtime_state->num_bytes_load_total());
2070
43.4k
    } else {
2071
4.30k
        DCHECK(!req.runtime_states.empty());
2072
4.30k
        if (!req.runtime_state->output_files().empty()) {
2073
0
            params.__isset.delta_urls = true;
2074
0
            for (auto& it : req.runtime_state->output_files()) {
2075
0
                params.delta_urls.push_back(_to_http_path(it));
2076
0
            }
2077
0
        }
2078
4.30k
        if (!params.delta_urls.empty()) {
2079
0
            params.__isset.delta_urls = true;
2080
0
        }
2081
4.30k
    }
2082
2083
47.7k
    static std::string s_dpp_normal_all = "dpp.norm.ALL";
2084
47.7k
    static std::string s_dpp_abnormal_all = "dpp.abnorm.ALL";
2085
47.7k
    static std::string s_unselected_rows = "unselected.rows";
2086
47.7k
    int64_t num_rows_load_success = 0;
2087
47.7k
    int64_t num_rows_load_filtered = 0;
2088
47.7k
    int64_t num_rows_load_unselected = 0;
2089
47.7k
    if (req.runtime_state->num_rows_load_total() > 0 ||
2090
47.7k
        req.runtime_state->num_rows_load_filtered() > 0 ||
2091
47.7k
        req.runtime_state->num_finished_range() > 0) {
2092
0
        params.__isset.load_counters = true;
2093
2094
0
        num_rows_load_success = req.runtime_state->num_rows_load_success();
2095
0
        num_rows_load_filtered = req.runtime_state->num_rows_load_filtered();
2096
0
        num_rows_load_unselected = req.runtime_state->num_rows_load_unselected();
2097
0
        params.__isset.fragment_instance_reports = true;
2098
0
        TFragmentInstanceReport t;
2099
0
        t.__set_fragment_instance_id(req.runtime_state->fragment_instance_id());
2100
0
        t.__set_num_finished_range(cast_set<int>(req.runtime_state->num_finished_range()));
2101
0
        t.__set_loaded_rows(req.runtime_state->num_rows_load_total());
2102
0
        t.__set_loaded_bytes(req.runtime_state->num_bytes_load_total());
2103
0
        params.fragment_instance_reports.push_back(t);
2104
47.7k
    } else if (!req.runtime_states.empty()) {
2105
173k
        for (auto* rs : req.runtime_states) {
2106
173k
            if (rs->num_rows_load_total() > 0 || rs->num_rows_load_filtered() > 0 ||
2107
173k
                rs->num_finished_range() > 0) {
2108
36.1k
                params.__isset.load_counters = true;
2109
36.1k
                num_rows_load_success += rs->num_rows_load_success();
2110
36.1k
                num_rows_load_filtered += rs->num_rows_load_filtered();
2111
36.1k
                num_rows_load_unselected += rs->num_rows_load_unselected();
2112
36.1k
                params.__isset.fragment_instance_reports = true;
2113
36.1k
                TFragmentInstanceReport t;
2114
36.1k
                t.__set_fragment_instance_id(rs->fragment_instance_id());
2115
36.1k
                t.__set_num_finished_range(cast_set<int>(rs->num_finished_range()));
2116
36.1k
                t.__set_loaded_rows(rs->num_rows_load_total());
2117
36.1k
                t.__set_loaded_bytes(rs->num_bytes_load_total());
2118
36.1k
                params.fragment_instance_reports.push_back(t);
2119
36.1k
            }
2120
173k
        }
2121
47.7k
    }
2122
47.7k
    params.load_counters.emplace(s_dpp_normal_all, std::to_string(num_rows_load_success));
2123
47.7k
    params.load_counters.emplace(s_dpp_abnormal_all, std::to_string(num_rows_load_filtered));
2124
47.7k
    params.load_counters.emplace(s_unselected_rows, std::to_string(num_rows_load_unselected));
2125
2126
47.7k
    if (!req.load_error_url.empty()) {
2127
158
        params.__set_tracking_url(req.load_error_url);
2128
158
    }
2129
47.7k
    if (!req.first_error_msg.empty()) {
2130
158
        params.__set_first_error_msg(req.first_error_msg);
2131
158
    }
2132
173k
    for (auto* rs : req.runtime_states) {
2133
173k
        if (rs->wal_id() > 0) {
2134
112
            params.__set_txn_id(rs->wal_id());
2135
112
            params.__set_label(rs->import_label());
2136
112
        }
2137
173k
    }
2138
47.7k
    if (!req.runtime_state->export_output_files().empty()) {
2139
0
        params.__isset.export_files = true;
2140
0
        params.export_files = req.runtime_state->export_output_files();
2141
47.7k
    } else if (!req.runtime_states.empty()) {
2142
173k
        for (auto* rs : req.runtime_states) {
2143
173k
            if (!rs->export_output_files().empty()) {
2144
0
                params.__isset.export_files = true;
2145
0
                params.export_files.insert(params.export_files.end(),
2146
0
                                           rs->export_output_files().begin(),
2147
0
                                           rs->export_output_files().end());
2148
0
            }
2149
173k
        }
2150
47.7k
    }
2151
47.7k
    if (auto tci = req.runtime_state->tablet_commit_infos(); !tci.empty()) {
2152
0
        params.__isset.commitInfos = true;
2153
0
        params.commitInfos.insert(params.commitInfos.end(), tci.begin(), tci.end());
2154
47.7k
    } else if (!req.runtime_states.empty()) {
2155
173k
        for (auto* rs : req.runtime_states) {
2156
173k
            if (auto rs_tci = rs->tablet_commit_infos(); !rs_tci.empty()) {
2157
27.0k
                params.__isset.commitInfos = true;
2158
27.0k
                params.commitInfos.insert(params.commitInfos.end(), rs_tci.begin(), rs_tci.end());
2159
27.0k
            }
2160
173k
        }
2161
47.7k
    }
2162
47.7k
    if (auto eti = req.runtime_state->error_tablet_infos(); !eti.empty()) {
2163
0
        params.__isset.errorTabletInfos = true;
2164
0
        params.errorTabletInfos.insert(params.errorTabletInfos.end(), eti.begin(), eti.end());
2165
47.7k
    } else if (!req.runtime_states.empty()) {
2166
173k
        for (auto* rs : req.runtime_states) {
2167
173k
            if (auto rs_eti = rs->error_tablet_infos(); !rs_eti.empty()) {
2168
0
                params.__isset.errorTabletInfos = true;
2169
0
                params.errorTabletInfos.insert(params.errorTabletInfos.end(), rs_eti.begin(),
2170
0
                                               rs_eti.end());
2171
0
            }
2172
173k
        }
2173
47.7k
    }
2174
47.7k
    if (auto hpu = req.runtime_state->hive_partition_updates(); !hpu.empty()) {
2175
0
        params.__isset.hive_partition_updates = true;
2176
0
        params.hive_partition_updates.insert(params.hive_partition_updates.end(), hpu.begin(),
2177
0
                                             hpu.end());
2178
47.7k
    } else if (!req.runtime_states.empty()) {
2179
173k
        for (auto* rs : req.runtime_states) {
2180
173k
            if (auto rs_hpu = rs->hive_partition_updates(); !rs_hpu.empty()) {
2181
2.12k
                params.__isset.hive_partition_updates = true;
2182
2.12k
                params.hive_partition_updates.insert(params.hive_partition_updates.end(),
2183
2.12k
                                                     rs_hpu.begin(), rs_hpu.end());
2184
2.12k
            }
2185
173k
        }
2186
47.7k
    }
2187
47.7k
    if (auto icd = req.runtime_state->iceberg_commit_datas(); !icd.empty()) {
2188
0
        params.__isset.iceberg_commit_datas = true;
2189
0
        params.iceberg_commit_datas.insert(params.iceberg_commit_datas.end(), icd.begin(),
2190
0
                                           icd.end());
2191
47.7k
    } else if (!req.runtime_states.empty()) {
2192
173k
        for (auto* rs : req.runtime_states) {
2193
173k
            if (auto rs_icd = rs->iceberg_commit_datas(); !rs_icd.empty()) {
2194
2.08k
                params.__isset.iceberg_commit_datas = true;
2195
2.08k
                params.iceberg_commit_datas.insert(params.iceberg_commit_datas.end(),
2196
2.08k
                                                   rs_icd.begin(), rs_icd.end());
2197
2.08k
            }
2198
173k
        }
2199
47.7k
    }
2200
2201
47.7k
    if (auto mcd = req.runtime_state->mc_commit_datas(); !mcd.empty()) {
2202
0
        params.__isset.mc_commit_datas = true;
2203
0
        params.mc_commit_datas.insert(params.mc_commit_datas.end(), mcd.begin(), mcd.end());
2204
47.7k
    } else if (!req.runtime_states.empty()) {
2205
173k
        for (auto* rs : req.runtime_states) {
2206
173k
            if (auto rs_mcd = rs->mc_commit_datas(); !rs_mcd.empty()) {
2207
0
                params.__isset.mc_commit_datas = true;
2208
0
                params.mc_commit_datas.insert(params.mc_commit_datas.end(), rs_mcd.begin(),
2209
0
                                              rs_mcd.end());
2210
0
            }
2211
173k
        }
2212
47.7k
    }
2213
2214
47.7k
    req.runtime_state->get_unreported_errors(&(params.error_log));
2215
47.7k
    params.__isset.error_log = (!params.error_log.empty());
2216
2217
47.7k
    if (_exec_env->cluster_info()->backend_id != 0) {
2218
47.7k
        params.__set_backend_id(_exec_env->cluster_info()->backend_id);
2219
47.7k
    }
2220
2221
47.7k
    TReportExecStatusResult res;
2222
47.7k
    Status rpc_status;
2223
2224
47.7k
    VLOG_DEBUG << "reportExecStatus params is "
2225
5
               << apache::thrift::ThriftDebugString(params).c_str();
2226
47.7k
    if (!exec_status.ok()) {
2227
1.66k
        LOG(WARNING) << "report error status: " << exec_status.msg()
2228
1.66k
                     << " to coordinator: " << req.coord_addr
2229
1.66k
                     << ", query id: " << print_id(req.query_id);
2230
1.66k
    }
2231
47.7k
    try {
2232
47.7k
        try {
2233
47.7k
            (*coord)->reportExecStatus(res, params);
2234
47.7k
        } catch ([[maybe_unused]] apache::thrift::transport::TTransportException& e) {
2235
#ifndef ADDRESS_SANITIZER
2236
            LOG(WARNING) << "Retrying ReportExecStatus. query id: " << print_id(req.query_id)
2237
                         << ", instance id: " << print_id(req.fragment_instance_id) << " to "
2238
                         << req.coord_addr << ", err: " << e.what();
2239
#endif
2240
0
            rpc_status = coord->reopen();
2241
2242
0
            if (!rpc_status.ok()) {
2243
0
                req.cancel_fn(rpc_status);
2244
0
                return;
2245
0
            }
2246
0
            (*coord)->reportExecStatus(res, params);
2247
0
        }
2248
2249
47.7k
        rpc_status = Status::create<false>(res.status);
2250
47.7k
    } catch (apache::thrift::TException& e) {
2251
0
        rpc_status = Status::InternalError("ReportExecStatus() to {} failed: {}",
2252
0
                                           PrintThriftNetworkAddress(req.coord_addr), e.what());
2253
0
    }
2254
2255
47.7k
    if (!rpc_status.ok()) {
2256
0
        LOG_INFO("Going to cancel query {} since report exec status got rpc failed: {}",
2257
0
                 print_id(req.query_id), rpc_status.to_string());
2258
0
        req.cancel_fn(rpc_status);
2259
0
    }
2260
47.7k
}
2261
2262
434k
Status PipelineFragmentContext::send_report(bool done) {
2263
434k
    Status exec_status = _query_ctx->exec_status();
2264
    // If plan is done successfully, but _is_report_success is false,
2265
    // no need to send report.
2266
    // Load will set _is_report_success to true because load wants to know
2267
    // the process.
2268
434k
    if (!_is_report_success && done && exec_status.ok()) {
2269
386k
        return Status::OK();
2270
386k
    }
2271
2272
    // If both _is_report_success and _is_report_on_cancel are false,
2273
    // which means no matter query is success or failed, no report is needed.
2274
    // This may happen when the query limit reached and
2275
    // a internal cancellation being processed
2276
    // When limit is reached the fragment is also cancelled, but _is_report_on_cancel will
2277
    // be set to false, to avoid sending fault report to FE.
2278
48.1k
    if (!_is_report_success && !_is_report_on_cancel) {
2279
334
        if (done) {
2280
            // if done is true, which means the query is finished successfully, we can safely close the fragment instance without sending report to FE, and just return OK status here.
2281
334
            return Status::OK();
2282
334
        }
2283
0
        return Status::NeedSendAgain("");
2284
334
    }
2285
2286
47.7k
    std::vector<RuntimeState*> runtime_states;
2287
2288
118k
    for (auto& tasks : _tasks) {
2289
173k
        for (auto& task : tasks) {
2290
173k
            runtime_states.push_back(task.second.get());
2291
173k
        }
2292
118k
    }
2293
2294
47.7k
    std::string load_eror_url = _query_ctx->get_load_error_url().empty()
2295
47.7k
                                        ? get_load_error_url()
2296
47.7k
                                        : _query_ctx->get_load_error_url();
2297
47.7k
    std::string first_error_msg = _query_ctx->get_first_error_msg().empty()
2298
47.7k
                                          ? get_first_error_msg()
2299
47.7k
                                          : _query_ctx->get_first_error_msg();
2300
2301
47.7k
    ReportStatusRequest req {.status = exec_status,
2302
47.7k
                             .runtime_states = runtime_states,
2303
47.7k
                             .done = done || !exec_status.ok(),
2304
47.7k
                             .coord_addr = _query_ctx->coord_addr,
2305
47.7k
                             .query_id = _query_id,
2306
47.7k
                             .fragment_id = _fragment_id,
2307
47.7k
                             .fragment_instance_id = TUniqueId(),
2308
47.7k
                             .backend_num = -1,
2309
47.7k
                             .runtime_state = _runtime_state.get(),
2310
47.7k
                             .load_error_url = load_eror_url,
2311
47.7k
                             .first_error_msg = first_error_msg,
2312
47.7k
                             .cancel_fn = [this](const Status& reason) { cancel(reason); }};
2313
47.7k
    auto ctx = std::dynamic_pointer_cast<PipelineFragmentContext>(shared_from_this());
2314
47.7k
    return _exec_env->fragment_mgr()->get_thread_pool()->submit_func([this, req, ctx]() {
2315
47.7k
        SCOPED_ATTACH_TASK(ctx->get_query_ctx()->query_mem_tracker());
2316
47.7k
        _coordinator_callback(req);
2317
47.7k
        if (!req.done) {
2318
4.83k
            ctx->refresh_next_report_time();
2319
4.83k
        }
2320
47.7k
    });
2321
48.1k
}
2322
2323
0
size_t PipelineFragmentContext::get_revocable_size(bool* has_running_task) const {
2324
0
    size_t res = 0;
2325
    // _tasks will be cleared during ~PipelineFragmentContext, so that it's safe
2326
    // here to traverse the vector.
2327
0
    for (const auto& task_instances : _tasks) {
2328
0
        for (const auto& task : task_instances) {
2329
0
            if (task.first->is_running()) {
2330
0
                LOG_EVERY_N(INFO, 50) << "Query: " << print_id(_query_id)
2331
0
                                      << " is running, task: " << (void*)task.first.get()
2332
0
                                      << ", is_running: " << task.first->is_running();
2333
0
                *has_running_task = true;
2334
0
                return 0;
2335
0
            }
2336
2337
0
            size_t revocable_size = task.first->get_revocable_size();
2338
0
            if (revocable_size >= SpillFile::MIN_SPILL_WRITE_BATCH_MEM) {
2339
0
                res += revocable_size;
2340
0
            }
2341
0
        }
2342
0
    }
2343
0
    return res;
2344
0
}
2345
2346
0
std::vector<PipelineTask*> PipelineFragmentContext::get_revocable_tasks() const {
2347
0
    std::vector<PipelineTask*> revocable_tasks;
2348
0
    for (const auto& task_instances : _tasks) {
2349
0
        for (const auto& task : task_instances) {
2350
0
            size_t revocable_size_ = task.first->get_revocable_size();
2351
2352
0
            if (revocable_size_ >= SpillFile::MIN_SPILL_WRITE_BATCH_MEM) {
2353
0
                revocable_tasks.emplace_back(task.first.get());
2354
0
            }
2355
0
        }
2356
0
    }
2357
0
    return revocable_tasks;
2358
0
}
2359
2360
73
std::string PipelineFragmentContext::debug_string() {
2361
73
    std::lock_guard<std::mutex> l(_task_mutex);
2362
73
    fmt::memory_buffer debug_string_buffer;
2363
73
    fmt::format_to(debug_string_buffer,
2364
73
                   "PipelineFragmentContext Info: _closed_tasks={}, _total_tasks={}, "
2365
73
                   "need_notify_close={}, fragment_id={}, _rec_cte_stage={}\n",
2366
73
                   _closed_tasks, _total_tasks, _need_notify_close, _fragment_id, _rec_cte_stage);
2367
515
    for (size_t j = 0; j < _tasks.size(); j++) {
2368
442
        fmt::format_to(debug_string_buffer, "Tasks in instance {}:\n", j);
2369
1.20k
        for (size_t i = 0; i < _tasks[j].size(); i++) {
2370
758
            fmt::format_to(debug_string_buffer, "Task {}: {}\n", i,
2371
758
                           _tasks[j][i].first->debug_string());
2372
758
        }
2373
442
    }
2374
2375
73
    return fmt::to_string(debug_string_buffer);
2376
73
}
2377
2378
std::vector<std::shared_ptr<TRuntimeProfileTree>>
2379
2.23k
PipelineFragmentContext::collect_realtime_profile() const {
2380
2.23k
    std::vector<std::shared_ptr<TRuntimeProfileTree>> res;
2381
2382
    // we do not have mutex to protect pipeline_id_to_profile
2383
    // so we need to make sure this funciton is invoked after fragment context
2384
    // has already been prepared.
2385
2.23k
    if (!_prepared) {
2386
0
        std::string msg =
2387
0
                "Query " + print_id(_query_id) + " collecting profile, but its not prepared";
2388
0
        DCHECK(false) << msg;
2389
0
        LOG_ERROR(msg);
2390
0
        return res;
2391
0
    }
2392
2393
    // Make sure first profile is fragment level profile
2394
2.23k
    auto fragment_profile = std::make_shared<TRuntimeProfileTree>();
2395
2.23k
    _fragment_level_profile->to_thrift(fragment_profile.get(), _runtime_state->profile_level());
2396
2.23k
    res.push_back(fragment_profile);
2397
2398
    // pipeline_id_to_profile is initialized in prepare stage
2399
4.15k
    for (auto pipeline_profile : _runtime_state->pipeline_id_to_profile()) {
2400
4.15k
        auto profile_ptr = std::make_shared<TRuntimeProfileTree>();
2401
4.15k
        pipeline_profile->to_thrift(profile_ptr.get(), _runtime_state->profile_level());
2402
4.15k
        res.push_back(profile_ptr);
2403
4.15k
    }
2404
2405
2.23k
    return res;
2406
2.23k
}
2407
2408
std::shared_ptr<TRuntimeProfileTree>
2409
2.23k
PipelineFragmentContext::collect_realtime_load_channel_profile() const {
2410
    // we do not have mutex to protect pipeline_id_to_profile
2411
    // so we need to make sure this funciton is invoked after fragment context
2412
    // has already been prepared.
2413
2.23k
    if (!_prepared) {
2414
0
        std::string msg =
2415
0
                "Query " + print_id(_query_id) + " collecting profile, but its not prepared";
2416
0
        DCHECK(false) << msg;
2417
0
        LOG_ERROR(msg);
2418
0
        return nullptr;
2419
0
    }
2420
2421
5.80k
    for (const auto& tasks : _tasks) {
2422
11.7k
        for (const auto& task : tasks) {
2423
11.7k
            if (task.second->load_channel_profile() == nullptr) {
2424
0
                continue;
2425
0
            }
2426
2427
11.7k
            auto tmp_load_channel_profile = std::make_shared<TRuntimeProfileTree>();
2428
2429
11.7k
            task.second->load_channel_profile()->to_thrift(tmp_load_channel_profile.get(),
2430
11.7k
                                                           _runtime_state->profile_level());
2431
11.7k
            _runtime_state->load_channel_profile()->update(*tmp_load_channel_profile);
2432
11.7k
        }
2433
5.80k
    }
2434
2435
2.23k
    auto load_channel_profile = std::make_shared<TRuntimeProfileTree>();
2436
2.23k
    _runtime_state->load_channel_profile()->to_thrift(load_channel_profile.get(),
2437
2.23k
                                                      _runtime_state->profile_level());
2438
2.23k
    return load_channel_profile;
2439
2.23k
}
2440
2441
// Collect runtime filter IDs registered by all tasks in this PFC.
2442
// Used during recursive CTE stage transitions to know which filters to deregister
2443
// before creating the new PFC for the next recursion round.
2444
// Called from rerun_fragment(wait_for_destroy) while tasks are still closing.
2445
// Thread safety: safe because _tasks is structurally immutable after prepare() —
2446
// the vector sizes do not change, and individual RuntimeState filter sets are
2447
// written only during open() which has completed by the time we reach rerun.
2448
3.28k
std::set<int> PipelineFragmentContext::get_deregister_runtime_filter() const {
2449
3.28k
    std::set<int> result;
2450
6.99k
    for (const auto& _task : _tasks) {
2451
12.2k
        for (const auto& task : _task) {
2452
12.2k
            auto set = task.first->runtime_state()->get_deregister_runtime_filter();
2453
12.2k
            result.merge(set);
2454
12.2k
        }
2455
6.99k
    }
2456
3.28k
    if (_runtime_state) {
2457
3.28k
        auto set = _runtime_state->get_deregister_runtime_filter();
2458
3.28k
        result.merge(set);
2459
3.28k
    }
2460
3.28k
    return result;
2461
3.28k
}
2462
2463
434k
void PipelineFragmentContext::_release_resource() {
2464
434k
    std::lock_guard<std::mutex> l(_task_mutex);
2465
    // The memory released by the query end is recorded in the query mem tracker.
2466
434k
    SCOPED_SWITCH_THREAD_MEM_TRACKER_LIMITER(_query_ctx->query_mem_tracker());
2467
434k
    auto st = _query_ctx->exec_status();
2468
1.21M
    for (auto& _task : _tasks) {
2469
1.21M
        if (!_task.empty()) {
2470
1.21M
            _call_back(_task.front().first->runtime_state(), &st);
2471
1.21M
        }
2472
1.21M
    }
2473
434k
    _tasks.clear();
2474
434k
    _dag.clear();
2475
434k
    _pip_id_to_pipeline.clear();
2476
434k
    _pipelines.clear();
2477
434k
    _sink.reset();
2478
434k
    _root_op.reset();
2479
434k
    _runtime_filter_mgr_map.clear();
2480
434k
    _op_id_to_shared_state.clear();
2481
434k
}
2482
2483
} // namespace doris