Coverage Report

Created: 2026-06-24 17:16

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
392k
        : _query_id(std::move(query_id)),
144
392k
          _fragment_id(request.fragment_id),
145
392k
          _exec_env(exec_env),
146
392k
          _query_ctx(std::move(query_ctx)),
147
392k
          _call_back(call_back),
148
392k
          _is_report_on_cancel(true),
149
392k
          _params(request),
150
392k
          _parallel_instances(_params.__isset.parallel_instances ? _params.parallel_instances : 0),
151
392k
          _need_notify_close(request.__isset.need_notify_close ? request.need_notify_close
152
392k
                                                               : false) {
153
392k
    _fragment_watcher.start();
154
392k
}
155
156
392k
PipelineFragmentContext::~PipelineFragmentContext() {
157
392k
    LOG_INFO("PipelineFragmentContext::~PipelineFragmentContext")
158
392k
            .tag("query_id", print_id(_query_id))
159
392k
            .tag("fragment_id", _fragment_id);
160
392k
    _release_resource();
161
392k
    {
162
        // The memory released by the query end is recorded in the query mem tracker.
163
392k
        SCOPED_SWITCH_THREAD_MEM_TRACKER_LIMITER(_query_ctx->query_mem_tracker());
164
392k
        _runtime_state.reset();
165
392k
        _query_ctx.reset();
166
392k
    }
167
392k
}
168
169
71
bool PipelineFragmentContext::is_timeout(timespec now) const {
170
71
    if (_timeout <= 0) {
171
0
        return false;
172
0
    }
173
71
    return _fragment_watcher.elapsed_time_seconds(now) > _timeout;
174
71
}
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
9.89k
bool PipelineFragmentContext::notify_close() {
181
9.89k
    bool all_closed = false;
182
9.89k
    bool need_remove = false;
183
9.89k
    {
184
9.89k
        std::lock_guard<std::mutex> l(_task_mutex);
185
9.89k
        if (_closed_tasks >= _total_tasks) {
186
3.70k
            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
3.64k
                need_remove = true;
193
3.64k
            }
194
3.70k
            all_closed = true;
195
3.70k
        }
196
        // make fragment release by self after cancel
197
9.89k
        _need_notify_close = false;
198
9.89k
    }
199
9.89k
    if (need_remove) {
200
3.64k
        _exec_env->fragment_mgr()->remove_pipeline_context({_query_id, _fragment_id});
201
3.64k
    }
202
9.89k
    return all_closed;
203
9.89k
}
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
73
        return;
216
73
    }
217
    // Timeout is a special error code, we need print current stack to debug timeout issue.
218
6.10k
    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
6.10k
    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
6.10k
    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
6.10k
    if (auto error_url = get_load_error_url(); !error_url.empty()) {
235
23
        _query_ctx->set_load_error_url(error_url);
236
23
    }
237
238
6.10k
    if (auto first_error_msg = get_first_error_msg(); !first_error_msg.empty()) {
239
23
        _query_ctx->set_first_error_msg(first_error_msg);
240
23
    }
241
242
6.10k
    _query_ctx->cancel(reason, _fragment_id);
243
6.10k
    if (reason.is<ErrorCode::LIMIT_REACH>()) {
244
222
        _is_report_on_cancel = false;
245
5.88k
    } else {
246
34.7k
        for (auto& id : _fragment_instance_ids) {
247
34.7k
            LOG(WARNING) << "PipelineFragmentContext cancel instance: " << print_id(id);
248
34.7k
        }
249
5.88k
    }
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
6.10k
    auto stream_load_ctx = _exec_env->new_load_stream_mgr()->get(_query_id);
253
6.10k
    if (stream_load_ctx != nullptr) {
254
31
        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
31
        stream_load_ctx->error_url = get_load_error_url();
259
31
        stream_load_ctx->first_error_msg = get_first_error_msg();
260
31
    }
261
262
35.3k
    for (auto& tasks : _tasks) {
263
77.5k
        for (auto& task : tasks) {
264
77.5k
            task.first->unblock_all_dependencies();
265
77.5k
        }
266
35.3k
    }
267
6.10k
}
268
269
607k
PipelinePtr PipelineFragmentContext::add_pipeline(PipelinePtr parent, int idx) {
270
607k
    PipelineId id = _next_pipeline_id++;
271
607k
    auto pipeline = std::make_shared<Pipeline>(
272
607k
            id, parent ? std::min(parent->num_tasks(), _num_instances) : _num_instances,
273
607k
            parent ? parent->num_tasks() : _num_instances);
274
607k
    if (idx >= 0) {
275
756
        _pipelines.insert(_pipelines.begin() + idx, pipeline);
276
606k
    } else {
277
606k
        _pipelines.emplace_back(pipeline);
278
606k
    }
279
607k
    if (parent) {
280
210k
        parent->set_children(pipeline);
281
210k
    }
282
607k
    return pipeline;
283
607k
}
284
285
392k
Status PipelineFragmentContext::_build_and_prepare_full_pipeline(ThreadPool* thread_pool) {
286
392k
    {
287
392k
        SCOPED_TIMER(_build_pipelines_timer);
288
        // 2. Build pipelines with operators in this fragment.
289
392k
        auto root_pipeline = add_pipeline();
290
392k
        RETURN_IF_ERROR(_build_pipelines(_runtime_state->obj_pool(), *_query_ctx->desc_tbl,
291
392k
                                         &_root_op, root_pipeline));
292
293
        // Propagate _num_instances from LOCAL_EXCHANGE pipelines to ancestor pipelines
294
        // that inherited reduced num_tasks from a serial operator.
295
392k
        _propagate_local_exchange_num_tasks();
296
297
        // Create deferred local exchangers now that all pipelines have final num_tasks.
298
392k
        RETURN_IF_ERROR(_create_deferred_local_exchangers());
299
300
        // Raise num_tasks for pipelines whose serial non-scan operators (e.g.,
301
        // UNPARTITIONED Exchange) reduced num_tasks below _num_instances.
302
        // Without this, fragment instances 1+ have no task for these pipelines
303
        // and downstream operators fail with "must set shared state".
304
        //
305
        // This applies to ALL pipelines (not just deferred exchanger upstreams):
306
        // fragments with UNION/INTERSECT/EXCEPT + serial Exchange in child
307
        // pipelines also need the raise, even without FE-planned local exchange.
308
        //
309
        // Exception: serial scan sources (pooling scan) keep num_tasks=1 — the
310
        // PassthroughExchanger(1, N) handles the fan-out correctly.
311
        // NOTE: Do NOT raise pipelines whose source is a serial operator
312
        // (Exchange or scan) — they legitimately have 1 task, and raising
313
        // them causes crashes (e.g., 4 Exchange tasks but only 1 receives
314
        // data).  The correct fix for shared state injection across
315
        // instances is handled by the FE: it inserts local exchange nodes
316
        // between serial operators and their downstream consumers, creating
317
        // proper pipeline boundaries with _num_instances tasks.
318
319
        // 3. Create sink operator
320
392k
        if (!_params.fragment.__isset.output_sink) {
321
0
            return Status::InternalError("No output sink in this fragment!");
322
0
        }
323
392k
        RETURN_IF_ERROR(_create_data_sink(_runtime_state->obj_pool(), _params.fragment.output_sink,
324
392k
                                          _params.fragment.output_exprs, _params,
325
392k
                                          root_pipeline->output_row_desc(), _runtime_state.get(),
326
392k
                                          *_desc_tbl, root_pipeline->id()));
327
392k
        RETURN_IF_ERROR(_sink->init(_params.fragment.output_sink));
328
392k
        RETURN_IF_ERROR(root_pipeline->set_sink(_sink));
329
330
606k
        for (PipelinePtr& pipeline : _pipelines) {
331
606k
            DCHECK(pipeline->sink() != nullptr) << pipeline->operators().size();
332
606k
            RETURN_IF_ERROR(pipeline->sink()->set_child(pipeline->operators().back()));
333
606k
        }
334
392k
    }
335
    // 4. Build local exchanger
336
392k
    if (_runtime_state->plan_local_shuffle()) {
337
116k
        SCOPED_TIMER(_plan_local_exchanger_timer);
338
116k
        RETURN_IF_ERROR(_plan_local_exchange(_params.num_buckets,
339
116k
                                             _params.bucket_seq_to_instance_idx,
340
116k
                                             _params.shuffle_idx_to_instance_idx));
341
116k
    }
342
343
    // 5. Initialize global states in pipelines.
344
608k
    for (PipelinePtr& pipeline : _pipelines) {
345
608k
        SCOPED_TIMER(_prepare_all_pipelines_timer);
346
608k
        pipeline->children().clear();
347
608k
        RETURN_IF_ERROR(pipeline->prepare(_runtime_state.get()));
348
608k
    }
349
350
391k
    {
351
391k
        SCOPED_TIMER(_build_tasks_timer);
352
        // 6. Build pipeline tasks and initialize local state.
353
391k
        RETURN_IF_ERROR(_build_pipeline_tasks(thread_pool));
354
391k
    }
355
356
391k
    return Status::OK();
357
391k
}
358
359
392k
Status PipelineFragmentContext::prepare(ThreadPool* thread_pool) {
360
392k
    if (_prepared) {
361
0
        return Status::InternalError("Already prepared");
362
0
    }
363
392k
    if (_params.__isset.query_options && _params.query_options.__isset.execution_timeout) {
364
392k
        _timeout = _params.query_options.execution_timeout;
365
392k
    }
366
367
392k
    _fragment_level_profile = std::make_unique<RuntimeProfile>("PipelineContext");
368
392k
    _prepare_timer = ADD_TIMER(_fragment_level_profile, "PrepareTime");
369
392k
    SCOPED_TIMER(_prepare_timer);
370
392k
    _build_pipelines_timer = ADD_TIMER(_fragment_level_profile, "BuildPipelinesTime");
371
392k
    _init_context_timer = ADD_TIMER(_fragment_level_profile, "InitContextTime");
372
392k
    _plan_local_exchanger_timer = ADD_TIMER(_fragment_level_profile, "PlanLocalLocalExchangerTime");
373
392k
    _build_tasks_timer = ADD_TIMER(_fragment_level_profile, "BuildTasksTime");
374
392k
    _prepare_all_pipelines_timer = ADD_TIMER(_fragment_level_profile, "PrepareAllPipelinesTime");
375
392k
    {
376
392k
        SCOPED_TIMER(_init_context_timer);
377
392k
        cast_set(_num_instances, _params.local_params.size());
378
392k
        _total_instances =
379
392k
                _params.__isset.total_instances ? _params.total_instances : _num_instances;
380
381
392k
        auto* fragment_context = this;
382
383
392k
        if (_params.query_options.__isset.is_report_success) {
384
390k
            fragment_context->set_is_report_success(_params.query_options.is_report_success);
385
390k
        }
386
387
        // 1. Set up the global runtime state.
388
392k
        _runtime_state = RuntimeState::create_unique(
389
392k
                _params.query_id, _params.fragment_id, _params.query_options,
390
392k
                _query_ctx->query_globals, _exec_env, _query_ctx.get());
391
392k
        _runtime_state->set_task_execution_context(shared_from_this());
392
392k
        SCOPED_SWITCH_THREAD_MEM_TRACKER_LIMITER(_runtime_state->query_mem_tracker());
393
392k
        if (_params.__isset.backend_id) {
394
388k
            _runtime_state->set_backend_id(_params.backend_id);
395
388k
        }
396
392k
        if (_params.__isset.import_label) {
397
240
            _runtime_state->set_import_label(_params.import_label);
398
240
        }
399
392k
        if (_params.__isset.db_name) {
400
192
            _runtime_state->set_db_name(_params.db_name);
401
192
        }
402
392k
        if (_params.__isset.load_job_id) {
403
0
            _runtime_state->set_load_job_id(_params.load_job_id);
404
0
        }
405
406
392k
        if (_params.is_simplified_param) {
407
135k
            _desc_tbl = _query_ctx->desc_tbl;
408
257k
        } else {
409
257k
            DCHECK(_params.__isset.desc_tbl);
410
257k
            RETURN_IF_ERROR(DescriptorTbl::create(_runtime_state->obj_pool(), _params.desc_tbl,
411
257k
                                                  &_desc_tbl));
412
257k
        }
413
392k
        _runtime_state->set_desc_tbl(_desc_tbl);
414
392k
        _runtime_state->set_num_per_fragment_instances(_params.num_senders);
415
392k
        _runtime_state->set_load_stream_per_node(_params.load_stream_per_node);
416
392k
        _runtime_state->set_total_load_streams(_params.total_load_streams);
417
392k
        _runtime_state->set_num_local_sink(_params.num_local_sink);
418
419
        // init fragment_instance_ids
420
392k
        const auto target_size = _params.local_params.size();
421
392k
        _fragment_instance_ids.resize(target_size);
422
1.53M
        for (size_t i = 0; i < _params.local_params.size(); i++) {
423
1.14M
            auto fragment_instance_id = _params.local_params[i].fragment_instance_id;
424
1.14M
            _fragment_instance_ids[i] = fragment_instance_id;
425
1.14M
        }
426
392k
    }
427
428
392k
    RETURN_IF_ERROR(_build_and_prepare_full_pipeline(thread_pool));
429
430
391k
    _init_next_report_time();
431
432
391k
    _prepared = true;
433
391k
    return Status::OK();
434
392k
}
435
436
Status PipelineFragmentContext::_build_pipeline_tasks_for_instance(
437
        int instance_idx,
438
1.14M
        const std::vector<std::shared_ptr<RuntimeProfile>>& pipeline_id_to_profile) {
439
1.14M
    const auto& local_params = _params.local_params[instance_idx];
440
1.14M
    auto fragment_instance_id = local_params.fragment_instance_id;
441
1.14M
    auto runtime_filter_mgr = std::make_unique<RuntimeFilterMgr>(false);
442
1.14M
    std::map<PipelineId, PipelineTask*> pipeline_id_to_task;
443
1.14M
    auto get_shared_state = [&](PipelinePtr pipeline)
444
1.14M
            -> std::map<int, std::pair<std::shared_ptr<BasicSharedState>,
445
1.91M
                                       std::vector<std::shared_ptr<Dependency>>>> {
446
1.91M
        std::map<int, std::pair<std::shared_ptr<BasicSharedState>,
447
1.91M
                                std::vector<std::shared_ptr<Dependency>>>>
448
1.91M
                shared_state_map;
449
2.51M
        for (auto& op : pipeline->operators()) {
450
2.51M
            auto source_id = op->operator_id();
451
2.51M
            if (auto iter = _op_id_to_shared_state.find(source_id);
452
2.51M
                iter != _op_id_to_shared_state.end()) {
453
832k
                shared_state_map.insert({source_id, iter->second});
454
832k
            }
455
2.51M
        }
456
1.91M
        for (auto sink_to_source_id : pipeline->sink()->dests_id()) {
457
1.91M
            if (auto iter = _op_id_to_shared_state.find(sink_to_source_id);
458
1.91M
                iter != _op_id_to_shared_state.end()) {
459
361k
                shared_state_map.insert({sink_to_source_id, iter->second});
460
361k
            }
461
1.91M
        }
462
1.91M
        return shared_state_map;
463
1.91M
    };
464
465
3.51M
    for (size_t pip_idx = 0; pip_idx < _pipelines.size(); pip_idx++) {
466
2.37M
        auto& pipeline = _pipelines[pip_idx];
467
2.37M
        if (pipeline->num_tasks() > 1 || instance_idx == 0) {
468
1.90M
            auto task_runtime_state = RuntimeState::create_unique(
469
1.90M
                    local_params.fragment_instance_id, _params.query_id, _params.fragment_id,
470
1.90M
                    _params.query_options, _query_ctx->query_globals, _exec_env, _query_ctx.get());
471
1.90M
            {
472
                // Initialize runtime state for this task
473
1.90M
                task_runtime_state->set_query_mem_tracker(_query_ctx->query_mem_tracker());
474
475
1.90M
                task_runtime_state->set_task_execution_context(shared_from_this());
476
1.90M
                task_runtime_state->set_be_number(local_params.backend_num);
477
478
1.91M
                if (_params.__isset.backend_id) {
479
1.91M
                    task_runtime_state->set_backend_id(_params.backend_id);
480
1.91M
                }
481
1.90M
                if (_params.__isset.import_label) {
482
241
                    task_runtime_state->set_import_label(_params.import_label);
483
241
                }
484
1.90M
                if (_params.__isset.db_name) {
485
193
                    task_runtime_state->set_db_name(_params.db_name);
486
193
                }
487
1.90M
                if (_params.__isset.load_job_id) {
488
0
                    task_runtime_state->set_load_job_id(_params.load_job_id);
489
0
                }
490
1.90M
                if (_params.__isset.wal_id) {
491
115
                    task_runtime_state->set_wal_id(_params.wal_id);
492
115
                }
493
1.90M
                if (_params.__isset.content_length) {
494
34
                    task_runtime_state->set_content_length(_params.content_length);
495
34
                }
496
497
1.90M
                task_runtime_state->set_desc_tbl(_desc_tbl);
498
1.90M
                task_runtime_state->set_per_fragment_instance_idx(local_params.sender_id);
499
1.90M
                task_runtime_state->set_num_per_fragment_instances(_params.num_senders);
500
1.90M
                task_runtime_state->resize_op_id_to_local_state(max_operator_id());
501
1.90M
                task_runtime_state->set_max_operator_id(max_operator_id());
502
1.90M
                task_runtime_state->set_load_stream_per_node(_params.load_stream_per_node);
503
1.90M
                task_runtime_state->set_total_load_streams(_params.total_load_streams);
504
1.90M
                task_runtime_state->set_num_local_sink(_params.num_local_sink);
505
506
1.90M
                task_runtime_state->set_runtime_filter_mgr(runtime_filter_mgr.get());
507
1.90M
            }
508
1.90M
            auto cur_task_id = _total_tasks++;
509
1.90M
            task_runtime_state->set_task_id(cur_task_id);
510
1.90M
            task_runtime_state->set_task_num(pipeline->num_tasks());
511
1.90M
            auto task = std::make_shared<PipelineTask>(
512
1.90M
                    pipeline, cur_task_id, task_runtime_state.get(),
513
1.90M
                    std::dynamic_pointer_cast<PipelineFragmentContext>(shared_from_this()),
514
1.90M
                    pipeline_id_to_profile[pip_idx].get(), get_shared_state(pipeline),
515
1.90M
                    instance_idx);
516
1.90M
            pipeline->incr_created_tasks(instance_idx, task.get());
517
1.90M
            pipeline_id_to_task.insert({pipeline->id(), task.get()});
518
1.90M
            _tasks[instance_idx].emplace_back(
519
1.90M
                    std::pair<std::shared_ptr<PipelineTask>, std::unique_ptr<RuntimeState>> {
520
1.90M
                            std::move(task), std::move(task_runtime_state)});
521
1.90M
        }
522
2.37M
    }
523
524
    /**
525
         * Build DAG for pipeline tasks.
526
         * For example, we have
527
         *
528
         *   ExchangeSink (Pipeline1)     JoinBuildSink (Pipeline2)
529
         *            \                      /
530
         *          JoinProbeOperator1 (Pipeline1)    JoinBuildSink (Pipeline3)
531
         *                 \                          /
532
         *               JoinProbeOperator2 (Pipeline1)
533
         *
534
         * In this fragment, we have three pipelines and pipeline 1 depends on pipeline 2 and pipeline 3.
535
         * To build this DAG, `_dag` manage dependencies between pipelines by pipeline ID and
536
         * `pipeline_id_to_task` is used to find the task by a unique pipeline ID.
537
         *
538
         * Finally, we have two upstream dependencies in Pipeline1 corresponding to JoinProbeOperator1
539
         * and JoinProbeOperator2.
540
         */
541
2.37M
    for (auto& _pipeline : _pipelines) {
542
2.37M
        if (pipeline_id_to_task.contains(_pipeline->id())) {
543
1.90M
            auto* task = pipeline_id_to_task[_pipeline->id()];
544
1.90M
            DCHECK(task != nullptr);
545
546
            // If this task has upstream dependency, then inject it into this task.
547
1.90M
            if (_dag.contains(_pipeline->id())) {
548
1.23M
                auto& deps = _dag[_pipeline->id()];
549
1.24M
                for (auto& dep : deps) {
550
1.24M
                    if (pipeline_id_to_task.contains(dep)) {
551
769k
                        auto ss = pipeline_id_to_task[dep]->get_sink_shared_state();
552
769k
                        if (ss) {
553
395k
                            task->inject_shared_state(ss);
554
395k
                        } else {
555
374k
                            pipeline_id_to_task[dep]->inject_shared_state(
556
374k
                                    task->get_source_shared_state());
557
374k
                        }
558
769k
                    }
559
1.24M
                }
560
1.23M
            }
561
1.90M
        }
562
2.37M
    }
563
3.52M
    for (size_t pip_idx = 0; pip_idx < _pipelines.size(); pip_idx++) {
564
2.37M
        if (pipeline_id_to_task.contains(_pipelines[pip_idx]->id())) {
565
1.90M
            auto* task = pipeline_id_to_task[_pipelines[pip_idx]->id()];
566
1.90M
            DCHECK(pipeline_id_to_profile[pip_idx]);
567
1.90M
            std::vector<TScanRangeParams> scan_ranges;
568
1.90M
            auto node_id = _pipelines[pip_idx]->operators().front()->node_id();
569
1.90M
            if (local_params.per_node_scan_ranges.contains(node_id)) {
570
291k
                scan_ranges = local_params.per_node_scan_ranges.find(node_id)->second;
571
291k
            }
572
1.90M
            RETURN_IF_ERROR_OR_CATCH_EXCEPTION(task->prepare(scan_ranges, local_params.sender_id,
573
1.90M
                                                             _params.fragment.output_sink));
574
1.90M
        }
575
2.37M
    }
576
1.14M
    {
577
1.14M
        std::lock_guard<std::mutex> l(_state_map_lock);
578
1.14M
        _runtime_filter_mgr_map[instance_idx] = std::move(runtime_filter_mgr);
579
1.14M
    }
580
1.14M
    return Status::OK();
581
1.14M
}
582
583
391k
Status PipelineFragmentContext::_build_pipeline_tasks(ThreadPool* thread_pool) {
584
391k
    _total_tasks = 0;
585
391k
    _closed_tasks = 0;
586
391k
    const auto target_size = _params.local_params.size();
587
391k
    _tasks.resize(target_size);
588
391k
    _runtime_filter_mgr_map.resize(target_size);
589
998k
    for (size_t pip_idx = 0; pip_idx < _pipelines.size(); pip_idx++) {
590
607k
        _pip_id_to_pipeline[_pipelines[pip_idx]->id()] = _pipelines[pip_idx].get();
591
607k
    }
592
391k
    auto pipeline_id_to_profile = _runtime_state->build_pipeline_profile(_pipelines.size());
593
594
391k
    if (target_size > 1 &&
595
391k
        (_runtime_state->query_options().__isset.parallel_prepare_threshold &&
596
127k
         target_size > _runtime_state->query_options().parallel_prepare_threshold)) {
597
        // If instances parallelism is big enough ( > parallel_prepare_threshold), we will prepare all tasks by multi-threads
598
21.6k
        std::vector<Status> prepare_status(target_size);
599
21.6k
        int submitted_tasks = 0;
600
21.6k
        Status submit_status;
601
21.6k
        CountDownLatch latch((int)target_size);
602
206k
        for (int i = 0; i < target_size; i++) {
603
184k
            submit_status = thread_pool->submit_func([&, i]() {
604
184k
                SCOPED_ATTACH_TASK(_query_ctx.get());
605
184k
                prepare_status[i] = _build_pipeline_tasks_for_instance(i, pipeline_id_to_profile);
606
184k
                latch.count_down();
607
184k
            });
608
184k
            if (LIKELY(submit_status.ok())) {
609
184k
                submitted_tasks++;
610
18.4E
            } else {
611
18.4E
                break;
612
18.4E
            }
613
184k
        }
614
21.6k
        latch.arrive_and_wait(target_size - submitted_tasks);
615
21.6k
        if (UNLIKELY(!submit_status.ok())) {
616
0
            return submit_status;
617
0
        }
618
206k
        for (int i = 0; i < submitted_tasks; i++) {
619
184k
            if (!prepare_status[i].ok()) {
620
0
                return prepare_status[i];
621
0
            }
622
184k
        }
623
370k
    } else {
624
1.33M
        for (int i = 0; i < target_size; i++) {
625
961k
            RETURN_IF_ERROR(_build_pipeline_tasks_for_instance(i, pipeline_id_to_profile));
626
961k
        }
627
370k
    }
628
391k
    _pipeline_parent_map.clear();
629
391k
    _op_id_to_shared_state.clear();
630
    // Record task cardinality once when this fragment context finishes task initialization.
631
391k
    _query_ctx->add_total_task_num(_total_tasks.load(std::memory_order_relaxed));
632
633
391k
    return Status::OK();
634
391k
}
635
636
390k
void PipelineFragmentContext::_init_next_report_time() {
637
390k
    auto interval_s = config::pipeline_status_report_interval;
638
390k
    if (_is_report_success && interval_s > 0 && _timeout > interval_s) {
639
38.0k
        VLOG_FILE << "enable period report: fragment id=" << _fragment_id;
640
38.0k
        uint64_t report_fragment_offset = (uint64_t)(rand() % interval_s) * NANOS_PER_SEC;
641
        // We don't want to wait longer than it takes to run the entire fragment.
642
38.0k
        _previous_report_time =
643
38.0k
                MonotonicNanos() + report_fragment_offset - (uint64_t)(interval_s)*NANOS_PER_SEC;
644
38.0k
        _disable_period_report = false;
645
38.0k
    }
646
390k
}
647
648
4.54k
void PipelineFragmentContext::refresh_next_report_time() {
649
4.54k
    auto disable = _disable_period_report.load(std::memory_order_acquire);
650
4.54k
    DCHECK(disable == true);
651
4.54k
    _previous_report_time.store(MonotonicNanos(), std::memory_order_release);
652
4.54k
    _disable_period_report.compare_exchange_strong(disable, false);
653
4.54k
}
654
655
6.82M
void PipelineFragmentContext::trigger_report_if_necessary() {
656
6.82M
    if (!_is_report_success) {
657
6.41M
        return;
658
6.41M
    }
659
407k
    auto disable = _disable_period_report.load(std::memory_order_acquire);
660
407k
    if (disable) {
661
7.46k
        return;
662
7.46k
    }
663
400k
    int32_t interval_s = config::pipeline_status_report_interval;
664
400k
    if (interval_s <= 0) {
665
0
        LOG(WARNING) << "config::status_report_interval is equal to or less than zero, do not "
666
0
                        "trigger "
667
0
                        "report.";
668
0
    }
669
400k
    uint64_t next_report_time = _previous_report_time.load(std::memory_order_acquire) +
670
400k
                                (uint64_t)(interval_s)*NANOS_PER_SEC;
671
400k
    if (MonotonicNanos() > next_report_time) {
672
4.55k
        if (!_disable_period_report.compare_exchange_strong(disable, true,
673
4.55k
                                                            std::memory_order_acq_rel)) {
674
13
            return;
675
13
        }
676
4.54k
        if (VLOG_FILE_IS_ON) {
677
0
            VLOG_FILE << "Reporting "
678
0
                      << "profile for query_id " << print_id(_query_id)
679
0
                      << ", fragment id: " << _fragment_id;
680
681
0
            std::stringstream ss;
682
0
            _runtime_state->runtime_profile()->compute_time_in_profile();
683
0
            _runtime_state->runtime_profile()->pretty_print(&ss);
684
0
            if (_runtime_state->load_channel_profile()) {
685
0
                _runtime_state->load_channel_profile()->pretty_print(&ss);
686
0
            }
687
688
0
            VLOG_FILE << "Query " << print_id(get_query_id()) << " fragment " << get_fragment_id()
689
0
                      << " profile:\n"
690
0
                      << ss.str();
691
0
        }
692
4.54k
        auto st = send_report(false);
693
4.54k
        if (!st.ok()) {
694
0
            disable = true;
695
0
            _disable_period_report.compare_exchange_strong(disable, false,
696
0
                                                           std::memory_order_acq_rel);
697
0
        }
698
4.54k
    }
699
400k
}
700
701
Status PipelineFragmentContext::_build_pipelines(ObjectPool* pool, const DescriptorTbl& descs,
702
388k
                                                 OperatorPtr* root, PipelinePtr cur_pipe) {
703
388k
    if (_params.fragment.plan.nodes.empty()) {
704
0
        throw Exception(ErrorCode::INTERNAL_ERROR, "Invalid plan which has no plan node!");
705
0
    }
706
707
388k
    int node_idx = 0;
708
709
388k
    RETURN_IF_ERROR(_create_tree_helper(pool, _params.fragment.plan.nodes, descs, nullptr,
710
388k
                                        &node_idx, root, cur_pipe, 0, false, false));
711
712
388k
    if (node_idx + 1 != _params.fragment.plan.nodes.size()) {
713
0
        return Status::InternalError(
714
0
                "Plan tree only partially reconstructed. Not all thrift nodes were used.");
715
0
    }
716
388k
    return Status::OK();
717
388k
}
718
719
391k
Status PipelineFragmentContext::_create_deferred_local_exchangers() {
720
391k
    for (auto& info : _deferred_exchangers) {
721
        // DANGER ZONE — do not "fix" this line without reading the history.
722
        //
723
        // sender_count seeds Exchanger::_running_sink_operators, which the source side
724
        // waits to reach 0 via sub_running_sink_operators on each sink LocalState close.
725
        // The correct value is THIS pipeline-instance's sink task count, which is exactly
726
        // info.upstream_pipe->num_tasks() — one PipelineTask per task, one close per task.
727
        //
728
        // Tempting wrong fix #1: `std::max(num_tasks, _num_instances)` to mirror the
729
        //   BE-planned path in _add_local_exchange_impl (~line 1023).  THIS BREAKS the
730
        //   common FE-planned shape of `serial scan → LE(PT) → ...`: upstream_pipe
731
        //   genuinely has num_tasks=1, only 1 close arrives, but seed becomes
732
        //   _num_instances so _running_sink_operators never reaches 0 — downstream
733
        //   sources hang on SHUFFLE_DATA_DEPENDENCY (e.g. MTMV refresh from
734
        //   mtmv_up_down_job_p0/load.groovy stays at Status=RUNNING and regressed
735
        //   exactly this way).  BE-planned mode uses max() because its
736
        //   `cur_pipe` is the source-side pipeline (always raised to _num_instances by
737
        //   add_pipeline) — not analogous to our `upstream_pipe` here, which is the
738
        //   sink-side pipeline that may legitimately stay at 1 for serial sources.
739
        //
740
        // Tempting wrong fix #2: multiply by _num_instances on the theory shared_state
741
        //   is shared across all instances.  Same hang — each fragment-instance
742
        //   PipelineFragmentContext has its OWN _op_id_to_shared_state map, so the
743
        //   exchanger is per-instance, not per-BE.  num_tasks() is already the right
744
        //   close-count for one instance.
745
        //
746
        // If a hang shows up with `_running_sink_operators < 0`, the bug is upstream:
747
        // _propagate_local_exchange_num_tasks left num_tasks too low (or too high) for
748
        // this fragment shape.  Fix THAT pass, not this seed value.
749
109k
        const int sender_count = info.upstream_pipe->num_tasks();
750
109k
        switch (info.partition_type) {
751
22.6k
        case TLocalPartitionType::LOCAL_EXECUTION_HASH_SHUFFLE:
752
22.6k
        case TLocalPartitionType::GLOBAL_EXECUTION_HASH_SHUFFLE:
753
22.6k
            info.shared_state->exchanger = ShuffleExchanger::create_unique(
754
22.6k
                    sender_count, _num_instances, info.num_partitions, info.free_blocks_limit,
755
22.6k
                    info.partition_type);
756
22.6k
            break;
757
523
        case TLocalPartitionType::BUCKET_HASH_SHUFFLE:
758
523
            info.shared_state->exchanger = BucketShuffleExchanger::create_unique(
759
523
                    sender_count, _num_instances, info.num_partitions, info.free_blocks_limit);
760
523
            break;
761
81.8k
        case TLocalPartitionType::PASSTHROUGH:
762
81.8k
            info.shared_state->exchanger = PassthroughExchanger::create_unique(
763
81.8k
                    sender_count, _num_instances, info.free_blocks_limit);
764
81.8k
            break;
765
505
        case TLocalPartitionType::BROADCAST:
766
505
            info.shared_state->exchanger = BroadcastExchanger::create_unique(
767
505
                    sender_count, _num_instances, info.free_blocks_limit);
768
505
            break;
769
2.67k
        case TLocalPartitionType::PASS_TO_ONE:
770
2.67k
            if (_runtime_state->enable_share_hash_table_for_broadcast_join()) {
771
1.47k
                info.shared_state->exchanger = PassToOneExchanger::create_unique(
772
1.47k
                        sender_count, _num_instances, info.free_blocks_limit);
773
1.47k
            } else {
774
1.20k
                info.shared_state->exchanger = BroadcastExchanger::create_unique(
775
1.20k
                        sender_count, _num_instances, info.free_blocks_limit);
776
1.20k
            }
777
2.67k
            break;
778
858
        case TLocalPartitionType::ADAPTIVE_PASSTHROUGH:
779
858
            info.shared_state->exchanger = AdaptivePassthroughExchanger::create_unique(
780
858
                    sender_count, _num_instances, info.free_blocks_limit);
781
858
            break;
782
0
        case TLocalPartitionType::NOOP:
783
0
        case TLocalPartitionType::LOCAL_MERGE_SORT:
784
            // FE-planned LocalExchangeNode currently never emits NOOP or LOCAL_MERGE_SORT
785
            // through the deferred-exchanger path.  NOOP means "no exchange needed" and
786
            // is filtered out before reaching here; LOCAL_MERGE_SORT is planned by the
787
            // legacy BE path only.  Crash in debug to surface the protocol violation if
788
            // that ever changes; return an error in release to avoid silently corrupting
789
            // execution.
790
0
            DCHECK(false) << "FE-planned local exchange should not emit partition_type="
791
0
                          << static_cast<int>(info.partition_type);
792
0
            return Status::InternalError("FE-planned local exchange emitted unsupported type: " +
793
0
                                         std::to_string(static_cast<int>(info.partition_type)));
794
0
        default:
795
            // New TLocalPartitionType added on FE side without a BE handler here.
796
0
            DCHECK(false) << "Unhandled TLocalPartitionType in deferred exchangers: "
797
0
                          << static_cast<int>(info.partition_type);
798
0
            return Status::InternalError("Unsupported FE-planned local exchange type: " +
799
0
                                         std::to_string(static_cast<int>(info.partition_type)));
800
109k
        }
801
109k
    }
802
391k
    _deferred_exchangers.clear();
803
391k
    return Status::OK();
804
391k
}
805
806
391k
void PipelineFragmentContext::_propagate_local_exchange_num_tasks() {
807
    // Only runs when FE has planned local exchanges and BE deferred their construction.
808
    // In legacy mode (enable_local_shuffle_planner=false) BE plans LE itself via
809
    // _plan_local_exchange and _deferred_exchangers stays empty — the legacy path
810
    // already gets its num_tasks right at construction time, so the propagate passes
811
    // would be no-ops and are skipped.  This is a transitional design: once the FE
812
    // planner is the only planner, the propagation logic itself should degrade into
813
    // a pure assertion that the FE plan already wired the right num_tasks everywhere.
814
391k
    if (_deferred_exchangers.empty()) {
815
311k
        return;
816
311k
    }
817
    // Reconcile num_tasks across paired pipelines created by pipeline-splitting operators
818
    // (AGG, SORT, JOIN): they share state via inject_shared_state and must agree, or
819
    // instance 1+ tasks access null shared_state.  A pipeline's num_tasks is fully
820
    // determined by its source operator plus its upstreams:
821
    //   - LocalExchangeSource  -> _num_instances (the LE re-parallelizes)
822
    //   - serial source        -> its reduced count (kept as-is, typically 1)
823
    //   - otherwise (splitter) -> inherit from upstreams: raise to _num_instances if any
824
    //                             upstream was raised by an LE, then lower to a serial
825
    //                             upstream's count (lower wins).
826
    // Visiting each pipeline only after all its upstreams (topological order over _dag) lets
827
    // a single sweep reach the same fixpoint the previous two while-loops iterated to — those
828
    // only existed to reconcile the top-down build's parent-inherited num_tasks guesses.
829
80.2k
    std::map<PipelineId, PipelinePtr> id_to_pipe;
830
80.2k
    std::map<PipelineId, std::vector<PipelineId>> downstreams_of;
831
80.2k
    std::map<PipelineId, int> in_degree;
832
238k
    for (auto& p : _pipelines) {
833
238k
        id_to_pipe[p->id()] = p;
834
238k
        in_degree.try_emplace(p->id(), 0);
835
238k
    }
836
152k
    for (const auto& [downstream_id, upstream_ids] : _dag) {
837
158k
        for (auto upstream_id : upstream_ids) {
838
158k
            downstreams_of[upstream_id].push_back(downstream_id);
839
158k
            in_degree[downstream_id]++;
840
158k
        }
841
152k
    }
842
80.2k
    std::vector<PipelineId> ready;
843
238k
    for (const auto& [id, deg] : in_degree) {
844
238k
        if (deg == 0) {
845
85.9k
            ready.push_back(id);
846
85.9k
        }
847
238k
    }
848
80.2k
    size_t visited = 0;
849
318k
    while (!ready.empty()) {
850
238k
        const auto id = ready.back();
851
238k
        ready.pop_back();
852
238k
        visited++;
853
238k
        auto pit = id_to_pipe.find(id);
854
238k
        if (pit != id_to_pipe.end()) {
855
238k
            auto& pipe = pit->second;
856
238k
            const auto& ops = pipe->operators();
857
238k
            const bool le_source =
858
238k
                    !ops.empty() && dynamic_cast<LocalExchangeSourceOperatorX*>(ops.front().get());
859
238k
            const bool serial_source = !ops.empty() && ops.front()->is_serial_operator();
860
238k
            if (le_source) {
861
109k
                pipe->set_num_tasks(_num_instances);
862
129k
            } else if (!serial_source) {
863
60.0k
                int target = pipe->num_tasks();
864
60.0k
                const auto up_it = _dag.find(id);
865
60.0k
                if (up_it != _dag.end()) {
866
                    // raise: any upstream already at _num_instances (e.g. an LE source)
867
43.6k
                    for (auto upstream_id : up_it->second) {
868
43.6k
                        auto uit = id_to_pipe.find(upstream_id);
869
43.6k
                        if (uit != id_to_pipe.end() && uit->second->num_tasks() >= _num_instances) {
870
43.6k
                            target = _num_instances;
871
43.6k
                            break;
872
43.6k
                        }
873
43.6k
                    }
874
                    // lower: a serial upstream with fewer tasks (wins over the raise above)
875
44.3k
                    for (auto upstream_id : up_it->second) {
876
44.3k
                        auto uit = id_to_pipe.find(upstream_id);
877
44.3k
                        if (uit != id_to_pipe.end() && uit->second->num_tasks() < target &&
878
44.3k
                            !uit->second->operators().empty() &&
879
44.3k
                            uit->second->operators().front()->is_serial_operator()) {
880
0
                            target = uit->second->num_tasks();
881
0
                        }
882
44.3k
                    }
883
43.6k
                }
884
60.0k
                pipe->set_num_tasks(target);
885
60.0k
            }
886
238k
        }
887
238k
        for (auto down : downstreams_of[id]) {
888
158k
            if (--in_degree[down] == 0) {
889
152k
                ready.push_back(down);
890
152k
            }
891
158k
        }
892
238k
    }
893
    // The pipeline DAG is acyclic; if a future change introduces a back-edge, some pipelines
894
    // stay unvisited (in_degree never reaches 0) — fail loudly rather than silently leaving
895
    // their num_tasks unreconciled.
896
80.2k
    DCHECK_EQ(visited, in_degree.size())
897
0
            << "pipeline num_tasks topological sweep visited " << visited << " of "
898
0
            << in_degree.size() << " pipelines (cycle in _dag?)";
899
80.2k
}
900
901
Status PipelineFragmentContext::_create_tree_helper(
902
        ObjectPool* pool, const std::vector<TPlanNode>& tnodes, const DescriptorTbl& descs,
903
        OperatorPtr parent, int* node_idx, OperatorPtr* root, PipelinePtr& cur_pipe, int child_idx,
904
701k
        const bool followed_by_shuffled_operator, const bool require_bucket_distribution) {
905
    // propagate error case
906
701k
    if (*node_idx >= tnodes.size()) {
907
0
        return Status::InternalError(
908
0
                "Failed to reconstruct plan tree from thrift. Node id: {}, number of nodes: {}",
909
0
                *node_idx, tnodes.size());
910
0
    }
911
701k
    const TPlanNode& tnode = tnodes[*node_idx];
912
913
701k
    int num_children = tnodes[*node_idx].num_children;
914
701k
    bool current_followed_by_shuffled_operator = followed_by_shuffled_operator;
915
701k
    bool current_require_bucket_distribution = require_bucket_distribution;
916
    // TODO: Create CacheOperator is confused now
917
701k
    OperatorPtr op = nullptr;
918
701k
    OperatorPtr cache_op = nullptr;
919
701k
    RETURN_IF_ERROR(_create_operator(pool, tnodes[*node_idx], descs, op, cur_pipe,
920
701k
                                     parent == nullptr ? -1 : parent->node_id(), child_idx,
921
701k
                                     followed_by_shuffled_operator,
922
701k
                                     current_require_bucket_distribution, cache_op));
923
    // Initialization must be done here. For example, group by expressions in agg will be used to
924
    // decide if a local shuffle should be planed, so it must be initialized here.
925
701k
    RETURN_IF_ERROR(op->init(tnode, _runtime_state.get()));
926
    // assert(parent != nullptr || (node_idx == 0 && root_expr != nullptr));
927
701k
    if (parent != nullptr) {
928
        // add to parent's child(s)
929
311k
        RETURN_IF_ERROR(parent->set_child(cache_op ? cache_op : op));
930
389k
    } else {
931
389k
        *root = op;
932
389k
    }
933
    /**
934
     * `TLocalPartitionType::GLOBAL_EXECUTION_HASH_SHUFFLE` should be used if an operator is followed by a shuffled operator (shuffled hash join, union operator followed by co-located operators).
935
     *
936
     * For plan:
937
     * LocalExchange(id=0) -> Aggregation(id=1) -> ShuffledHashJoin(id=2)
938
     *                           Exchange(id=3) -> ShuffledHashJoinBuild(id=2)
939
     * We must ensure data distribution of `LocalExchange(id=0)` is same as Exchange(id=3).
940
     *
941
     * If an operator's is followed by a local exchange without shuffle (e.g. passthrough), a
942
     * shuffled local exchanger will be used before join so it is not followed by shuffle join.
943
     */
944
701k
    auto required_data_distribution =
945
701k
            cur_pipe->operators().empty()
946
701k
                    ? cur_pipe->sink()->required_data_distribution(_runtime_state.get())
947
701k
                    : op->required_data_distribution(_runtime_state.get());
948
701k
    current_followed_by_shuffled_operator =
949
701k
            ((followed_by_shuffled_operator ||
950
701k
              (cur_pipe->operators().empty() ? cur_pipe->sink()->is_shuffled_operator()
951
634k
                                             : op->is_shuffled_operator())) &&
952
701k
             Pipeline::is_hash_exchange(required_data_distribution.distribution_type)) ||
953
701k
            (followed_by_shuffled_operator &&
954
587k
             required_data_distribution.distribution_type == TLocalPartitionType::NOOP);
955
956
701k
    current_require_bucket_distribution =
957
701k
            ((require_bucket_distribution ||
958
701k
              (cur_pipe->operators().empty() ? cur_pipe->sink()->is_colocated_operator()
959
641k
                                             : op->is_colocated_operator())) &&
960
701k
             Pipeline::is_hash_exchange(required_data_distribution.distribution_type)) ||
961
701k
            (require_bucket_distribution &&
962
594k
             required_data_distribution.distribution_type == TLocalPartitionType::NOOP);
963
964
701k
    if (num_children == 0) {
965
406k
        _use_serial_source = op->is_serial_operator();
966
406k
    }
967
    // rely on that tnodes is preorder of the plan
968
1.01M
    for (int i = 0; i < num_children; i++) {
969
311k
        ++*node_idx;
970
311k
        RETURN_IF_ERROR(_create_tree_helper(pool, tnodes, descs, op, node_idx, nullptr, cur_pipe, i,
971
311k
                                            current_followed_by_shuffled_operator,
972
311k
                                            current_require_bucket_distribution));
973
974
        // we are expecting a child, but have used all nodes
975
        // this means we have been given a bad tree and must fail
976
311k
        if (*node_idx >= tnodes.size()) {
977
0
            return Status::InternalError(
978
0
                    "Failed to reconstruct plan tree from thrift. Node id: {}, number of "
979
0
                    "nodes: {}",
980
0
                    *node_idx, tnodes.size());
981
0
        }
982
311k
    }
983
984
701k
    return Status::OK();
985
701k
}
986
987
void PipelineFragmentContext::_inherit_pipeline_properties(
988
        const DataDistribution& data_distribution, PipelinePtr pipe_with_source,
989
756
        PipelinePtr pipe_with_sink) {
990
756
    pipe_with_sink->set_num_tasks(pipe_with_source->num_tasks());
991
756
    pipe_with_source->set_num_tasks(_num_instances);
992
756
    pipe_with_source->set_data_distribution(data_distribution);
993
756
}
994
995
Status PipelineFragmentContext::_add_local_exchange_impl(
996
        int idx, ObjectPool* pool, PipelinePtr cur_pipe, PipelinePtr new_pip,
997
        DataDistribution data_distribution, bool* do_local_exchange, int num_buckets,
998
        const std::map<int, int>& bucket_seq_to_instance_idx,
999
756
        const std::map<int, int>& shuffle_idx_to_instance_idx) {
1000
756
    auto& operators = cur_pipe->operators();
1001
756
    const auto downstream_pipeline_id = cur_pipe->id();
1002
756
    auto local_exchange_id = next_operator_id();
1003
    // 1. Create a new pipeline with local exchange sink.
1004
756
    DataSinkOperatorPtr sink;
1005
756
    auto sink_id = next_sink_operator_id();
1006
1007
    /**
1008
     * `bucket_seq_to_instance_idx` is empty if no scan operator is contained in this fragment.
1009
     * So co-located operators(e.g. Agg, Analytic) should use `HASH_SHUFFLE` instead of `BUCKET_HASH_SHUFFLE`.
1010
     */
1011
756
    const bool followed_by_shuffled_operator =
1012
756
            operators.size() > idx ? operators[idx]->followed_by_shuffled_operator()
1013
756
                                   : cur_pipe->sink()->followed_by_shuffled_operator();
1014
756
    const bool use_global_hash_shuffle = bucket_seq_to_instance_idx.empty() &&
1015
756
                                         !shuffle_idx_to_instance_idx.contains(-1) &&
1016
756
                                         followed_by_shuffled_operator && !_use_serial_source;
1017
756
    sink = std::make_shared<LocalExchangeSinkOperatorX>(
1018
756
            sink_id, local_exchange_id, use_global_hash_shuffle ? _total_instances : _num_instances,
1019
756
            data_distribution.partition_exprs, bucket_seq_to_instance_idx);
1020
756
    if (bucket_seq_to_instance_idx.empty() &&
1021
756
        data_distribution.distribution_type == TLocalPartitionType::BUCKET_HASH_SHUFFLE) {
1022
2
        data_distribution.distribution_type =
1023
2
                use_global_hash_shuffle ? TLocalPartitionType::GLOBAL_EXECUTION_HASH_SHUFFLE
1024
2
                                        : TLocalPartitionType::LOCAL_EXECUTION_HASH_SHUFFLE;
1025
2
    }
1026
756
    if (!use_global_hash_shuffle &&
1027
756
        data_distribution.distribution_type == TLocalPartitionType::GLOBAL_EXECUTION_HASH_SHUFFLE) {
1028
71
        data_distribution.distribution_type = TLocalPartitionType::LOCAL_EXECUTION_HASH_SHUFFLE;
1029
71
    }
1030
756
    RETURN_IF_ERROR(new_pip->set_sink(sink));
1031
756
    RETURN_IF_ERROR(new_pip->sink()->init(_runtime_state.get(), data_distribution.distribution_type,
1032
756
                                          num_buckets, shuffle_idx_to_instance_idx));
1033
1034
    // 2. Create and initialize LocalExchangeSharedState.
1035
756
    std::shared_ptr<LocalExchangeSharedState> shared_state =
1036
756
            LocalExchangeSharedState::create_shared(_num_instances);
1037
756
    switch (data_distribution.distribution_type) {
1038
71
    case TLocalPartitionType::LOCAL_EXECUTION_HASH_SHUFFLE:
1039
74
    case TLocalPartitionType::GLOBAL_EXECUTION_HASH_SHUFFLE:
1040
74
        shared_state->exchanger = ShuffleExchanger::create_unique(
1041
74
                std::max(cur_pipe->num_tasks(), _num_instances), _num_instances,
1042
74
                use_global_hash_shuffle ? _total_instances : _num_instances,
1043
74
                _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
1044
74
                        ? cast_set<int>(
1045
74
                                  _runtime_state->query_options().local_exchange_free_blocks_limit)
1046
74
                        : 0,
1047
74
                data_distribution.distribution_type);
1048
74
        break;
1049
10
    case TLocalPartitionType::BUCKET_HASH_SHUFFLE:
1050
10
        shared_state->exchanger = BucketShuffleExchanger::create_unique(
1051
10
                std::max(cur_pipe->num_tasks(), _num_instances), _num_instances, num_buckets,
1052
10
                _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
1053
10
                        ? cast_set<int>(
1054
10
                                  _runtime_state->query_options().local_exchange_free_blocks_limit)
1055
10
                        : 0);
1056
10
        break;
1057
579
    case TLocalPartitionType::PASSTHROUGH:
1058
579
        shared_state->exchanger = PassthroughExchanger::create_unique(
1059
579
                cur_pipe->num_tasks(), _num_instances,
1060
579
                _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
1061
579
                        ? cast_set<int>(
1062
579
                                  _runtime_state->query_options().local_exchange_free_blocks_limit)
1063
579
                        : 0);
1064
579
        break;
1065
8
    case TLocalPartitionType::BROADCAST:
1066
8
        shared_state->exchanger = BroadcastExchanger::create_unique(
1067
8
                cur_pipe->num_tasks(), _num_instances,
1068
8
                _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
1069
8
                        ? cast_set<int>(
1070
8
                                  _runtime_state->query_options().local_exchange_free_blocks_limit)
1071
8
                        : 0);
1072
8
        break;
1073
2
    case TLocalPartitionType::PASS_TO_ONE:
1074
2
        if (_runtime_state->enable_share_hash_table_for_broadcast_join()) {
1075
            // If shared hash table is enabled for BJ, hash table will be built by only one task
1076
0
            shared_state->exchanger = PassToOneExchanger::create_unique(
1077
0
                    cur_pipe->num_tasks(), _num_instances,
1078
0
                    _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
1079
0
                            ? cast_set<int>(_runtime_state->query_options()
1080
0
                                                    .local_exchange_free_blocks_limit)
1081
0
                            : 0);
1082
2
        } else {
1083
2
            shared_state->exchanger = BroadcastExchanger::create_unique(
1084
2
                    cur_pipe->num_tasks(), _num_instances,
1085
2
                    _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
1086
2
                            ? cast_set<int>(_runtime_state->query_options()
1087
2
                                                    .local_exchange_free_blocks_limit)
1088
2
                            : 0);
1089
2
        }
1090
2
        break;
1091
83
    case TLocalPartitionType::ADAPTIVE_PASSTHROUGH:
1092
83
        shared_state->exchanger = AdaptivePassthroughExchanger::create_unique(
1093
83
                std::max(cur_pipe->num_tasks(), _num_instances), _num_instances,
1094
83
                _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
1095
83
                        ? cast_set<int>(
1096
83
                                  _runtime_state->query_options().local_exchange_free_blocks_limit)
1097
83
                        : 0);
1098
83
        break;
1099
0
    default:
1100
0
        return Status::InternalError("Unsupported local exchange type : " +
1101
0
                                     std::to_string((int)data_distribution.distribution_type));
1102
756
    }
1103
756
    shared_state->create_source_dependencies(_num_instances, local_exchange_id, local_exchange_id,
1104
756
                                             "LOCAL_EXCHANGE_OPERATOR");
1105
756
    shared_state->create_sink_dependency(sink_id, local_exchange_id, "LOCAL_EXCHANGE_SINK");
1106
756
    _op_id_to_shared_state.insert({local_exchange_id, {shared_state, shared_state->sink_deps}});
1107
1108
    // 3. Set two pipelines' operator list. For example, split pipeline [Scan - AggSink] to
1109
    // pipeline1 [Scan - LocalExchangeSink] and pipeline2 [LocalExchangeSource - AggSink].
1110
1111
    // 3.1 Initialize new pipeline's operator list.
1112
756
    std::copy(operators.begin(), operators.begin() + idx,
1113
756
              std::inserter(new_pip->operators(), new_pip->operators().end()));
1114
1115
    // 3.2 Erase unused operators in previous pipeline.
1116
756
    operators.erase(operators.begin(), operators.begin() + idx);
1117
1118
    // 4. Initialize LocalExchangeSource and insert it into this pipeline.
1119
756
    OperatorPtr source_op;
1120
756
    source_op = std::make_shared<LocalExchangeSourceOperatorX>(pool, local_exchange_id);
1121
756
    RETURN_IF_ERROR(source_op->set_child(new_pip->operators().back()));
1122
756
    RETURN_IF_ERROR(source_op->init(data_distribution.distribution_type));
1123
756
    if (!operators.empty()) {
1124
198
        RETURN_IF_ERROR(operators.front()->set_child(nullptr));
1125
198
        RETURN_IF_ERROR(operators.front()->set_child(source_op));
1126
198
    }
1127
756
    operators.insert(operators.begin(), source_op);
1128
1129
    // 5. Set children for two pipelines separately.
1130
756
    std::vector<std::shared_ptr<Pipeline>> new_children;
1131
756
    std::vector<PipelineId> edges_with_source;
1132
1.53k
    for (auto child : cur_pipe->children()) {
1133
1.53k
        bool found = false;
1134
2.12k
        for (auto op : new_pip->operators()) {
1135
2.12k
            if (child->sink()->node_id() == op->node_id()) {
1136
559
                new_pip->set_children(child);
1137
559
                found = true;
1138
559
            };
1139
2.12k
        }
1140
1.53k
        if (!found) {
1141
972
            new_children.push_back(child);
1142
972
            edges_with_source.push_back(child->id());
1143
972
        }
1144
1.53k
    }
1145
756
    new_children.push_back(new_pip);
1146
756
    edges_with_source.push_back(new_pip->id());
1147
1148
    // 6. Set DAG for new pipelines.
1149
756
    if (!new_pip->children().empty()) {
1150
339
        std::vector<PipelineId> edges_with_sink;
1151
559
        for (auto child : new_pip->children()) {
1152
559
            edges_with_sink.push_back(child->id());
1153
559
        }
1154
339
        _dag.insert({new_pip->id(), edges_with_sink});
1155
339
    }
1156
756
    cur_pipe->set_children(new_children);
1157
756
    _dag[downstream_pipeline_id] = edges_with_source;
1158
756
    RETURN_IF_ERROR(new_pip->sink()->set_child(new_pip->operators().back()));
1159
756
    RETURN_IF_ERROR(cur_pipe->sink()->set_child(nullptr));
1160
756
    RETURN_IF_ERROR(cur_pipe->sink()->set_child(cur_pipe->operators().back()));
1161
1162
    // 7. Inherit properties from current pipeline.
1163
756
    _inherit_pipeline_properties(data_distribution, cur_pipe, new_pip);
1164
756
    return Status::OK();
1165
756
}
1166
1167
Status PipelineFragmentContext::_add_local_exchange(
1168
        int pip_idx, int idx, int node_id, ObjectPool* pool, PipelinePtr cur_pipe,
1169
        DataDistribution data_distribution, bool* do_local_exchange, int num_buckets,
1170
        const std::map<int, int>& bucket_seq_to_instance_idx,
1171
8.93k
        const std::map<int, int>& shuffle_idx_to_instance_idx) {
1172
8.93k
    if (_num_instances <= 1 || cur_pipe->num_tasks_of_parent() <= 1) {
1173
7.50k
        return Status::OK();
1174
7.50k
    }
1175
1176
1.43k
    if (!cur_pipe->need_to_local_exchange(data_distribution, idx)) {
1177
694
        return Status::OK();
1178
694
    }
1179
738
    *do_local_exchange = true;
1180
1181
738
    auto& operators = cur_pipe->operators();
1182
738
    auto total_op_num = operators.size();
1183
738
    auto new_pip = add_pipeline(cur_pipe, pip_idx + 1);
1184
738
    RETURN_IF_ERROR(_add_local_exchange_impl(
1185
738
            idx, pool, cur_pipe, new_pip, data_distribution, do_local_exchange, num_buckets,
1186
738
            bucket_seq_to_instance_idx, shuffle_idx_to_instance_idx));
1187
1188
738
    CHECK(total_op_num + 1 == cur_pipe->operators().size() + new_pip->operators().size())
1189
0
            << "total_op_num: " << total_op_num
1190
0
            << " cur_pipe->operators().size(): " << cur_pipe->operators().size()
1191
0
            << " new_pip->operators().size(): " << new_pip->operators().size();
1192
1193
    // There are some local shuffles with relatively heavy operations on the sink.
1194
    // If the local sink concurrency is 1 and the local source concurrency is n, the sink becomes a bottleneck.
1195
    // Therefore, local passthrough is used to increase the concurrency of the sink.
1196
    // op -> local sink(1) -> local source (n)
1197
    // op -> local passthrough(1) -> local passthrough(n) ->  local sink(n) -> local source (n)
1198
738
    if (cur_pipe->num_tasks() > 1 && new_pip->num_tasks() == 1 &&
1199
738
        Pipeline::heavy_operations_on_the_sink(data_distribution.distribution_type)) {
1200
18
        RETURN_IF_ERROR(_add_local_exchange_impl(
1201
18
                cast_set<int>(new_pip->operators().size()), pool, new_pip,
1202
18
                add_pipeline(new_pip, pip_idx + 2),
1203
18
                DataDistribution(TLocalPartitionType::PASSTHROUGH), do_local_exchange, num_buckets,
1204
18
                bucket_seq_to_instance_idx, shuffle_idx_to_instance_idx));
1205
18
    }
1206
738
    return Status::OK();
1207
738
}
1208
1209
Status PipelineFragmentContext::_plan_local_exchange(
1210
        int num_buckets, const std::map<int, int>& bucket_seq_to_instance_idx,
1211
116k
        const std::map<int, int>& shuffle_idx_to_instance_idx) {
1212
262k
    for (int pip_idx = cast_set<int>(_pipelines.size()) - 1; pip_idx >= 0; pip_idx--) {
1213
145k
        _pipelines[pip_idx]->init_data_distribution(_runtime_state.get());
1214
        // Set property if child pipeline is not join operator's child.
1215
145k
        if (!_pipelines[pip_idx]->children().empty()) {
1216
25.7k
            for (auto& child : _pipelines[pip_idx]->children()) {
1217
25.7k
                if (child->sink()->node_id() ==
1218
25.7k
                    _pipelines[pip_idx]->operators().front()->node_id()) {
1219
20.7k
                    _pipelines[pip_idx]->set_data_distribution(child->data_distribution());
1220
20.7k
                }
1221
25.7k
            }
1222
23.4k
        }
1223
1224
        // if 'num_buckets == 0' means the fragment is colocated by exchange node not the
1225
        // scan node. so here use `_num_instance` to replace the `num_buckets` to prevent dividing 0
1226
        // still keep colocate plan after local shuffle
1227
145k
        RETURN_IF_ERROR(_plan_local_exchange(num_buckets, pip_idx, _pipelines[pip_idx],
1228
145k
                                             bucket_seq_to_instance_idx,
1229
145k
                                             shuffle_idx_to_instance_idx));
1230
145k
    }
1231
116k
    return Status::OK();
1232
116k
}
1233
1234
Status PipelineFragmentContext::_plan_local_exchange(
1235
        int num_buckets, int pip_idx, PipelinePtr pip,
1236
        const std::map<int, int>& bucket_seq_to_instance_idx,
1237
145k
        const std::map<int, int>& shuffle_idx_to_instance_idx) {
1238
145k
    int idx = 1;
1239
145k
    bool do_local_exchange = false;
1240
145k
    do {
1241
145k
        auto& ops = pip->operators();
1242
145k
        do_local_exchange = false;
1243
        // Plan local exchange for each operator.
1244
152k
        for (; idx < ops.size();) {
1245
6.56k
            auto _le_req = ops[idx]->required_data_distribution(_runtime_state.get());
1246
6.56k
            if (_le_req.need_local_exchange()) {
1247
4.98k
                RETURN_IF_ERROR(_add_local_exchange(
1248
4.98k
                        pip_idx, idx, ops[idx]->node_id(), _runtime_state->obj_pool(), pip, _le_req,
1249
4.98k
                        &do_local_exchange, num_buckets, bucket_seq_to_instance_idx,
1250
4.98k
                        shuffle_idx_to_instance_idx));
1251
4.98k
            }
1252
6.56k
            if (do_local_exchange) {
1253
                // If local exchange is needed for current operator, we will split this pipeline to
1254
                // two pipelines by local exchange sink/source. And then we need to process remaining
1255
                // operators in this pipeline so we set idx to 2 (0 is local exchange source and 1
1256
                // is current operator was already processed) and continue to plan local exchange.
1257
198
                idx = 2;
1258
198
                break;
1259
198
            }
1260
6.37k
            idx++;
1261
6.37k
        }
1262
145k
    } while (do_local_exchange);
1263
145k
    if (pip->sink()->required_data_distribution(_runtime_state.get()).need_local_exchange()) {
1264
3.95k
        RETURN_IF_ERROR(_add_local_exchange(
1265
3.95k
                pip_idx, idx, pip->sink()->node_id(), _runtime_state->obj_pool(), pip,
1266
3.95k
                pip->sink()->required_data_distribution(_runtime_state.get()), &do_local_exchange,
1267
3.95k
                num_buckets, bucket_seq_to_instance_idx, shuffle_idx_to_instance_idx));
1268
3.95k
    }
1269
145k
    return Status::OK();
1270
145k
}
1271
1272
Status PipelineFragmentContext::_create_data_sink(ObjectPool* pool, const TDataSink& thrift_sink,
1273
                                                  const std::vector<TExpr>& output_exprs,
1274
                                                  const TPipelineFragmentParams& params,
1275
                                                  const RowDescriptor& row_desc,
1276
                                                  RuntimeState* state, DescriptorTbl& desc_tbl,
1277
392k
                                                  PipelineId cur_pipeline_id) {
1278
392k
    switch (thrift_sink.type) {
1279
133k
    case TDataSinkType::DATA_STREAM_SINK: {
1280
133k
        if (!thrift_sink.__isset.stream_sink) {
1281
0
            return Status::InternalError("Missing data stream sink.");
1282
0
        }
1283
133k
        _sink = std::make_shared<ExchangeSinkOperatorX>(
1284
133k
                state, row_desc, next_sink_operator_id(), thrift_sink.stream_sink,
1285
133k
                params.destinations, _fragment_instance_ids);
1286
133k
        break;
1287
133k
    }
1288
222k
    case TDataSinkType::RESULT_SINK: {
1289
222k
        if (!thrift_sink.__isset.result_sink) {
1290
0
            return Status::InternalError("Missing data buffer sink.");
1291
0
        }
1292
1293
222k
        auto& pipeline = _pipelines[cur_pipeline_id];
1294
222k
        int child_node_id = pipeline->operators().back()->node_id();
1295
222k
        _sink = std::make_shared<ResultSinkOperatorX>(next_sink_operator_id(), child_node_id + 1,
1296
222k
                                                      row_desc, output_exprs,
1297
222k
                                                      thrift_sink.result_sink);
1298
222k
        break;
1299
222k
    }
1300
105
    case TDataSinkType::DICTIONARY_SINK: {
1301
105
        if (!thrift_sink.__isset.dictionary_sink) {
1302
0
            return Status::InternalError("Missing dict sink.");
1303
0
        }
1304
1305
105
        _sink = std::make_shared<DictSinkOperatorX>(next_sink_operator_id(), row_desc, output_exprs,
1306
105
                                                    thrift_sink.dictionary_sink);
1307
105
        break;
1308
105
    }
1309
0
    case TDataSinkType::GROUP_COMMIT_OLAP_TABLE_SINK:
1310
31.9k
    case TDataSinkType::OLAP_TABLE_SINK: {
1311
31.9k
        auto& pipeline = _pipelines[cur_pipeline_id];
1312
31.9k
        int child_node_id = pipeline->operators().back()->node_id();
1313
31.9k
        if (state->query_options().enable_memtable_on_sink_node &&
1314
31.9k
            !_has_inverted_index_v1_or_partial_update(thrift_sink.olap_table_sink) &&
1315
31.9k
            !_has_row_binlog(thrift_sink.olap_table_sink) && !config::is_cloud_mode()) {
1316
1.47k
            _sink = std::make_shared<OlapTableSinkV2OperatorX>(
1317
1.47k
                    pool, next_sink_operator_id(), child_node_id + 1, row_desc, output_exprs);
1318
30.4k
        } else {
1319
30.4k
            _sink = std::make_shared<OlapTableSinkOperatorX>(
1320
30.4k
                    pool, next_sink_operator_id(), child_node_id + 1, row_desc, output_exprs);
1321
30.4k
        }
1322
31.9k
        break;
1323
0
    }
1324
168
    case TDataSinkType::GROUP_COMMIT_BLOCK_SINK: {
1325
168
        DCHECK(thrift_sink.__isset.olap_table_sink);
1326
168
        DCHECK(state->get_query_ctx() != nullptr);
1327
168
        state->get_query_ctx()->query_mem_tracker()->is_group_commit_load = true;
1328
168
        _sink = std::make_shared<GroupCommitBlockSinkOperatorX>(next_sink_operator_id(), row_desc,
1329
168
                                                                output_exprs);
1330
168
        break;
1331
0
    }
1332
742
    case TDataSinkType::HIVE_TABLE_SINK: {
1333
742
        if (!thrift_sink.__isset.hive_table_sink) {
1334
0
            return Status::InternalError("Missing hive table sink.");
1335
0
        }
1336
742
        _sink = std::make_shared<HiveTableSinkOperatorX>(pool, next_sink_operator_id(), row_desc,
1337
742
                                                         output_exprs);
1338
742
        break;
1339
742
    }
1340
868
    case TDataSinkType::ICEBERG_TABLE_SINK: {
1341
868
        if (!thrift_sink.__isset.iceberg_table_sink) {
1342
0
            return Status::InternalError("Missing iceberg table sink.");
1343
0
        }
1344
868
        if (thrift_sink.iceberg_table_sink.__isset.sort_info) {
1345
2
            _sink = std::make_shared<SpillIcebergTableSinkOperatorX>(pool, next_sink_operator_id(),
1346
2
                                                                     row_desc, output_exprs);
1347
866
        } else {
1348
866
            _sink = std::make_shared<IcebergTableSinkOperatorX>(pool, next_sink_operator_id(),
1349
866
                                                                row_desc, output_exprs);
1350
866
        }
1351
868
        break;
1352
868
    }
1353
10
    case TDataSinkType::ICEBERG_DELETE_SINK: {
1354
10
        if (!thrift_sink.__isset.iceberg_delete_sink) {
1355
0
            return Status::InternalError("Missing iceberg delete sink.");
1356
0
        }
1357
10
        _sink = std::make_shared<IcebergDeleteSinkOperatorX>(pool, next_sink_operator_id(),
1358
10
                                                             row_desc, output_exprs);
1359
10
        break;
1360
10
    }
1361
40
    case TDataSinkType::ICEBERG_MERGE_SINK: {
1362
40
        if (!thrift_sink.__isset.iceberg_merge_sink) {
1363
0
            return Status::InternalError("Missing iceberg merge sink.");
1364
0
        }
1365
40
        _sink = std::make_shared<IcebergMergeSinkOperatorX>(pool, next_sink_operator_id(), row_desc,
1366
40
                                                            output_exprs);
1367
40
        break;
1368
40
    }
1369
0
    case TDataSinkType::MAXCOMPUTE_TABLE_SINK: {
1370
0
        if (!thrift_sink.__isset.max_compute_table_sink) {
1371
0
            return Status::InternalError("Missing max compute table sink.");
1372
0
        }
1373
0
        _sink = std::make_shared<MCTableSinkOperatorX>(pool, next_sink_operator_id(), row_desc,
1374
0
                                                       output_exprs);
1375
0
        break;
1376
0
    }
1377
44
    case TDataSinkType::JDBC_TABLE_SINK: {
1378
44
        if (!thrift_sink.__isset.jdbc_table_sink) {
1379
0
            return Status::InternalError("Missing data jdbc sink.");
1380
0
        }
1381
44
        if (config::enable_java_support) {
1382
44
            _sink = std::make_shared<JdbcTableSinkOperatorX>(row_desc, next_sink_operator_id(),
1383
44
                                                             output_exprs);
1384
44
        } else {
1385
0
            return Status::InternalError(
1386
0
                    "Jdbc table sink is not enabled, you can change be config "
1387
0
                    "enable_java_support to true and restart be.");
1388
0
        }
1389
44
        break;
1390
44
    }
1391
44
    case TDataSinkType::MEMORY_SCRATCH_SINK: {
1392
3
        if (!thrift_sink.__isset.memory_scratch_sink) {
1393
0
            return Status::InternalError("Missing data buffer sink.");
1394
0
        }
1395
1396
3
        _sink = std::make_shared<MemoryScratchSinkOperatorX>(row_desc, next_sink_operator_id(),
1397
3
                                                             output_exprs);
1398
3
        break;
1399
3
    }
1400
421
    case TDataSinkType::RESULT_FILE_SINK: {
1401
421
        if (!thrift_sink.__isset.result_file_sink) {
1402
0
            return Status::InternalError("Missing result file sink.");
1403
0
        }
1404
1405
        // Result file sink is not the top sink
1406
421
        if (params.__isset.destinations && !params.destinations.empty()) {
1407
0
            _sink = std::make_shared<ResultFileSinkOperatorX>(
1408
0
                    next_sink_operator_id(), row_desc, thrift_sink.result_file_sink,
1409
0
                    params.destinations, output_exprs, desc_tbl);
1410
421
        } else {
1411
421
            _sink = std::make_shared<ResultFileSinkOperatorX>(next_sink_operator_id(), row_desc,
1412
421
                                                              output_exprs);
1413
421
        }
1414
421
        break;
1415
421
    }
1416
1.57k
    case TDataSinkType::MULTI_CAST_DATA_STREAM_SINK: {
1417
1.57k
        DCHECK(thrift_sink.__isset.multi_cast_stream_sink);
1418
1.57k
        DCHECK_GT(thrift_sink.multi_cast_stream_sink.sinks.size(), 0);
1419
1.57k
        auto sink_id = next_sink_operator_id();
1420
1.57k
        const int multi_cast_node_id = sink_id;
1421
1.57k
        auto sender_size = thrift_sink.multi_cast_stream_sink.sinks.size();
1422
        // one sink has multiple sources.
1423
1.57k
        std::vector<int> sources;
1424
6.09k
        for (int i = 0; i < sender_size; ++i) {
1425
4.51k
            auto source_id = next_operator_id();
1426
4.51k
            sources.push_back(source_id);
1427
4.51k
        }
1428
1429
1.57k
        _sink = std::make_shared<MultiCastDataStreamSinkOperatorX>(
1430
1.57k
                sink_id, multi_cast_node_id, sources, pool, thrift_sink.multi_cast_stream_sink);
1431
6.09k
        for (int i = 0; i < sender_size; ++i) {
1432
4.51k
            auto new_pipeline = add_pipeline();
1433
            // use to exchange sink
1434
4.51k
            RowDescriptor* exchange_row_desc = nullptr;
1435
4.51k
            {
1436
4.51k
                const auto& tmp_row_desc =
1437
4.51k
                        !thrift_sink.multi_cast_stream_sink.sinks[i].output_exprs.empty()
1438
4.51k
                                ? RowDescriptor(state->desc_tbl(),
1439
4.51k
                                                {thrift_sink.multi_cast_stream_sink.sinks[i]
1440
4.51k
                                                         .output_tuple_id})
1441
4.51k
                                : row_desc;
1442
4.51k
                exchange_row_desc = pool->add(new RowDescriptor(tmp_row_desc));
1443
4.51k
            }
1444
4.51k
            auto source_id = sources[i];
1445
4.51k
            OperatorPtr source_op;
1446
            // 1. create and set the source operator of multi_cast_data_stream_source for new pipeline
1447
4.51k
            source_op = std::make_shared<MultiCastDataStreamerSourceOperatorX>(
1448
4.51k
                    /*node_id*/ source_id, /*consumer_id*/ i, pool,
1449
4.51k
                    thrift_sink.multi_cast_stream_sink.sinks[i], row_desc,
1450
4.51k
                    /*operator_id=*/source_id);
1451
4.51k
            RETURN_IF_ERROR(new_pipeline->add_operator(
1452
4.51k
                    source_op, params.__isset.parallel_instances ? params.parallel_instances : 0));
1453
            // 2. create and set sink operator of data stream sender for new pipeline
1454
1455
4.51k
            DataSinkOperatorPtr sink_op;
1456
4.51k
            sink_op = std::make_shared<ExchangeSinkOperatorX>(
1457
4.51k
                    state, *exchange_row_desc, next_sink_operator_id(),
1458
4.51k
                    thrift_sink.multi_cast_stream_sink.sinks[i],
1459
4.51k
                    thrift_sink.multi_cast_stream_sink.destinations[i], _fragment_instance_ids);
1460
1461
4.51k
            RETURN_IF_ERROR(new_pipeline->set_sink(sink_op));
1462
4.51k
            {
1463
4.51k
                TDataSink* t = pool->add(new TDataSink());
1464
4.51k
                t->stream_sink = thrift_sink.multi_cast_stream_sink.sinks[i];
1465
4.51k
                RETURN_IF_ERROR(sink_op->init(*t));
1466
4.51k
            }
1467
1468
            // 3. set dependency dag
1469
4.51k
            _dag[new_pipeline->id()].push_back(cur_pipeline_id);
1470
4.51k
        }
1471
1.57k
        if (sources.empty()) {
1472
0
            return Status::InternalError("size of sources must be greater than 0");
1473
0
        }
1474
1.57k
        break;
1475
1.57k
    }
1476
1.57k
    case TDataSinkType::BLACKHOLE_SINK: {
1477
8
        if (!thrift_sink.__isset.blackhole_sink) {
1478
0
            return Status::InternalError("Missing blackhole sink.");
1479
0
        }
1480
1481
8
        _sink.reset(new BlackholeSinkOperatorX(next_sink_operator_id()));
1482
8
        break;
1483
8
    }
1484
78
    case TDataSinkType::TVF_TABLE_SINK: {
1485
78
        if (!thrift_sink.__isset.tvf_table_sink) {
1486
0
            return Status::InternalError("Missing TVF table sink.");
1487
0
        }
1488
78
        _sink = std::make_shared<TVFTableSinkOperatorX>(pool, next_sink_operator_id(), row_desc,
1489
78
                                                        output_exprs);
1490
78
        break;
1491
78
    }
1492
0
    default:
1493
0
        return Status::InternalError("Unsuported sink type in pipeline: {}", thrift_sink.type);
1494
392k
    }
1495
391k
    return Status::OK();
1496
392k
}
1497
1498
// NOLINTBEGIN(readability-function-size)
1499
// NOLINTBEGIN(readability-function-cognitive-complexity)
1500
Status PipelineFragmentContext::_create_operator(ObjectPool* pool, const TPlanNode& tnode,
1501
                                                 const DescriptorTbl& descs, OperatorPtr& op,
1502
                                                 PipelinePtr& cur_pipe, int parent_idx,
1503
                                                 int child_idx,
1504
                                                 const bool followed_by_shuffled_operator,
1505
                                                 const bool require_bucket_distribution,
1506
704k
                                                 OperatorPtr& cache_op) {
1507
704k
    std::vector<DataSinkOperatorPtr> sink_ops;
1508
704k
    Defer defer = Defer([&]() {
1509
703k
        if (op) {
1510
702k
            op->update_operator(tnode, followed_by_shuffled_operator, require_bucket_distribution);
1511
702k
        }
1512
703k
        for (auto& s : sink_ops) {
1513
209k
            s->update_operator(tnode, followed_by_shuffled_operator, require_bucket_distribution);
1514
209k
        }
1515
703k
    });
1516
    // We directly construct the operator from Thrift because the given array is in the order of preorder traversal.
1517
    // Therefore, here we need to use a stack-like structure.
1518
704k
    _pipeline_parent_map.pop(cur_pipe, parent_idx, child_idx);
1519
704k
    std::stringstream error_msg;
1520
704k
    bool enable_query_cache = _params.fragment.__isset.query_cache_param;
1521
1522
704k
    bool fe_with_old_version = false;
1523
704k
    switch (tnode.node_type) {
1524
192k
    case TPlanNodeType::OLAP_SCAN_NODE: {
1525
192k
        op = std::make_shared<OlapScanOperatorX>(
1526
192k
                pool, tnode, next_operator_id(), descs, _num_instances,
1527
192k
                enable_query_cache ? _params.fragment.query_cache_param : TQueryCacheParam {});
1528
192k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1529
192k
        fe_with_old_version = !tnode.__isset.is_serial_operator;
1530
192k
        break;
1531
192k
    }
1532
78
    case TPlanNodeType::GROUP_COMMIT_SCAN_NODE: {
1533
78
        DCHECK(_query_ctx != nullptr);
1534
78
        _query_ctx->query_mem_tracker()->is_group_commit_load = true;
1535
78
        op = std::make_shared<GroupCommitOperatorX>(pool, tnode, next_operator_id(), descs,
1536
78
                                                    _num_instances);
1537
78
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1538
78
        fe_with_old_version = !tnode.__isset.is_serial_operator;
1539
78
        break;
1540
78
    }
1541
0
    case TPlanNodeType::JDBC_SCAN_NODE: {
1542
0
        if (config::enable_java_support) {
1543
0
            op = std::make_shared<JDBCScanOperatorX>(pool, tnode, next_operator_id(), descs,
1544
0
                                                     _num_instances);
1545
0
            RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1546
0
        } else {
1547
0
            return Status::InternalError(
1548
0
                    "Jdbc scan node is disabled, you can change be config enable_java_support "
1549
0
                    "to true and restart be.");
1550
0
        }
1551
0
        fe_with_old_version = !tnode.__isset.is_serial_operator;
1552
0
        break;
1553
0
    }
1554
14.3k
    case TPlanNodeType::FILE_SCAN_NODE: {
1555
14.3k
        op = std::make_shared<FileScanOperatorX>(pool, tnode, next_operator_id(), descs,
1556
14.3k
                                                 _num_instances);
1557
14.3k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1558
14.3k
        fe_with_old_version = !tnode.__isset.is_serial_operator;
1559
14.3k
        break;
1560
14.3k
    }
1561
135k
    case TPlanNodeType::EXCHANGE_NODE: {
1562
135k
        int num_senders = _params.per_exch_num_senders.contains(tnode.node_id)
1563
135k
                                  ? _params.per_exch_num_senders.find(tnode.node_id)->second
1564
18.4E
                                  : 0;
1565
135k
        DCHECK_GT(num_senders, 0);
1566
135k
        op = std::make_shared<ExchangeSourceOperatorX>(pool, tnode, next_operator_id(), descs,
1567
135k
                                                       num_senders);
1568
135k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1569
135k
        fe_with_old_version = !tnode.__isset.is_serial_operator;
1570
135k
        break;
1571
135k
    }
1572
127k
    case TPlanNodeType::AGGREGATION_NODE: {
1573
127k
        if (tnode.agg_node.grouping_exprs.empty() &&
1574
127k
            descs.get_tuple_descriptor(tnode.agg_node.output_tuple_id)->slots().empty()) {
1575
0
            return Status::InternalError("Illegal aggregate node " + std::to_string(tnode.node_id) +
1576
0
                                         ": group by and output is empty");
1577
0
        }
1578
127k
        bool need_create_cache_op =
1579
127k
                enable_query_cache && tnode.node_id == _params.fragment.query_cache_param.node_id;
1580
127k
        auto create_query_cache_operator = [&](PipelinePtr& new_pipe) {
1581
10
            auto cache_node_id = _params.local_params[0].per_node_scan_ranges.begin()->first;
1582
10
            auto cache_source_id = next_operator_id();
1583
10
            op = std::make_shared<CacheSourceOperatorX>(pool, cache_node_id, cache_source_id,
1584
10
                                                        _params.fragment.query_cache_param);
1585
10
            RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1586
1587
10
            const auto downstream_pipeline_id = cur_pipe->id();
1588
10
            if (!_dag.contains(downstream_pipeline_id)) {
1589
10
                _dag.insert({downstream_pipeline_id, {}});
1590
10
            }
1591
10
            new_pipe = add_pipeline(cur_pipe);
1592
10
            _dag[downstream_pipeline_id].push_back(new_pipe->id());
1593
1594
10
            DataSinkOperatorPtr cache_sink(new CacheSinkOperatorX(
1595
10
                    next_sink_operator_id(), op->node_id(), op->operator_id()));
1596
10
            RETURN_IF_ERROR(new_pipe->set_sink(cache_sink));
1597
10
            return Status::OK();
1598
10
        };
1599
127k
        const bool group_by_limit_opt =
1600
127k
                tnode.agg_node.__isset.agg_sort_info_by_group_key && tnode.limit > 0;
1601
1602
        /// PartitionedAggSourceOperatorX does not support "group by limit opt(#29641)" yet.
1603
        /// If `group_by_limit_opt` is true, then it might not need to spill at all.
1604
127k
        const bool enable_spill = _runtime_state->enable_spill() &&
1605
127k
                                  !tnode.agg_node.grouping_exprs.empty() && !group_by_limit_opt;
1606
127k
        const bool is_streaming_agg = tnode.agg_node.__isset.use_streaming_preaggregation &&
1607
127k
                                      tnode.agg_node.use_streaming_preaggregation &&
1608
127k
                                      !tnode.agg_node.grouping_exprs.empty();
1609
        // TODO: distinct streaming agg does not support spill.
1610
127k
        const bool can_use_distinct_streaming_agg =
1611
127k
                (!enable_spill || is_streaming_agg) && tnode.agg_node.aggregate_functions.empty() &&
1612
127k
                !tnode.agg_node.__isset.agg_sort_info_by_group_key &&
1613
127k
                _params.query_options.__isset.enable_distinct_streaming_aggregation &&
1614
127k
                _params.query_options.enable_distinct_streaming_aggregation;
1615
1616
127k
        if (can_use_distinct_streaming_agg) {
1617
82.6k
            if (need_create_cache_op) {
1618
8
                PipelinePtr new_pipe;
1619
8
                RETURN_IF_ERROR(create_query_cache_operator(new_pipe));
1620
1621
8
                cache_op = op;
1622
8
                op = std::make_shared<DistinctStreamingAggOperatorX>(pool, next_operator_id(),
1623
8
                                                                     tnode, descs);
1624
8
                RETURN_IF_ERROR(new_pipe->add_operator(op, _parallel_instances));
1625
8
                RETURN_IF_ERROR(cur_pipe->operators().front()->set_child(op));
1626
8
                cur_pipe = new_pipe;
1627
82.6k
            } else {
1628
82.6k
                op = std::make_shared<DistinctStreamingAggOperatorX>(pool, next_operator_id(),
1629
82.6k
                                                                     tnode, descs);
1630
82.6k
                RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1631
82.6k
            }
1632
82.6k
        } else if (is_streaming_agg) {
1633
1.87k
            if (need_create_cache_op) {
1634
0
                PipelinePtr new_pipe;
1635
0
                RETURN_IF_ERROR(create_query_cache_operator(new_pipe));
1636
0
                cache_op = op;
1637
0
                op = std::make_shared<StreamingAggOperatorX>(pool, next_operator_id(), tnode,
1638
0
                                                             descs);
1639
0
                RETURN_IF_ERROR(cur_pipe->operators().front()->set_child(op));
1640
0
                RETURN_IF_ERROR(new_pipe->add_operator(op, _parallel_instances));
1641
0
                cur_pipe = new_pipe;
1642
1.87k
            } else {
1643
1.87k
                op = std::make_shared<StreamingAggOperatorX>(pool, next_operator_id(), tnode,
1644
1.87k
                                                             descs);
1645
1.87k
                RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1646
1.87k
            }
1647
43.1k
        } else {
1648
            // create new pipeline to add query cache operator
1649
43.1k
            PipelinePtr new_pipe;
1650
43.1k
            if (need_create_cache_op) {
1651
2
                RETURN_IF_ERROR(create_query_cache_operator(new_pipe));
1652
2
                cache_op = op;
1653
2
            }
1654
1655
43.1k
            if (enable_spill) {
1656
107
                op = std::make_shared<PartitionedAggSourceOperatorX>(pool, tnode,
1657
107
                                                                     next_operator_id(), descs);
1658
43.0k
            } else {
1659
43.0k
                op = std::make_shared<AggSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1660
43.0k
            }
1661
43.1k
            if (need_create_cache_op) {
1662
2
                RETURN_IF_ERROR(cur_pipe->operators().front()->set_child(op));
1663
2
                RETURN_IF_ERROR(new_pipe->add_operator(op, _parallel_instances));
1664
2
                cur_pipe = new_pipe;
1665
43.1k
            } else {
1666
43.1k
                RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1667
43.1k
            }
1668
1669
43.1k
            const auto downstream_pipeline_id = cur_pipe->id();
1670
43.1k
            if (!_dag.contains(downstream_pipeline_id)) {
1671
41.1k
                _dag.insert({downstream_pipeline_id, {}});
1672
41.1k
            }
1673
43.1k
            cur_pipe = add_pipeline(cur_pipe);
1674
43.1k
            _dag[downstream_pipeline_id].push_back(cur_pipe->id());
1675
1676
43.1k
            if (enable_spill) {
1677
107
                sink_ops.push_back(std::make_shared<PartitionedAggSinkOperatorX>(
1678
107
                        pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1679
43.0k
            } else {
1680
43.0k
                sink_ops.push_back(std::make_shared<AggSinkOperatorX>(
1681
43.0k
                        pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1682
43.0k
            }
1683
43.1k
            RETURN_IF_ERROR(cur_pipe->set_sink(sink_ops.back()));
1684
43.1k
            RETURN_IF_ERROR(cur_pipe->sink()->init(tnode, _runtime_state.get()));
1685
43.1k
        }
1686
127k
        break;
1687
127k
    }
1688
127k
    case TPlanNodeType::BUCKETED_AGGREGATION_NODE: {
1689
83
        if (tnode.bucketed_agg_node.grouping_exprs.empty()) {
1690
0
            return Status::InternalError(
1691
0
                    "Bucketed aggregation node {} should not be used without group by keys",
1692
0
                    tnode.node_id);
1693
0
        }
1694
1695
        // Create source operator (goes on the current / downstream pipeline).
1696
83
        op = std::make_shared<BucketedAggSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1697
83
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1698
1699
        // Create a new pipeline for the sink side.
1700
83
        const auto downstream_pipeline_id = cur_pipe->id();
1701
83
        if (!_dag.contains(downstream_pipeline_id)) {
1702
83
            _dag.insert({downstream_pipeline_id, {}});
1703
83
        }
1704
83
        cur_pipe = add_pipeline(cur_pipe);
1705
83
        _dag[downstream_pipeline_id].push_back(cur_pipe->id());
1706
1707
        // Create sink operator.
1708
83
        sink_ops.push_back(std::make_shared<BucketedAggSinkOperatorX>(
1709
83
                pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1710
83
        RETURN_IF_ERROR(cur_pipe->set_sink(sink_ops.back()));
1711
83
        RETURN_IF_ERROR(cur_pipe->sink()->init(tnode, _runtime_state.get()));
1712
1713
        // Pre-register a single shared state for ALL instances so that every
1714
        // sink instance writes its per-instance hash table into the same
1715
        // BucketedAggSharedState and every source instance can merge across
1716
        // all of them.
1717
83
        {
1718
83
            auto shared_state = BucketedAggSharedState::create_shared();
1719
83
            shared_state->id = op->operator_id();
1720
83
            shared_state->related_op_ids.insert(op->operator_id());
1721
1722
795
            for (int i = 0; i < _num_instances; i++) {
1723
712
                auto sink_dep = std::make_shared<Dependency>(op->operator_id(), op->node_id(),
1724
712
                                                             "BUCKETED_AGG_SINK_DEPENDENCY");
1725
712
                sink_dep->set_shared_state(shared_state.get());
1726
712
                shared_state->sink_deps.push_back(sink_dep);
1727
712
            }
1728
83
            shared_state->create_source_dependencies(_num_instances, op->operator_id(),
1729
83
                                                     op->node_id(), "BUCKETED_AGG_SOURCE");
1730
83
            _op_id_to_shared_state.insert(
1731
83
                    {op->operator_id(), {shared_state, shared_state->sink_deps}});
1732
83
        }
1733
83
        break;
1734
83
    }
1735
9.87k
    case TPlanNodeType::HASH_JOIN_NODE: {
1736
9.87k
        const auto is_broadcast_join = tnode.hash_join_node.__isset.is_broadcast_join &&
1737
9.87k
                                       tnode.hash_join_node.is_broadcast_join;
1738
9.87k
        const auto enable_spill = _runtime_state->enable_spill();
1739
9.87k
        if (enable_spill && !is_broadcast_join) {
1740
0
            auto tnode_ = tnode;
1741
0
            tnode_.runtime_filters.clear();
1742
0
            auto inner_probe_operator =
1743
0
                    std::make_shared<HashJoinProbeOperatorX>(pool, tnode_, 0, descs);
1744
1745
            // probe side inner sink operator is used to build hash table on probe side when data is spilled.
1746
            // So here use `tnode_` which has no runtime filters.
1747
0
            auto probe_side_inner_sink_operator =
1748
0
                    std::make_shared<HashJoinBuildSinkOperatorX>(pool, 0, 0, tnode_, descs);
1749
1750
0
            RETURN_IF_ERROR(inner_probe_operator->init(tnode_, _runtime_state.get()));
1751
0
            RETURN_IF_ERROR(probe_side_inner_sink_operator->init(tnode_, _runtime_state.get()));
1752
1753
0
            auto probe_operator = std::make_shared<PartitionedHashJoinProbeOperatorX>(
1754
0
                    pool, tnode_, next_operator_id(), descs);
1755
0
            probe_operator->set_inner_operators(probe_side_inner_sink_operator,
1756
0
                                                inner_probe_operator);
1757
0
            op = std::move(probe_operator);
1758
0
            RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1759
1760
0
            const auto downstream_pipeline_id = cur_pipe->id();
1761
0
            if (!_dag.contains(downstream_pipeline_id)) {
1762
0
                _dag.insert({downstream_pipeline_id, {}});
1763
0
            }
1764
0
            PipelinePtr build_side_pipe = add_pipeline(cur_pipe);
1765
0
            _dag[downstream_pipeline_id].push_back(build_side_pipe->id());
1766
1767
0
            auto inner_sink_operator =
1768
0
                    std::make_shared<HashJoinBuildSinkOperatorX>(pool, 0, 0, tnode, descs);
1769
0
            auto sink_operator = std::make_shared<PartitionedHashJoinSinkOperatorX>(
1770
0
                    pool, next_sink_operator_id(), op->operator_id(), tnode_, descs);
1771
0
            RETURN_IF_ERROR(inner_sink_operator->init(tnode, _runtime_state.get()));
1772
1773
0
            sink_operator->set_inner_operators(inner_sink_operator, inner_probe_operator);
1774
0
            sink_ops.push_back(std::move(sink_operator));
1775
0
            RETURN_IF_ERROR(build_side_pipe->set_sink(sink_ops.back()));
1776
0
            RETURN_IF_ERROR(build_side_pipe->sink()->init(tnode_, _runtime_state.get()));
1777
1778
0
            _pipeline_parent_map.push(op->node_id(), cur_pipe);
1779
0
            _pipeline_parent_map.push(op->node_id(), build_side_pipe);
1780
9.87k
        } else {
1781
9.87k
            op = std::make_shared<HashJoinProbeOperatorX>(pool, tnode, next_operator_id(), descs);
1782
9.87k
            RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1783
1784
9.87k
            const auto downstream_pipeline_id = cur_pipe->id();
1785
9.87k
            if (!_dag.contains(downstream_pipeline_id)) {
1786
8.24k
                _dag.insert({downstream_pipeline_id, {}});
1787
8.24k
            }
1788
9.87k
            PipelinePtr build_side_pipe = add_pipeline(cur_pipe);
1789
9.87k
            _dag[downstream_pipeline_id].push_back(build_side_pipe->id());
1790
1791
9.87k
            sink_ops.push_back(std::make_shared<HashJoinBuildSinkOperatorX>(
1792
9.87k
                    pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1793
9.87k
            RETURN_IF_ERROR(build_side_pipe->set_sink(sink_ops.back()));
1794
9.87k
            RETURN_IF_ERROR(build_side_pipe->sink()->init(tnode, _runtime_state.get()));
1795
1796
9.87k
            _pipeline_parent_map.push(op->node_id(), cur_pipe);
1797
9.87k
            _pipeline_parent_map.push(op->node_id(), build_side_pipe);
1798
9.87k
        }
1799
9.87k
        if (is_broadcast_join && _runtime_state->enable_share_hash_table_for_broadcast_join()) {
1800
4.44k
            std::shared_ptr<HashJoinSharedState> shared_state =
1801
4.44k
                    HashJoinSharedState::create_shared(_num_instances);
1802
22.7k
            for (int i = 0; i < _num_instances; i++) {
1803
18.3k
                auto sink_dep = std::make_shared<Dependency>(op->operator_id(), op->node_id(),
1804
18.3k
                                                             "HASH_JOIN_BUILD_DEPENDENCY");
1805
18.3k
                sink_dep->set_shared_state(shared_state.get());
1806
18.3k
                shared_state->sink_deps.push_back(sink_dep);
1807
18.3k
            }
1808
4.44k
            shared_state->create_source_dependencies(_num_instances, op->operator_id(),
1809
4.44k
                                                     op->node_id(), "HASH_JOIN_PROBE");
1810
4.44k
            _op_id_to_shared_state.insert(
1811
4.44k
                    {op->operator_id(), {shared_state, shared_state->sink_deps}});
1812
4.44k
        }
1813
9.87k
        break;
1814
9.87k
    }
1815
4.10k
    case TPlanNodeType::CROSS_JOIN_NODE: {
1816
4.10k
        op = std::make_shared<NestedLoopJoinProbeOperatorX>(pool, tnode, next_operator_id(), descs);
1817
4.10k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1818
1819
4.10k
        const auto downstream_pipeline_id = cur_pipe->id();
1820
4.10k
        if (!_dag.contains(downstream_pipeline_id)) {
1821
3.85k
            _dag.insert({downstream_pipeline_id, {}});
1822
3.85k
        }
1823
4.10k
        PipelinePtr build_side_pipe = add_pipeline(cur_pipe);
1824
4.10k
        _dag[downstream_pipeline_id].push_back(build_side_pipe->id());
1825
1826
4.10k
        sink_ops.push_back(std::make_shared<NestedLoopJoinBuildSinkOperatorX>(
1827
4.10k
                pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1828
4.10k
        RETURN_IF_ERROR(build_side_pipe->set_sink(sink_ops.back()));
1829
4.10k
        RETURN_IF_ERROR(build_side_pipe->sink()->init(tnode, _runtime_state.get()));
1830
4.10k
        _pipeline_parent_map.push(op->node_id(), cur_pipe);
1831
4.10k
        _pipeline_parent_map.push(op->node_id(), build_side_pipe);
1832
4.10k
        break;
1833
4.10k
    }
1834
52.3k
    case TPlanNodeType::UNION_NODE: {
1835
52.3k
        int child_count = tnode.num_children;
1836
52.3k
        op = std::make_shared<UnionSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1837
52.3k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1838
1839
52.3k
        const auto downstream_pipeline_id = cur_pipe->id();
1840
52.3k
        if (!_dag.contains(downstream_pipeline_id)) {
1841
51.7k
            _dag.insert({downstream_pipeline_id, {}});
1842
51.7k
        }
1843
53.6k
        for (int i = 0; i < child_count; i++) {
1844
1.27k
            PipelinePtr build_side_pipe = add_pipeline(cur_pipe);
1845
1.27k
            _dag[downstream_pipeline_id].push_back(build_side_pipe->id());
1846
1.27k
            sink_ops.push_back(std::make_shared<UnionSinkOperatorX>(
1847
1.27k
                    i, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
1848
1.27k
            RETURN_IF_ERROR(build_side_pipe->set_sink(sink_ops.back()));
1849
1.27k
            RETURN_IF_ERROR(build_side_pipe->sink()->init(tnode, _runtime_state.get()));
1850
            // preset children pipelines. if any pipeline found this as its father, will use the prepared pipeline to build.
1851
1.27k
            _pipeline_parent_map.push(op->node_id(), build_side_pipe);
1852
1.27k
        }
1853
52.3k
        break;
1854
52.3k
    }
1855
52.3k
    case TPlanNodeType::SORT_NODE: {
1856
40.0k
        const auto should_spill = _runtime_state->enable_spill() &&
1857
40.0k
                                  tnode.sort_node.algorithm == TSortAlgorithm::FULL_SORT;
1858
40.0k
        const bool use_local_merge =
1859
40.0k
                tnode.sort_node.__isset.use_local_merge && tnode.sort_node.use_local_merge;
1860
40.0k
        if (should_spill) {
1861
9
            op = std::make_shared<SpillSortSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1862
40.0k
        } else if (use_local_merge) {
1863
37.8k
            op = std::make_shared<LocalMergeSortSourceOperatorX>(pool, tnode, next_operator_id(),
1864
37.8k
                                                                 descs);
1865
37.8k
        } else {
1866
2.23k
            op = std::make_shared<SortSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1867
2.23k
        }
1868
40.0k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1869
1870
40.0k
        const auto downstream_pipeline_id = cur_pipe->id();
1871
40.0k
        if (!_dag.contains(downstream_pipeline_id)) {
1872
40.0k
            _dag.insert({downstream_pipeline_id, {}});
1873
40.0k
        }
1874
40.0k
        cur_pipe = add_pipeline(cur_pipe);
1875
40.0k
        _dag[downstream_pipeline_id].push_back(cur_pipe->id());
1876
1877
40.0k
        if (should_spill) {
1878
9
            sink_ops.push_back(std::make_shared<SpillSortSinkOperatorX>(
1879
9
                    pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1880
40.0k
        } else {
1881
40.0k
            sink_ops.push_back(std::make_shared<SortSinkOperatorX>(
1882
40.0k
                    pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1883
40.0k
        }
1884
40.0k
        RETURN_IF_ERROR(cur_pipe->set_sink(sink_ops.back()));
1885
40.0k
        RETURN_IF_ERROR(cur_pipe->sink()->init(tnode, _runtime_state.get()));
1886
40.0k
        break;
1887
40.0k
    }
1888
40.0k
    case TPlanNodeType::PARTITION_SORT_NODE: {
1889
68
        op = std::make_shared<PartitionSortSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1890
68
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1891
1892
68
        const auto downstream_pipeline_id = cur_pipe->id();
1893
68
        if (!_dag.contains(downstream_pipeline_id)) {
1894
68
            _dag.insert({downstream_pipeline_id, {}});
1895
68
        }
1896
68
        cur_pipe = add_pipeline(cur_pipe);
1897
68
        _dag[downstream_pipeline_id].push_back(cur_pipe->id());
1898
1899
68
        sink_ops.push_back(std::make_shared<PartitionSortSinkOperatorX>(
1900
68
                pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1901
68
        RETURN_IF_ERROR(cur_pipe->set_sink(sink_ops.back()));
1902
68
        RETURN_IF_ERROR(cur_pipe->sink()->init(tnode, _runtime_state.get()));
1903
68
        break;
1904
68
    }
1905
1.80k
    case TPlanNodeType::ANALYTIC_EVAL_NODE: {
1906
1.80k
        op = std::make_shared<AnalyticSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1907
1.80k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1908
1909
1.80k
        const auto downstream_pipeline_id = cur_pipe->id();
1910
1.80k
        if (!_dag.contains(downstream_pipeline_id)) {
1911
1.79k
            _dag.insert({downstream_pipeline_id, {}});
1912
1.79k
        }
1913
1.80k
        cur_pipe = add_pipeline(cur_pipe);
1914
1.80k
        _dag[downstream_pipeline_id].push_back(cur_pipe->id());
1915
1916
1.80k
        sink_ops.push_back(std::make_shared<AnalyticSinkOperatorX>(
1917
1.80k
                pool, next_sink_operator_id(), op->operator_id(), tnode, descs));
1918
1.80k
        RETURN_IF_ERROR(cur_pipe->set_sink(sink_ops.back()));
1919
1.80k
        RETURN_IF_ERROR(cur_pipe->sink()->init(tnode, _runtime_state.get()));
1920
1.80k
        break;
1921
1.80k
    }
1922
1.80k
    case TPlanNodeType::MATERIALIZATION_NODE: {
1923
1.17k
        op = std::make_shared<MaterializationOperator>(pool, tnode, next_operator_id(), descs);
1924
1.17k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1925
1.17k
        break;
1926
1.17k
    }
1927
1.17k
    case TPlanNodeType::INTERSECT_NODE: {
1928
168
        RETURN_IF_ERROR(_build_operators_for_set_operation_node<true>(pool, tnode, descs, op,
1929
168
                                                                      cur_pipe, sink_ops));
1930
168
        break;
1931
168
    }
1932
168
    case TPlanNodeType::EXCEPT_NODE: {
1933
159
        RETURN_IF_ERROR(_build_operators_for_set_operation_node<false>(pool, tnode, descs, op,
1934
159
                                                                       cur_pipe, sink_ops));
1935
159
        break;
1936
159
    }
1937
328
    case TPlanNodeType::REPEAT_NODE: {
1938
328
        op = std::make_shared<RepeatOperatorX>(pool, tnode, next_operator_id(), descs);
1939
328
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1940
328
        break;
1941
328
    }
1942
918
    case TPlanNodeType::TABLE_FUNCTION_NODE: {
1943
918
        op = std::make_shared<TableFunctionOperatorX>(pool, tnode, next_operator_id(), descs);
1944
918
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1945
918
        break;
1946
918
    }
1947
918
    case TPlanNodeType::ASSERT_NUM_ROWS_NODE: {
1948
118
        op = std::make_shared<AssertNumRowsOperatorX>(pool, tnode, next_operator_id(), descs);
1949
118
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1950
118
        break;
1951
118
    }
1952
1.58k
    case TPlanNodeType::EMPTY_SET_NODE: {
1953
1.58k
        op = std::make_shared<EmptySetSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1954
1.58k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1955
1.58k
        break;
1956
1.58k
    }
1957
1.58k
    case TPlanNodeType::DATA_GEN_SCAN_NODE: {
1958
387
        op = std::make_shared<DataGenSourceOperatorX>(pool, tnode, next_operator_id(), descs);
1959
387
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1960
387
        fe_with_old_version = !tnode.__isset.is_serial_operator;
1961
387
        break;
1962
387
    }
1963
1.81k
    case TPlanNodeType::SCHEMA_SCAN_NODE: {
1964
1.81k
        op = std::make_shared<SchemaScanOperatorX>(pool, tnode, next_operator_id(), descs);
1965
1.81k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1966
1.81k
        break;
1967
1.81k
    }
1968
6.31k
    case TPlanNodeType::META_SCAN_NODE: {
1969
6.31k
        op = std::make_shared<MetaScanOperatorX>(pool, tnode, next_operator_id(), descs);
1970
6.31k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1971
6.31k
        break;
1972
6.31k
    }
1973
6.31k
    case TPlanNodeType::SELECT_NODE: {
1974
1.55k
        op = std::make_shared<SelectOperatorX>(pool, tnode, next_operator_id(), descs);
1975
1.55k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1976
1.55k
        break;
1977
1.55k
    }
1978
1.55k
    case TPlanNodeType::REC_CTE_NODE: {
1979
163
        op = std::make_shared<RecCTESourceOperatorX>(pool, tnode, next_operator_id(), descs);
1980
163
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
1981
1982
163
        const auto downstream_pipeline_id = cur_pipe->id();
1983
163
        if (!_dag.contains(downstream_pipeline_id)) {
1984
159
            _dag.insert({downstream_pipeline_id, {}});
1985
159
        }
1986
1987
163
        PipelinePtr anchor_side_pipe = add_pipeline(cur_pipe);
1988
163
        _dag[downstream_pipeline_id].push_back(anchor_side_pipe->id());
1989
1990
163
        DataSinkOperatorPtr anchor_sink;
1991
163
        anchor_sink = std::make_shared<RecCTEAnchorSinkOperatorX>(next_sink_operator_id(),
1992
163
                                                                  op->operator_id(), tnode, descs);
1993
163
        RETURN_IF_ERROR(anchor_side_pipe->set_sink(anchor_sink));
1994
163
        RETURN_IF_ERROR(anchor_side_pipe->sink()->init(tnode, _runtime_state.get()));
1995
163
        _pipeline_parent_map.push(op->node_id(), anchor_side_pipe);
1996
1997
163
        PipelinePtr rec_side_pipe = add_pipeline(cur_pipe);
1998
163
        _dag[downstream_pipeline_id].push_back(rec_side_pipe->id());
1999
2000
163
        DataSinkOperatorPtr rec_sink;
2001
163
        rec_sink = std::make_shared<RecCTESinkOperatorX>(next_sink_operator_id(), op->operator_id(),
2002
163
                                                         tnode, descs);
2003
163
        RETURN_IF_ERROR(rec_side_pipe->set_sink(rec_sink));
2004
163
        RETURN_IF_ERROR(rec_side_pipe->sink()->init(tnode, _runtime_state.get()));
2005
163
        _pipeline_parent_map.push(op->node_id(), rec_side_pipe);
2006
2007
163
        break;
2008
163
    }
2009
2.18k
    case TPlanNodeType::REC_CTE_SCAN_NODE: {
2010
2.18k
        op = std::make_shared<RecCTEScanOperatorX>(pool, tnode, next_operator_id(), descs);
2011
2.18k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
2012
2.18k
        break;
2013
2.18k
    }
2014
109k
    case TPlanNodeType::LOCAL_EXCHANGE_NODE: {
2015
109k
        op = std::make_shared<LocalExchangeSourceOperatorX>(pool, tnode, next_operator_id(), descs);
2016
        // The downstream pipeline (containing LocalExchangeSource) must have
2017
        // _num_instances tasks — matching BE-native _inherit_pipeline_properties
2018
        // which sets pipe_with_source.set_num_tasks(_num_instances).
2019
        // Without this, when the parent pipeline was reduced by a serial operator
2020
        // (e.g., serial Exchange with use_serial_exchange=true, or UNPARTITIONED
2021
        // Exchange), the downstream inherits the reduced num_tasks via
2022
        // add_pipeline(parent).  The deferred exchanger creates _num_instances
2023
        // channels but only fewer source tasks initialize mem_counters — the
2024
        // sink round-robins to all channels and crashes on uninitialized ones.
2025
109k
        RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
2026
        // Restore downstream pipeline's num_tasks (mirroring _inherit_pipeline_properties:
2027
        // downstream keeps _num_instances, upstream gets the serial/reduced count)
2028
109k
        cur_pipe->set_num_tasks(_num_instances);
2029
2030
109k
        const auto downstream_pipeline_id = cur_pipe->id();
2031
109k
        if (!_dag.contains(downstream_pipeline_id)) {
2032
104k
            _dag.insert({downstream_pipeline_id, {}});
2033
104k
        }
2034
109k
        cur_pipe = add_pipeline(cur_pipe);
2035
        // If this local exchange was inserted because of a serial scan (is_serial_operator),
2036
        // the upstream pipeline (cur_pipe) should have num_tasks=1 (only 1 scan task).
2037
        // We set this now so the exchanger is created with the correct sender count.
2038
        // Child operators added later (serial scan) will also set num_tasks=1, which is
2039
        // consistent with this.
2040
109k
        if (op->is_serial_operator() && _parallel_instances > 0) {
2041
0
            cur_pipe->set_num_tasks(_parallel_instances);
2042
0
        }
2043
109k
        _dag[downstream_pipeline_id].push_back(cur_pipe->id());
2044
109k
        int num_partitions = 0;
2045
109k
        std::map<int, int> shuffle_id_to_instance_idx;
2046
109k
        auto partition_type = tnode.local_exchange_node.partition_type;
2047
109k
        switch (partition_type) {
2048
523
        case TLocalPartitionType::BUCKET_HASH_SHUFFLE:
2049
523
            num_partitions = _params.num_buckets;
2050
523
            shuffle_id_to_instance_idx = _params.bucket_seq_to_instance_idx;
2051
523
            break;
2052
22.6k
        case TLocalPartitionType::LOCAL_EXECUTION_HASH_SHUFFLE:
2053
211k
            for (int i = 0; i < _num_instances; i++) {
2054
188k
                shuffle_id_to_instance_idx[i] = i;
2055
188k
            }
2056
22.6k
            num_partitions = _num_instances;
2057
22.6k
            break;
2058
6
        case TLocalPartitionType::GLOBAL_EXECUTION_HASH_SHUFFLE:
2059
6
            num_partitions = _total_instances;
2060
6
            shuffle_id_to_instance_idx = _params.shuffle_idx_to_instance_idx;
2061
6
            break;
2062
85.7k
        default:
2063
85.7k
            break;
2064
109k
        }
2065
108k
        auto local_exchange_id = op->operator_id();
2066
108k
        auto sink_id = next_sink_operator_id();
2067
108k
        DataSinkOperatorPtr sink = std::make_shared<LocalExchangeSinkOperatorX>(
2068
108k
                sink_id, local_exchange_id, tnode, num_partitions, shuffle_id_to_instance_idx);
2069
108k
        sink_ops.push_back(sink);
2070
108k
        RETURN_IF_ERROR(cur_pipe->set_sink(sink));
2071
108k
        RETURN_IF_ERROR(cur_pipe->sink()->init(tnode, _runtime_state.get()));
2072
2073
        // For FE-planned local exchange, we need to:
2074
        // 1. Initialize the partitioner for hash shuffle types
2075
        // 2. Defer exchanger creation until after the full plan tree is built
2076
        //    (child operators like serial ExchangeNode may change cur_pipe->num_tasks())
2077
        // 3. Register shared state so pipeline tasks can find it
2078
108k
        RETURN_IF_ERROR(static_cast<LocalExchangeSinkOperatorX*>(cur_pipe->sink())
2079
108k
                                ->init_partitioner(_runtime_state.get()));
2080
2081
108k
        int free_blocks_limit =
2082
108k
                _runtime_state->query_options().__isset.local_exchange_free_blocks_limit
2083
108k
                        ? cast_set<int>(
2084
108k
                                  _runtime_state->query_options().local_exchange_free_blocks_limit)
2085
108k
                        : 0;
2086
108k
        auto shared_state = LocalExchangeSharedState::create_shared(_num_instances);
2087
108k
        shared_state->create_source_dependencies(_num_instances, local_exchange_id,
2088
108k
                                                 local_exchange_id, "LOCAL_EXCHANGE_OPERATOR");
2089
108k
        shared_state->create_sink_dependency(sink_id, local_exchange_id, "LOCAL_EXCHANGE_SINK");
2090
108k
        _op_id_to_shared_state.insert({local_exchange_id, {shared_state, shared_state->sink_deps}});
2091
        // Defer exchanger creation: sender count depends on final upstream num_tasks
2092
108k
        _deferred_exchangers.push_back({shared_state, cur_pipe, partition_type, num_partitions,
2093
108k
                                        free_blocks_limit, local_exchange_id, sink_id});
2094
108k
        break;
2095
108k
    }
2096
0
    default:
2097
0
        return Status::InternalError("Unsupported exec type in pipeline: {}",
2098
0
                                     print_plan_node_type(tnode.node_type));
2099
704k
    }
2100
702k
    if (_params.__isset.parallel_instances && fe_with_old_version) {
2101
0
        cur_pipe->set_num_tasks(_params.parallel_instances);
2102
0
        op->set_serial_operator();
2103
0
    }
2104
2105
702k
    return Status::OK();
2106
704k
}
2107
// NOLINTEND(readability-function-cognitive-complexity)
2108
// NOLINTEND(readability-function-size)
2109
2110
template <bool is_intersect>
2111
Status PipelineFragmentContext::_build_operators_for_set_operation_node(
2112
        ObjectPool* pool, const TPlanNode& tnode, const DescriptorTbl& descs, OperatorPtr& op,
2113
327
        PipelinePtr& cur_pipe, std::vector<DataSinkOperatorPtr>& sink_ops) {
2114
327
    op.reset(new SetSourceOperatorX<is_intersect>(pool, tnode, next_operator_id(), descs));
2115
327
    RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
2116
2117
327
    const auto downstream_pipeline_id = cur_pipe->id();
2118
327
    if (!_dag.contains(downstream_pipeline_id)) {
2119
312
        _dag.insert({downstream_pipeline_id, {}});
2120
312
    }
2121
2122
1.07k
    for (int child_id = 0; child_id < tnode.num_children; child_id++) {
2123
750
        PipelinePtr probe_side_pipe = add_pipeline(cur_pipe);
2124
750
        _dag[downstream_pipeline_id].push_back(probe_side_pipe->id());
2125
2126
750
        if (child_id == 0) {
2127
327
            sink_ops.push_back(std::make_shared<SetSinkOperatorX<is_intersect>>(
2128
327
                    child_id, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
2129
423
        } else {
2130
423
            sink_ops.push_back(std::make_shared<SetProbeSinkOperatorX<is_intersect>>(
2131
423
                    child_id, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
2132
423
        }
2133
750
        RETURN_IF_ERROR(probe_side_pipe->set_sink(sink_ops.back()));
2134
750
        RETURN_IF_ERROR(probe_side_pipe->sink()->init(tnode, _runtime_state.get()));
2135
        // prepare children pipelines. if any pipeline found this as its father, will use the prepared pipeline to build.
2136
750
        _pipeline_parent_map.push(op->node_id(), probe_side_pipe);
2137
750
    }
2138
2139
327
    return Status::OK();
2140
327
}
_ZN5doris23PipelineFragmentContext39_build_operators_for_set_operation_nodeILb1EEENS_6StatusEPNS_10ObjectPoolERKNS_9TPlanNodeERKNS_13DescriptorTblERSt10shared_ptrINS_13OperatorXBaseEERSB_INS_8PipelineEERSt6vectorISB_INS_21DataSinkOperatorXBaseEESaISK_EE
Line
Count
Source
2113
168
        PipelinePtr& cur_pipe, std::vector<DataSinkOperatorPtr>& sink_ops) {
2114
168
    op.reset(new SetSourceOperatorX<is_intersect>(pool, tnode, next_operator_id(), descs));
2115
168
    RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
2116
2117
168
    const auto downstream_pipeline_id = cur_pipe->id();
2118
168
    if (!_dag.contains(downstream_pipeline_id)) {
2119
159
        _dag.insert({downstream_pipeline_id, {}});
2120
159
    }
2121
2122
585
    for (int child_id = 0; child_id < tnode.num_children; child_id++) {
2123
417
        PipelinePtr probe_side_pipe = add_pipeline(cur_pipe);
2124
417
        _dag[downstream_pipeline_id].push_back(probe_side_pipe->id());
2125
2126
417
        if (child_id == 0) {
2127
168
            sink_ops.push_back(std::make_shared<SetSinkOperatorX<is_intersect>>(
2128
168
                    child_id, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
2129
249
        } else {
2130
249
            sink_ops.push_back(std::make_shared<SetProbeSinkOperatorX<is_intersect>>(
2131
249
                    child_id, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
2132
249
        }
2133
417
        RETURN_IF_ERROR(probe_side_pipe->set_sink(sink_ops.back()));
2134
417
        RETURN_IF_ERROR(probe_side_pipe->sink()->init(tnode, _runtime_state.get()));
2135
        // prepare children pipelines. if any pipeline found this as its father, will use the prepared pipeline to build.
2136
417
        _pipeline_parent_map.push(op->node_id(), probe_side_pipe);
2137
417
    }
2138
2139
168
    return Status::OK();
2140
168
}
_ZN5doris23PipelineFragmentContext39_build_operators_for_set_operation_nodeILb0EEENS_6StatusEPNS_10ObjectPoolERKNS_9TPlanNodeERKNS_13DescriptorTblERSt10shared_ptrINS_13OperatorXBaseEERSB_INS_8PipelineEERSt6vectorISB_INS_21DataSinkOperatorXBaseEESaISK_EE
Line
Count
Source
2113
159
        PipelinePtr& cur_pipe, std::vector<DataSinkOperatorPtr>& sink_ops) {
2114
159
    op.reset(new SetSourceOperatorX<is_intersect>(pool, tnode, next_operator_id(), descs));
2115
159
    RETURN_IF_ERROR(cur_pipe->add_operator(op, _parallel_instances));
2116
2117
159
    const auto downstream_pipeline_id = cur_pipe->id();
2118
159
    if (!_dag.contains(downstream_pipeline_id)) {
2119
153
        _dag.insert({downstream_pipeline_id, {}});
2120
153
    }
2121
2122
492
    for (int child_id = 0; child_id < tnode.num_children; child_id++) {
2123
333
        PipelinePtr probe_side_pipe = add_pipeline(cur_pipe);
2124
333
        _dag[downstream_pipeline_id].push_back(probe_side_pipe->id());
2125
2126
333
        if (child_id == 0) {
2127
159
            sink_ops.push_back(std::make_shared<SetSinkOperatorX<is_intersect>>(
2128
159
                    child_id, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
2129
174
        } else {
2130
174
            sink_ops.push_back(std::make_shared<SetProbeSinkOperatorX<is_intersect>>(
2131
174
                    child_id, next_sink_operator_id(), op->operator_id(), pool, tnode, descs));
2132
174
        }
2133
333
        RETURN_IF_ERROR(probe_side_pipe->set_sink(sink_ops.back()));
2134
333
        RETURN_IF_ERROR(probe_side_pipe->sink()->init(tnode, _runtime_state.get()));
2135
        // prepare children pipelines. if any pipeline found this as its father, will use the prepared pipeline to build.
2136
333
        _pipeline_parent_map.push(op->node_id(), probe_side_pipe);
2137
333
    }
2138
2139
159
    return Status::OK();
2140
159
}
2141
2142
388k
Status PipelineFragmentContext::submit() {
2143
388k
    if (_submitted) {
2144
0
        return Status::InternalError("submitted");
2145
0
    }
2146
388k
    _submitted = true;
2147
2148
388k
    int submit_tasks = 0;
2149
388k
    Status st;
2150
388k
    auto* scheduler = _query_ctx->get_pipe_exec_scheduler();
2151
1.14M
    for (auto& task : _tasks) {
2152
1.91M
        for (auto& t : task) {
2153
1.91M
            st = scheduler->submit(t.first);
2154
1.91M
            DBUG_EXECUTE_IF("PipelineFragmentContext.submit.failed",
2155
1.91M
                            { st = Status::Aborted("PipelineFragmentContext.submit.failed"); });
2156
1.91M
            if (!st) {
2157
0
                cancel(Status::InternalError("submit context to executor fail"));
2158
0
                std::lock_guard<std::mutex> l(_task_mutex);
2159
0
                _total_tasks = submit_tasks;
2160
0
                break;
2161
0
            }
2162
1.91M
            submit_tasks++;
2163
1.91M
        }
2164
1.14M
    }
2165
388k
    if (!st.ok()) {
2166
0
        bool need_remove = false;
2167
0
        {
2168
0
            std::lock_guard<std::mutex> l(_task_mutex);
2169
0
            if (_closed_tasks >= _total_tasks) {
2170
0
                need_remove = _close_fragment_instance();
2171
0
            }
2172
0
        }
2173
        // Call remove_pipeline_context() outside _task_mutex to avoid ABBA deadlock.
2174
0
        if (need_remove) {
2175
0
            _exec_env->fragment_mgr()->remove_pipeline_context({_query_id, _fragment_id});
2176
0
        }
2177
0
        return Status::InternalError("Submit pipeline failed. err = {}, BE: {}", st.to_string(),
2178
0
                                     BackendOptions::get_localhost());
2179
388k
    } else {
2180
388k
        return st;
2181
388k
    }
2182
388k
}
2183
2184
0
void PipelineFragmentContext::print_profile(const std::string& extra_info) {
2185
0
    if (_runtime_state->enable_profile()) {
2186
0
        std::stringstream ss;
2187
0
        for (auto runtime_profile_ptr : _runtime_state->pipeline_id_to_profile()) {
2188
0
            runtime_profile_ptr->pretty_print(&ss);
2189
0
        }
2190
2191
0
        if (_runtime_state->load_channel_profile()) {
2192
0
            _runtime_state->load_channel_profile()->pretty_print(&ss);
2193
0
        }
2194
2195
0
        auto profile_str =
2196
0
                fmt::format("Query {} fragment {} {}, profile, {}", print_id(this->_query_id),
2197
0
                            this->_fragment_id, extra_info, ss.str());
2198
0
        LOG_LONG_STRING(INFO, profile_str);
2199
0
    }
2200
0
}
2201
// If all pipeline tasks binded to the fragment instance are finished, then we could
2202
// close the fragment instance.
2203
// Returns true if the caller should call remove_pipeline_context() **after** releasing
2204
// _task_mutex. We must not call remove_pipeline_context() here because it acquires
2205
// _pipeline_map's shard lock, and this function is called while _task_mutex is held.
2206
// Acquiring _pipeline_map while holding _task_mutex creates an ABBA deadlock with
2207
// dump_pipeline_tasks(), which acquires _pipeline_map first and then _task_mutex
2208
// (via debug_string()).
2209
391k
bool PipelineFragmentContext::_close_fragment_instance() {
2210
391k
    if (_is_fragment_instance_closed) {
2211
0
        return false;
2212
0
    }
2213
391k
    Defer defer_op {[&]() { _is_fragment_instance_closed = true; }};
2214
391k
    _fragment_level_profile->total_time_counter()->update(_fragment_watcher.elapsed_time());
2215
391k
    if (!_need_notify_close) {
2216
388k
        auto st = send_report(true);
2217
388k
        if (!st) {
2218
0
            LOG(WARNING) << fmt::format("Failed to send report for query {}, fragment {}: {}",
2219
0
                                        print_id(_query_id), _fragment_id, st.to_string());
2220
0
        }
2221
388k
    }
2222
    // Print profile content in info log is a tempoeray solution for stream load and external_connector.
2223
    // Since stream load does not have someting like coordinator on FE, so
2224
    // backend can not report profile to FE, ant its profile can not be shown
2225
    // in the same way with other query. So we print the profile content to info log.
2226
2227
391k
    if (_runtime_state->enable_profile() &&
2228
391k
        (_query_ctx->get_query_source() == QuerySource::STREAM_LOAD ||
2229
2.95k
         _query_ctx->get_query_source() == QuerySource::EXTERNAL_CONNECTOR ||
2230
2.95k
         _query_ctx->get_query_source() == QuerySource::GROUP_COMMIT_LOAD)) {
2231
0
        std::stringstream ss;
2232
        // Compute the _local_time_percent before pretty_print the runtime_profile
2233
        // Before add this operation, the print out like that:
2234
        // UNION_NODE (id=0):(Active: 56.720us, non-child: 00.00%)
2235
        // After add the operation, the print out like that:
2236
        // UNION_NODE (id=0):(Active: 56.720us, non-child: 82.53%)
2237
        // We can easily know the exec node execute time without child time consumed.
2238
0
        for (auto runtime_profile_ptr : _runtime_state->pipeline_id_to_profile()) {
2239
0
            runtime_profile_ptr->pretty_print(&ss);
2240
0
        }
2241
2242
0
        if (_runtime_state->load_channel_profile()) {
2243
0
            _runtime_state->load_channel_profile()->pretty_print(&ss);
2244
0
        }
2245
2246
0
        LOG_INFO("Query {} fragment {} profile:\n {}", print_id(_query_id), _fragment_id, ss.str());
2247
0
    }
2248
2249
391k
    if (_query_ctx->enable_profile()) {
2250
2.95k
        _query_ctx->add_fragment_profile(_fragment_id, collect_realtime_profile(),
2251
2.95k
                                         collect_realtime_load_channel_profile());
2252
2.95k
    }
2253
2254
    // Return whether the caller needs to remove from the pipeline map.
2255
    // The caller must do this after releasing _task_mutex.
2256
391k
    return !_need_notify_close;
2257
391k
}
2258
2259
1.90M
void PipelineFragmentContext::decrement_running_task(PipelineId pipeline_id) {
2260
    // If all tasks of this pipeline has been closed, upstream tasks is never needed, and we just make those runnable here
2261
1.90M
    DCHECK(_pip_id_to_pipeline.contains(pipeline_id));
2262
1.90M
    if (_pip_id_to_pipeline[pipeline_id]->close_task()) {
2263
607k
        if (_dag.contains(pipeline_id)) {
2264
257k
            for (auto dep : _dag[pipeline_id]) {
2265
216k
                _pip_id_to_pipeline[dep]->make_all_runnable(pipeline_id);
2266
216k
            }
2267
257k
        }
2268
607k
    }
2269
1.90M
    bool need_remove = false;
2270
1.90M
    {
2271
1.90M
        std::lock_guard<std::mutex> l(_task_mutex);
2272
1.90M
        ++_closed_tasks;
2273
        // Update query-level finished task progress in real time.
2274
1.90M
        _query_ctx->inc_finished_task_num();
2275
1.90M
        if (_closed_tasks >= _total_tasks) {
2276
391k
            need_remove = _close_fragment_instance();
2277
391k
        }
2278
1.90M
    }
2279
    // Call remove_pipeline_context() outside _task_mutex to avoid ABBA deadlock.
2280
1.90M
    if (need_remove) {
2281
388k
        _exec_env->fragment_mgr()->remove_pipeline_context({_query_id, _fragment_id});
2282
388k
    }
2283
1.90M
}
2284
2285
50.1k
std::string PipelineFragmentContext::get_load_error_url() {
2286
50.1k
    if (const auto& str = _runtime_state->get_error_log_file_path(); !str.empty()) {
2287
0
        return to_load_error_http_path(str);
2288
0
    }
2289
128k
    for (auto& tasks : _tasks) {
2290
210k
        for (auto& task : tasks) {
2291
210k
            if (const auto& str = task.second->get_error_log_file_path(); !str.empty()) {
2292
186
                return to_load_error_http_path(str);
2293
186
            }
2294
210k
        }
2295
128k
    }
2296
49.9k
    return "";
2297
50.1k
}
2298
2299
50.1k
std::string PipelineFragmentContext::get_first_error_msg() {
2300
50.1k
    if (const auto& str = _runtime_state->get_first_error_msg(); !str.empty()) {
2301
0
        return str;
2302
0
    }
2303
128k
    for (auto& tasks : _tasks) {
2304
210k
        for (auto& task : tasks) {
2305
210k
            if (const auto& str = task.second->get_first_error_msg(); !str.empty()) {
2306
186
                return str;
2307
186
            }
2308
210k
        }
2309
128k
    }
2310
49.9k
    return "";
2311
50.1k
}
2312
2313
0
std::string PipelineFragmentContext::_to_http_path(const std::string& file_name) const {
2314
0
    std::stringstream url;
2315
0
    url << "http://" << BackendOptions::get_localhost() << ":" << config::webserver_port
2316
0
        << "/api/_download_load?"
2317
0
        << "token=" << _exec_env->token() << "&file=" << file_name;
2318
0
    return url.str();
2319
0
}
2320
2321
44.0k
void PipelineFragmentContext::_coordinator_callback(const ReportStatusRequest& req) {
2322
44.0k
    DBUG_EXECUTE_IF("FragmentMgr::coordinator_callback.report_delay", {
2323
44.0k
        int random_seconds = req.status.is<ErrorCode::DATA_QUALITY_ERROR>() ? 8 : 2;
2324
44.0k
        LOG_INFO("sleep : ").tag("time", random_seconds).tag("query_id", print_id(req.query_id));
2325
44.0k
        std::this_thread::sleep_for(std::chrono::seconds(random_seconds));
2326
44.0k
        LOG_INFO("sleep done").tag("query_id", print_id(req.query_id));
2327
44.0k
    });
2328
2329
44.0k
    DCHECK(req.status.ok() || req.done); // if !status.ok() => done
2330
44.0k
    if (req.coord_addr.hostname == "external") {
2331
        // External query (flink/spark read tablets) not need to report to FE.
2332
0
        return;
2333
0
    }
2334
44.0k
    int callback_retries = 10;
2335
44.0k
    const int sleep_ms = 1000;
2336
44.0k
    Status exec_status = req.status;
2337
44.0k
    Status coord_status;
2338
44.0k
    std::unique_ptr<FrontendServiceConnection> coord = nullptr;
2339
44.0k
    do {
2340
44.0k
        coord = std::make_unique<FrontendServiceConnection>(_exec_env->frontend_client_cache(),
2341
44.0k
                                                            req.coord_addr, &coord_status);
2342
44.0k
        if (!coord_status.ok()) {
2343
0
            std::this_thread::sleep_for(std::chrono::milliseconds(sleep_ms));
2344
0
        }
2345
44.0k
    } while (!coord_status.ok() && callback_retries-- > 0);
2346
2347
44.0k
    if (!coord_status.ok()) {
2348
0
        UniqueId uid(req.query_id.hi, req.query_id.lo);
2349
0
        static_cast<void>(req.cancel_fn(Status::InternalError(
2350
0
                "query_id: {}, couldn't get a client for {}, reason is {}", uid.to_string(),
2351
0
                PrintThriftNetworkAddress(req.coord_addr), coord_status.to_string())));
2352
0
        return;
2353
0
    }
2354
2355
44.0k
    TReportExecStatusParams params;
2356
44.0k
    params.protocol_version = FrontendServiceVersion::V1;
2357
44.0k
    params.__set_query_id(req.query_id);
2358
44.0k
    params.__set_backend_num(req.backend_num);
2359
44.0k
    params.__set_fragment_instance_id(req.fragment_instance_id);
2360
44.0k
    params.__set_fragment_id(req.fragment_id);
2361
44.0k
    params.__set_status(exec_status.to_thrift());
2362
44.0k
    params.__set_done(req.done);
2363
44.0k
    params.__set_query_type(req.runtime_state->query_type());
2364
44.0k
    params.__isset.profile = false;
2365
2366
44.0k
    DCHECK(req.runtime_state != nullptr);
2367
2368
44.0k
    if (req.runtime_state->query_type() == TQueryType::LOAD) {
2369
39.1k
        params.__set_loaded_rows(req.runtime_state->num_rows_load_total());
2370
39.1k
        params.__set_loaded_bytes(req.runtime_state->num_bytes_load_total());
2371
39.1k
    } else {
2372
4.91k
        DCHECK(!req.runtime_states.empty());
2373
4.91k
        if (!req.runtime_state->output_files().empty()) {
2374
0
            params.__isset.delta_urls = true;
2375
0
            for (auto& it : req.runtime_state->output_files()) {
2376
0
                params.delta_urls.push_back(_to_http_path(it));
2377
0
            }
2378
0
        }
2379
4.91k
        if (!params.delta_urls.empty()) {
2380
0
            params.__isset.delta_urls = true;
2381
0
        }
2382
4.91k
    }
2383
2384
44.0k
    static std::string s_dpp_normal_all = "dpp.norm.ALL";
2385
44.0k
    static std::string s_dpp_abnormal_all = "dpp.abnorm.ALL";
2386
44.0k
    static std::string s_unselected_rows = "unselected.rows";
2387
44.0k
    int64_t num_rows_load_success = 0;
2388
44.0k
    int64_t num_rows_load_filtered = 0;
2389
44.0k
    int64_t num_rows_load_unselected = 0;
2390
44.0k
    if (req.runtime_state->num_rows_load_total() > 0 ||
2391
44.0k
        req.runtime_state->num_rows_load_filtered() > 0 ||
2392
44.0k
        req.runtime_state->num_finished_range() > 0) {
2393
0
        params.__isset.load_counters = true;
2394
2395
0
        num_rows_load_success = req.runtime_state->num_rows_load_success();
2396
0
        num_rows_load_filtered = req.runtime_state->num_rows_load_filtered();
2397
0
        num_rows_load_unselected = req.runtime_state->num_rows_load_unselected();
2398
0
        params.__isset.fragment_instance_reports = true;
2399
0
        TFragmentInstanceReport t;
2400
0
        t.__set_fragment_instance_id(req.runtime_state->fragment_instance_id());
2401
0
        t.__set_num_finished_range(cast_set<int>(req.runtime_state->num_finished_range()));
2402
0
        t.__set_loaded_rows(req.runtime_state->num_rows_load_total());
2403
0
        t.__set_loaded_bytes(req.runtime_state->num_bytes_load_total());
2404
0
        params.fragment_instance_reports.push_back(t);
2405
44.0k
    } else if (!req.runtime_states.empty()) {
2406
133k
        for (auto* rs : req.runtime_states) {
2407
133k
            if (rs->num_rows_load_total() > 0 || rs->num_rows_load_filtered() > 0 ||
2408
133k
                rs->num_finished_range() > 0) {
2409
34.8k
                params.__isset.load_counters = true;
2410
34.8k
                num_rows_load_success += rs->num_rows_load_success();
2411
34.8k
                num_rows_load_filtered += rs->num_rows_load_filtered();
2412
34.8k
                num_rows_load_unselected += rs->num_rows_load_unselected();
2413
34.8k
                params.__isset.fragment_instance_reports = true;
2414
34.8k
                TFragmentInstanceReport t;
2415
34.8k
                t.__set_fragment_instance_id(rs->fragment_instance_id());
2416
34.8k
                t.__set_num_finished_range(cast_set<int>(rs->num_finished_range()));
2417
34.8k
                t.__set_loaded_rows(rs->num_rows_load_total());
2418
34.8k
                t.__set_loaded_bytes(rs->num_bytes_load_total());
2419
34.8k
                params.fragment_instance_reports.push_back(t);
2420
34.8k
            }
2421
133k
        }
2422
44.0k
    }
2423
44.0k
    params.load_counters.emplace(s_dpp_normal_all, std::to_string(num_rows_load_success));
2424
44.0k
    params.load_counters.emplace(s_dpp_abnormal_all, std::to_string(num_rows_load_filtered));
2425
44.0k
    params.load_counters.emplace(s_unselected_rows, std::to_string(num_rows_load_unselected));
2426
2427
44.0k
    if (!req.load_error_url.empty()) {
2428
171
        params.__set_tracking_url(req.load_error_url);
2429
171
    }
2430
44.0k
    if (!req.first_error_msg.empty()) {
2431
171
        params.__set_first_error_msg(req.first_error_msg);
2432
171
    }
2433
133k
    for (auto* rs : req.runtime_states) {
2434
133k
        if (rs->wal_id() > 0) {
2435
104
            params.__set_txn_id(rs->wal_id());
2436
104
            params.__set_label(rs->import_label());
2437
104
        }
2438
133k
    }
2439
44.0k
    if (!req.runtime_state->export_output_files().empty()) {
2440
0
        params.__isset.export_files = true;
2441
0
        params.export_files = req.runtime_state->export_output_files();
2442
44.0k
    } else if (!req.runtime_states.empty()) {
2443
133k
        for (auto* rs : req.runtime_states) {
2444
133k
            if (!rs->export_output_files().empty()) {
2445
0
                params.__isset.export_files = true;
2446
0
                params.export_files.insert(params.export_files.end(),
2447
0
                                           rs->export_output_files().begin(),
2448
0
                                           rs->export_output_files().end());
2449
0
            }
2450
133k
        }
2451
44.0k
    }
2452
44.0k
    if (auto tci = req.runtime_state->tablet_commit_infos(); !tci.empty()) {
2453
0
        params.__isset.commitInfos = true;
2454
0
        params.commitInfos.insert(params.commitInfos.end(), tci.begin(), tci.end());
2455
44.0k
    } else if (!req.runtime_states.empty()) {
2456
133k
        for (auto* rs : req.runtime_states) {
2457
133k
            if (auto rs_tci = rs->tablet_commit_infos(); !rs_tci.empty()) {
2458
26.7k
                params.__isset.commitInfos = true;
2459
26.7k
                params.commitInfos.insert(params.commitInfos.end(), rs_tci.begin(), rs_tci.end());
2460
26.7k
            }
2461
133k
        }
2462
44.0k
    }
2463
44.0k
    if (auto eti = req.runtime_state->error_tablet_infos(); !eti.empty()) {
2464
0
        params.__isset.errorTabletInfos = true;
2465
0
        params.errorTabletInfos.insert(params.errorTabletInfos.end(), eti.begin(), eti.end());
2466
44.0k
    } else if (!req.runtime_states.empty()) {
2467
133k
        for (auto* rs : req.runtime_states) {
2468
133k
            if (auto rs_eti = rs->error_tablet_infos(); !rs_eti.empty()) {
2469
0
                params.__isset.errorTabletInfos = true;
2470
0
                params.errorTabletInfos.insert(params.errorTabletInfos.end(), rs_eti.begin(),
2471
0
                                               rs_eti.end());
2472
0
            }
2473
133k
        }
2474
44.0k
    }
2475
44.0k
    if (auto hpu = req.runtime_state->hive_partition_updates(); !hpu.empty()) {
2476
0
        params.__isset.hive_partition_updates = true;
2477
0
        params.hive_partition_updates.insert(params.hive_partition_updates.end(), hpu.begin(),
2478
0
                                             hpu.end());
2479
44.0k
    } else if (!req.runtime_states.empty()) {
2480
133k
        for (auto* rs : req.runtime_states) {
2481
133k
            if (auto rs_hpu = rs->hive_partition_updates(); !rs_hpu.empty()) {
2482
1.06k
                params.__isset.hive_partition_updates = true;
2483
1.06k
                params.hive_partition_updates.insert(params.hive_partition_updates.end(),
2484
1.06k
                                                     rs_hpu.begin(), rs_hpu.end());
2485
1.06k
            }
2486
133k
        }
2487
44.0k
    }
2488
44.0k
    if (auto icd = req.runtime_state->iceberg_commit_datas(); !icd.empty()) {
2489
0
        params.__isset.iceberg_commit_datas = true;
2490
0
        params.iceberg_commit_datas.insert(params.iceberg_commit_datas.end(), icd.begin(),
2491
0
                                           icd.end());
2492
44.0k
    } else if (!req.runtime_states.empty()) {
2493
133k
        for (auto* rs : req.runtime_states) {
2494
133k
            if (auto rs_icd = rs->iceberg_commit_datas(); !rs_icd.empty()) {
2495
1.04k
                params.__isset.iceberg_commit_datas = true;
2496
1.04k
                params.iceberg_commit_datas.insert(params.iceberg_commit_datas.end(),
2497
1.04k
                                                   rs_icd.begin(), rs_icd.end());
2498
1.04k
            }
2499
133k
        }
2500
44.0k
    }
2501
2502
44.0k
    if (auto mcd = req.runtime_state->mc_commit_datas(); !mcd.empty()) {
2503
0
        params.__isset.mc_commit_datas = true;
2504
0
        params.mc_commit_datas.insert(params.mc_commit_datas.end(), mcd.begin(), mcd.end());
2505
44.0k
    } else if (!req.runtime_states.empty()) {
2506
133k
        for (auto* rs : req.runtime_states) {
2507
133k
            if (auto rs_mcd = rs->mc_commit_datas(); !rs_mcd.empty()) {
2508
0
                params.__isset.mc_commit_datas = true;
2509
0
                params.mc_commit_datas.insert(params.mc_commit_datas.end(), rs_mcd.begin(),
2510
0
                                              rs_mcd.end());
2511
0
            }
2512
133k
        }
2513
44.0k
    }
2514
2515
44.0k
    req.runtime_state->get_unreported_errors(&(params.error_log));
2516
44.0k
    params.__isset.error_log = (!params.error_log.empty());
2517
2518
44.0k
    if (_exec_env->cluster_info()->backend_id != 0) {
2519
44.0k
        params.__set_backend_id(_exec_env->cluster_info()->backend_id);
2520
44.0k
    }
2521
2522
44.0k
    TReportExecStatusResult res;
2523
44.0k
    Status rpc_status;
2524
2525
44.0k
    VLOG_DEBUG << "reportExecStatus params is "
2526
12
               << apache::thrift::ThriftDebugString(params).c_str();
2527
44.0k
    if (!exec_status.ok()) {
2528
1.61k
        LOG(WARNING) << "report error status: " << exec_status.msg()
2529
1.61k
                     << " to coordinator: " << req.coord_addr
2530
1.61k
                     << ", query id: " << print_id(req.query_id);
2531
1.61k
    }
2532
44.0k
    try {
2533
44.0k
        try {
2534
44.0k
            (*coord)->reportExecStatus(res, params);
2535
44.0k
        } catch ([[maybe_unused]] apache::thrift::transport::TTransportException& e) {
2536
#ifndef ADDRESS_SANITIZER
2537
            LOG(WARNING) << "Retrying ReportExecStatus. query id: " << print_id(req.query_id)
2538
                         << ", instance id: " << print_id(req.fragment_instance_id) << " to "
2539
                         << req.coord_addr << ", err: " << e.what();
2540
#endif
2541
0
            rpc_status = coord->reopen();
2542
2543
0
            if (!rpc_status.ok()) {
2544
0
                req.cancel_fn(rpc_status);
2545
0
                return;
2546
0
            }
2547
0
            (*coord)->reportExecStatus(res, params);
2548
0
        }
2549
2550
44.0k
        rpc_status = Status::create<false>(res.status);
2551
44.0k
    } catch (apache::thrift::TException& e) {
2552
0
        rpc_status = Status::InternalError("ReportExecStatus() to {} failed: {}",
2553
0
                                           PrintThriftNetworkAddress(req.coord_addr), e.what());
2554
0
    }
2555
2556
44.0k
    if (!rpc_status.ok()) {
2557
0
        LOG_INFO("Going to cancel query {} since report exec status got rpc failed: {}",
2558
0
                 print_id(req.query_id), rpc_status.to_string());
2559
0
        req.cancel_fn(rpc_status);
2560
0
    }
2561
44.0k
}
2562
2563
392k
Status PipelineFragmentContext::send_report(bool done) {
2564
392k
    Status exec_status = _query_ctx->exec_status();
2565
    // If plan is done successfully, but _is_report_success is false,
2566
    // no need to send report.
2567
    // Load will set _is_report_success to true because load wants to know
2568
    // the process.
2569
392k
    if (!_is_report_success && done && exec_status.ok()) {
2570
348k
        return Status::OK();
2571
348k
    }
2572
2573
    // If both _is_report_success and _is_report_on_cancel are false,
2574
    // which means no matter query is success or failed, no report is needed.
2575
    // This may happen when the query limit reached and
2576
    // a internal cancellation being processed
2577
    // When limit is reached the fragment is also cancelled, but _is_report_on_cancel will
2578
    // be set to false, to avoid sending fault report to FE.
2579
44.2k
    if (!_is_report_success && !_is_report_on_cancel) {
2580
160
        if (done) {
2581
            // 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.
2582
160
            return Status::OK();
2583
160
        }
2584
0
        return Status::NeedSendAgain("");
2585
160
    }
2586
2587
44.0k
    std::vector<RuntimeState*> runtime_states;
2588
2589
93.3k
    for (auto& tasks : _tasks) {
2590
133k
        for (auto& task : tasks) {
2591
133k
            runtime_states.push_back(task.second.get());
2592
133k
        }
2593
93.3k
    }
2594
2595
44.0k
    std::string load_eror_url = _query_ctx->get_load_error_url().empty()
2596
44.0k
                                        ? get_load_error_url()
2597
44.0k
                                        : _query_ctx->get_load_error_url();
2598
44.0k
    std::string first_error_msg = _query_ctx->get_first_error_msg().empty()
2599
44.0k
                                          ? get_first_error_msg()
2600
44.0k
                                          : _query_ctx->get_first_error_msg();
2601
2602
44.0k
    ReportStatusRequest req {.status = exec_status,
2603
44.0k
                             .runtime_states = runtime_states,
2604
44.0k
                             .done = done || !exec_status.ok(),
2605
44.0k
                             .coord_addr = _query_ctx->coord_addr,
2606
44.0k
                             .query_id = _query_id,
2607
44.0k
                             .fragment_id = _fragment_id,
2608
44.0k
                             .fragment_instance_id = TUniqueId(),
2609
44.0k
                             .backend_num = -1,
2610
44.0k
                             .runtime_state = _runtime_state.get(),
2611
44.0k
                             .load_error_url = load_eror_url,
2612
44.0k
                             .first_error_msg = first_error_msg,
2613
44.0k
                             .cancel_fn = [this](const Status& reason) { cancel(reason); }};
2614
44.0k
    auto ctx = std::dynamic_pointer_cast<PipelineFragmentContext>(shared_from_this());
2615
44.0k
    return _exec_env->fragment_mgr()->get_thread_pool()->submit_func([this, req, ctx]() {
2616
44.0k
        SCOPED_ATTACH_TASK(ctx->get_query_ctx()->query_mem_tracker());
2617
44.0k
        _coordinator_callback(req);
2618
44.0k
        if (!req.done) {
2619
4.54k
            ctx->refresh_next_report_time();
2620
4.54k
        }
2621
44.0k
    });
2622
44.2k
}
2623
2624
4
size_t PipelineFragmentContext::get_revocable_size(bool* has_running_task) const {
2625
4
    size_t res = 0;
2626
    // _tasks will be cleared during ~PipelineFragmentContext, so that it's safe
2627
    // here to traverse the vector.
2628
4
    for (const auto& task_instances : _tasks) {
2629
6
        for (const auto& task : task_instances) {
2630
6
            if (task.first->is_running()) {
2631
0
                LOG_EVERY_N(INFO, 50) << "Query: " << print_id(_query_id)
2632
0
                                      << " is running, task: " << (void*)task.first.get()
2633
0
                                      << ", is_running: " << task.first->is_running();
2634
0
                *has_running_task = true;
2635
0
                return 0;
2636
0
            }
2637
2638
6
            size_t revocable_size = task.first->get_revocable_size();
2639
6
            if (revocable_size >= SpillFile::MIN_SPILL_WRITE_BATCH_MEM) {
2640
2
                res += revocable_size;
2641
2
            }
2642
6
        }
2643
4
    }
2644
4
    return res;
2645
4
}
2646
2647
8
std::vector<PipelineTask*> PipelineFragmentContext::get_revocable_tasks() const {
2648
8
    std::vector<PipelineTask*> revocable_tasks;
2649
8
    for (const auto& task_instances : _tasks) {
2650
12
        for (const auto& task : task_instances) {
2651
12
            size_t revocable_size_ = task.first->get_revocable_size();
2652
2653
12
            if (revocable_size_ >= SpillFile::MIN_SPILL_WRITE_BATCH_MEM) {
2654
4
                revocable_tasks.emplace_back(task.first.get());
2655
4
            }
2656
12
        }
2657
8
    }
2658
8
    return revocable_tasks;
2659
8
}
2660
2661
72
std::string PipelineFragmentContext::debug_string() {
2662
72
    std::lock_guard<std::mutex> l(_task_mutex);
2663
72
    fmt::memory_buffer debug_string_buffer;
2664
72
    fmt::format_to(debug_string_buffer,
2665
72
                   "PipelineFragmentContext Info: _closed_tasks={}, _total_tasks={}, "
2666
72
                   "need_notify_close={}, fragment_id={}, _rec_cte_stage={}\n",
2667
72
                   _closed_tasks, _total_tasks, _need_notify_close, _fragment_id, _rec_cte_stage);
2668
220
    for (size_t j = 0; j < _tasks.size(); j++) {
2669
148
        fmt::format_to(debug_string_buffer, "Tasks in instance {}:\n", j);
2670
494
        for (size_t i = 0; i < _tasks[j].size(); i++) {
2671
346
            fmt::format_to(debug_string_buffer, "Task {}: {}\n", i,
2672
346
                           _tasks[j][i].first->debug_string());
2673
346
        }
2674
148
    }
2675
2676
72
    return fmt::to_string(debug_string_buffer);
2677
72
}
2678
2679
std::vector<std::shared_ptr<TRuntimeProfileTree>>
2680
2.95k
PipelineFragmentContext::collect_realtime_profile() const {
2681
2.95k
    std::vector<std::shared_ptr<TRuntimeProfileTree>> res;
2682
2683
    // we do not have mutex to protect pipeline_id_to_profile
2684
    // so we need to make sure this funciton is invoked after fragment context
2685
    // has already been prepared.
2686
2.95k
    if (!_prepared) {
2687
0
        std::string msg =
2688
0
                "Query " + print_id(_query_id) + " collecting profile, but its not prepared";
2689
0
        DCHECK(false) << msg;
2690
0
        LOG_ERROR(msg);
2691
0
        return res;
2692
0
    }
2693
2694
    // Make sure first profile is fragment level profile
2695
2.95k
    auto fragment_profile = std::make_shared<TRuntimeProfileTree>();
2696
2.95k
    _fragment_level_profile->to_thrift(fragment_profile.get(), _runtime_state->profile_level());
2697
2.95k
    res.push_back(fragment_profile);
2698
2699
    // pipeline_id_to_profile is initialized in prepare stage
2700
5.40k
    for (auto pipeline_profile : _runtime_state->pipeline_id_to_profile()) {
2701
5.40k
        auto profile_ptr = std::make_shared<TRuntimeProfileTree>();
2702
5.40k
        pipeline_profile->to_thrift(profile_ptr.get(), _runtime_state->profile_level());
2703
5.40k
        res.push_back(profile_ptr);
2704
5.40k
    }
2705
2706
2.95k
    return res;
2707
2.95k
}
2708
2709
std::shared_ptr<TRuntimeProfileTree>
2710
2.95k
PipelineFragmentContext::collect_realtime_load_channel_profile() const {
2711
    // we do not have mutex to protect pipeline_id_to_profile
2712
    // so we need to make sure this funciton is invoked after fragment context
2713
    // has already been prepared.
2714
2.95k
    if (!_prepared) {
2715
0
        std::string msg =
2716
0
                "Query " + print_id(_query_id) + " collecting profile, but its not prepared";
2717
0
        DCHECK(false) << msg;
2718
0
        LOG_ERROR(msg);
2719
0
        return nullptr;
2720
0
    }
2721
2722
7.34k
    for (const auto& tasks : _tasks) {
2723
15.0k
        for (const auto& task : tasks) {
2724
15.0k
            if (task.second->load_channel_profile() == nullptr) {
2725
0
                continue;
2726
0
            }
2727
2728
15.0k
            auto tmp_load_channel_profile = std::make_shared<TRuntimeProfileTree>();
2729
2730
15.0k
            task.second->load_channel_profile()->to_thrift(tmp_load_channel_profile.get(),
2731
15.0k
                                                           _runtime_state->profile_level());
2732
15.0k
            _runtime_state->load_channel_profile()->update(*tmp_load_channel_profile);
2733
15.0k
        }
2734
7.34k
    }
2735
2736
2.95k
    auto load_channel_profile = std::make_shared<TRuntimeProfileTree>();
2737
2.95k
    _runtime_state->load_channel_profile()->to_thrift(load_channel_profile.get(),
2738
2.95k
                                                      _runtime_state->profile_level());
2739
2.95k
    return load_channel_profile;
2740
2.95k
}
2741
2742
// Collect runtime filter IDs registered by all tasks in this PFC.
2743
// Used during recursive CTE stage transitions to know which filters to deregister
2744
// before creating the new PFC for the next recursion round.
2745
// Called from rerun_fragment(wait_for_destroy) while tasks are still closing.
2746
// Thread safety: safe because _tasks is structurally immutable after prepare() —
2747
// the vector sizes do not change, and individual RuntimeState filter sets are
2748
// written only during open() which has completed by the time we reach rerun.
2749
3.50k
std::set<int> PipelineFragmentContext::get_deregister_runtime_filter() const {
2750
3.50k
    std::set<int> result;
2751
14.7k
    for (const auto& _task : _tasks) {
2752
25.1k
        for (const auto& task : _task) {
2753
25.1k
            auto set = task.first->runtime_state()->get_deregister_runtime_filter();
2754
25.1k
            result.merge(set);
2755
25.1k
        }
2756
14.7k
    }
2757
3.50k
    if (_runtime_state) {
2758
3.50k
        auto set = _runtime_state->get_deregister_runtime_filter();
2759
3.50k
        result.merge(set);
2760
3.50k
    }
2761
3.50k
    return result;
2762
3.50k
}
2763
2764
393k
void PipelineFragmentContext::_release_resource() {
2765
393k
    std::lock_guard<std::mutex> l(_task_mutex);
2766
    // The memory released by the query end is recorded in the query mem tracker.
2767
393k
    SCOPED_SWITCH_THREAD_MEM_TRACKER_LIMITER(_query_ctx->query_mem_tracker());
2768
393k
    auto st = _query_ctx->exec_status();
2769
1.14M
    for (auto& _task : _tasks) {
2770
1.14M
        if (!_task.empty()) {
2771
1.14M
            _call_back(_task.front().first->runtime_state(), &st);
2772
1.14M
        }
2773
1.14M
    }
2774
393k
    _tasks.clear();
2775
393k
    _dag.clear();
2776
393k
    _pip_id_to_pipeline.clear();
2777
393k
    _pipelines.clear();
2778
393k
    _sink.reset();
2779
393k
    _root_op.reset();
2780
393k
    _runtime_filter_mgr_map.clear();
2781
393k
    _op_id_to_shared_state.clear();
2782
393k
}
2783
2784
} // namespace doris