Coverage Report

Created: 2026-05-13 15:15

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
be/src/exec/exchange/vdata_stream_sender.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/exchange/vdata_stream_sender.h"
19
20
#include <fmt/format.h>
21
#include <fmt/ranges.h> // IWYU pragma: keep
22
#include <gen_cpp/DataSinks_types.h>
23
#include <gen_cpp/Metrics_types.h>
24
#include <gen_cpp/Types_types.h>
25
#include <gen_cpp/data.pb.h>
26
#include <gen_cpp/internal_service.pb.h>
27
#include <glog/logging.h>
28
#include <stddef.h>
29
30
#include <algorithm>
31
#include <cstdint>
32
#include <functional>
33
#include <map>
34
#include <memory>
35
#include <random>
36
37
#include "common/object_pool.h"
38
#include "common/status.h"
39
#include "core/column/column_const.h"
40
#include "exec/common/sip_hash.h"
41
#include "exec/exchange/vdata_stream_mgr.h"
42
#include "exec/exchange/vdata_stream_recvr.h"
43
#include "exec/operator/exchange_sink_operator.h"
44
#include "exec/operator/result_file_sink_operator.h"
45
#include "exec/sink/vrow_distribution.h"
46
#include "exec/sink/writer/vtablet_writer_v2.h"
47
#include "exprs/vexpr.h"
48
#include "runtime/descriptors.h"
49
#include "runtime/runtime_state.h"
50
#include "runtime/thread_context.h"
51
#include "storage/tablet_info.h"
52
#include "util/proto_util.h"
53
54
namespace doris {
55
56
4.90M
Status Channel::init(RuntimeState* state) {
57
    // only enable_local_exchange() is true and the destination address is localhost, then the channel is local
58
4.90M
    _is_local &= state->enable_local_exchange();
59
60
4.90M
    if (_is_local) {
61
2.07M
        return Status::OK();
62
2.07M
    }
63
64
2.83M
    RETURN_IF_ERROR(_init_brpc_stub(state));
65
66
2.83M
    return Status::OK();
67
2.83M
}
68
69
2.83M
Status Channel::_init_brpc_stub(RuntimeState* state) {
70
2.83M
    if (_brpc_dest_addr.hostname.empty()) {
71
0
        LOG(WARNING) << "there is no brpc destination address's hostname"
72
0
                        ", maybe version is not compatible.";
73
0
        return Status::InternalError("no brpc destination");
74
0
    }
75
76
2.83M
    auto network_address = _brpc_dest_addr;
77
2.85M
    if (_brpc_dest_addr.hostname == BackendOptions::get_localhost()) {
78
2.85M
        _brpc_stub = state->exec_env()->brpc_internal_client_cache()->get_client(
79
2.85M
                "127.0.0.1", _brpc_dest_addr.port);
80
2.85M
        network_address.hostname = "127.0.0.1";
81
18.4E
    } else {
82
18.4E
        _brpc_stub = state->exec_env()->brpc_internal_client_cache()->get_client(_brpc_dest_addr);
83
18.4E
    }
84
85
2.83M
    if (!_brpc_stub) {
86
0
        std::string msg = fmt::format("Get rpc stub failed, dest_addr={}:{}",
87
0
                                      _brpc_dest_addr.hostname, _brpc_dest_addr.port);
88
0
        LOG(WARNING) << msg;
89
0
        return Status::InternalError(msg);
90
0
    }
91
92
2.86M
    if (config::enable_brpc_connection_check) {
93
2.86M
        state->get_query_ctx()->add_using_brpc_stub(_brpc_dest_addr, _brpc_stub);
94
2.86M
    }
95
2.83M
    return Status::OK();
96
2.83M
}
97
98
4.74M
Status Channel::open(RuntimeState* state) {
99
4.74M
    if (_is_local) {
100
2.22M
        RETURN_IF_ERROR(_find_local_recvr(state));
101
2.22M
    }
102
4.74M
    _be_number = state->be_number();
103
4.74M
    _brpc_timeout_ms = get_execution_rpc_timeout_ms(state->execution_timeout());
104
4.74M
    _serializer.set_is_local(_is_local);
105
106
    // In bucket shuffle join will set fragment_instance_id (-1, -1)
107
    // to build a camouflaged empty channel. the ip and port is '0.0.0.0:0"
108
    // so the empty channel not need call function close_internal()
109
4.78M
    _need_close = (_fragment_instance_id.hi != -1 && _fragment_instance_id.lo != -1);
110
111
4.74M
    return Status::OK();
112
4.74M
}
113
114
2.22M
Status Channel::_find_local_recvr(RuntimeState* state) {
115
2.22M
    auto st = _parent->state()->exec_env()->vstream_mgr()->find_recvr(_fragment_instance_id,
116
2.22M
                                                                      _dest_node_id, &_local_recvr);
117
2.22M
    if (!st.ok()) {
118
        // If could not find local receiver, then it means the channel is EOF.
119
        // Maybe downstream task is finished already.
120
        //if (_receiver_status.ok()) {
121
        //    _receiver_status = Status::EndOfFile("local data stream receiver is deconstructed");
122
        //}
123
6.89k
        LOG(INFO) << "Query: " << print_id(state->query_id())
124
6.89k
                  << " recvr is not found, maybe downstream task is finished. error st is: "
125
6.89k
                  << st.to_string();
126
6.89k
    }
127
2.22M
    return Status::OK();
128
2.22M
}
129
Status Channel::add_rows(Block* block, const uint32_t* data, const uint32_t offset,
130
4.59M
                         const uint32_t size, bool eos) {
131
4.59M
    if (_fragment_instance_id.lo == -1) {
132
562
        return Status::OK();
133
562
    }
134
135
4.59M
    bool serialized = false;
136
4.59M
    if (_pblock == nullptr) {
137
4.53M
        _pblock = std::make_unique<PBlock>();
138
4.53M
    }
139
4.59M
    RETURN_IF_ERROR(_serializer.next_serialized_block(block, _pblock.get(), 1, &serialized, eos,
140
4.59M
                                                      data, offset, size));
141
4.59M
    if (serialized) {
142
4.51M
        RETURN_IF_ERROR(_send_current_block(eos));
143
4.51M
    }
144
145
4.49M
    return Status::OK();
146
4.59M
}
147
148
2.23M
std::shared_ptr<Dependency> Channel::get_local_channel_dependency() {
149
2.23M
    if (!_local_recvr) {
150
6.91k
        return nullptr;
151
6.91k
    }
152
2.23M
    return _local_recvr->get_local_channel_dependency(_parent->sender_id());
153
2.23M
}
154
155
10.7M
int64_t Channel::mem_usage() const {
156
10.7M
    return _serializer.mem_usage();
157
10.7M
}
158
159
5.30M
Status Channel::send_remote_block(std::unique_ptr<PBlock>&& block, bool eos) {
160
5.30M
    COUNTER_UPDATE(_parent->blocks_sent_counter(), 1);
161
162
5.38M
    if (eos) {
163
5.38M
        if (_eos_send) {
164
2.68M
            return Status::OK();
165
2.70M
        } else {
166
2.70M
            _eos_send = true;
167
2.70M
        }
168
5.38M
    }
169
2.72M
    if (eos || block->column_metas_size()) {
170
2.72M
        RETURN_IF_ERROR(_buffer->add_block(this, {std::move(block), eos}));
171
2.72M
    }
172
2.53M
    return Status::OK();
173
2.62M
}
174
175
137k
Status Channel::send_broadcast_block(std::shared_ptr<BroadcastPBlockHolder>& block, bool eos) {
176
137k
    COUNTER_UPDATE(_parent->blocks_sent_counter(), 1);
177
137k
    if (eos) {
178
137k
        if (_eos_send) {
179
0
            return Status::OK();
180
0
        }
181
137k
        _eos_send = true;
182
137k
    }
183
138k
    if (eos || block->get_block()->column_metas_size()) {
184
138k
        RETURN_IF_ERROR(_buffer->add_block(this, {block, eos}));
185
138k
    }
186
137k
    return Status::OK();
187
137k
}
188
189
9.43M
Status Channel::_send_current_block(bool eos) {
190
9.43M
    if (is_local()) {
191
4.12M
        return _send_local_block(eos);
192
4.12M
    }
193
    // here _pblock maybe nullptr , but we must send the eos to the receiver
194
5.30M
    return send_remote_block(std::move(_pblock), eos);
195
9.43M
}
196
197
4.12M
Status Channel::_send_local_block(bool eos) {
198
4.12M
    Block block;
199
4.12M
    if (_serializer.get_block() != nullptr) {
200
3.77M
        block = _serializer.get_block()->to_block();
201
3.77M
        _serializer.get_block()->set_mutable_columns(block.clone_empty_columns());
202
3.77M
    }
203
204
4.12M
    if (!block.empty() || eos) { // if eos is true, we MUST to send an empty block
205
4.11M
        RETURN_IF_ERROR(send_local_block(&block, eos, true));
206
4.11M
    }
207
4.11M
    return Status::OK();
208
4.12M
}
209
210
4.50M
Status Channel::send_local_block(Block* block, bool eos, bool can_be_moved) {
211
4.50M
    SCOPED_TIMER(_parent->local_send_timer());
212
213
4.50M
    if (eos) {
214
4.42M
        if (_eos_send) {
215
2.18M
            return Status::OK();
216
2.24M
        } else {
217
2.24M
            _eos_send = true;
218
2.24M
        }
219
4.42M
    }
220
221
2.31M
    if (is_receiver_eof()) {
222
0
        return _receiver_status;
223
0
    }
224
2.31M
    auto receiver_status = _recvr_status();
225
    // _local_recvr depdend on ExchangeLocalState* _parent to do some memory counter settings
226
    // but it only owns a raw pointer, so that the ExchangeLocalState object may be deconstructed.
227
    // Lock the fragment context to ensure the runtime state and other objects are not deconstructed
228
2.31M
    TaskExecutionContextSPtr ctx_lock = nullptr;
229
2.33M
    if (receiver_status.ok() && _local_recvr != nullptr) {
230
2.33M
        ctx_lock = _local_recvr->task_exec_ctx();
231
        // Do not return internal error, because when query finished, the downstream node
232
        // may finish before upstream node. And the object maybe deconstructed. If return error
233
        // then the upstream node may report error status to FE, the query is failed.
234
2.33M
        if (ctx_lock == nullptr) {
235
0
            receiver_status = Status::EndOfFile("local data stream receiver is deconstructed");
236
0
        }
237
2.33M
    }
238
2.34M
    if (receiver_status.ok()) {
239
2.34M
        COUNTER_UPDATE(_parent->local_bytes_send_counter(), block->bytes());
240
2.34M
        COUNTER_UPDATE(_parent->local_sent_rows(), block->rows());
241
2.34M
        COUNTER_UPDATE(_parent->blocks_sent_counter(), 1);
242
243
2.34M
        const auto sender_id = _parent->sender_id();
244
2.34M
        if (!block->empty()) [[likely]] {
245
176k
            _local_recvr->add_block(block, sender_id, can_be_moved);
246
176k
        }
247
248
2.34M
        if (eos) [[unlikely]] {
249
2.24M
            _local_recvr->remove_sender(sender_id, _be_number, Status::OK());
250
2.24M
            _parent->on_channel_finished(_fragment_instance_id.lo);
251
2.24M
        }
252
2.34M
        return Status::OK();
253
18.4E
    } else {
254
18.4E
        _receiver_status = std::move(receiver_status);
255
18.4E
        _parent->on_channel_finished(_fragment_instance_id.lo);
256
18.4E
        return _receiver_status;
257
18.4E
    }
258
2.31M
}
259
260
0
std::string Channel::debug_string() const {
261
0
    fmt::memory_buffer debug_string_buffer;
262
0
    fmt::format_to(debug_string_buffer,
263
0
                   "fragment_instance_id: {}, _dest_node_id: {}, _is_local: {}, _receiver_status: "
264
0
                   "{}, _closed: {}, _need_close: {}, _be_number: {}, _eos_send: {}",
265
0
                   print_id(_fragment_instance_id), _dest_node_id, _is_local,
266
0
                   _receiver_status.to_string(), _closed, _need_close, _be_number, _eos_send);
267
0
    if (_is_local) {
268
0
        fmt::format_to(debug_string_buffer, "_local_recvr: {}", _local_recvr->debug_string());
269
0
    }
270
0
    return fmt::to_string(debug_string_buffer);
271
0
}
272
273
5.17M
Status Channel::close(RuntimeState* state) {
274
5.17M
    if (_closed) {
275
99.8k
        return Status::OK();
276
99.8k
    }
277
5.07M
    _closed = true;
278
5.07M
    if (!_need_close) {
279
10
        return Status::OK();
280
10
    }
281
282
5.07M
    if (is_receiver_eof()) {
283
98.6k
        _serializer.reset_block();
284
98.6k
        return Status::OK();
285
4.97M
    } else {
286
4.97M
        return _send_current_block(true);
287
4.97M
    }
288
5.07M
}
289
290
BlockSerializer::BlockSerializer(ExchangeSinkLocalState* parent, bool is_local)
291
5.63M
        : _parent(parent), _is_local(is_local), _batch_size(parent->state()->batch_size()) {}
292
293
Status BlockSerializer::next_serialized_block(Block* block, PBlock* dest, size_t num_receivers,
294
                                              bool* serialized, bool eos, const uint32_t* data,
295
4.76M
                                              const uint32_t offset, const uint32_t size) {
296
4.76M
    if (_mutable_block == nullptr) {
297
4.65M
        _mutable_block = MutableBlock::create_unique(block->clone_empty());
298
4.65M
    }
299
300
4.76M
    {
301
4.76M
        SCOPED_TIMER(_parent->merge_block_timer());
302
4.76M
        if (data) {
303
4.61M
            if (size > 0) {
304
64.2k
                RETURN_IF_ERROR(
305
64.2k
                        _mutable_block->add_rows(block, data + offset, data + offset + size));
306
64.2k
            }
307
4.61M
        } else if (!block->empty()) {
308
63.2k
            RETURN_IF_ERROR(_mutable_block->merge(*block));
309
63.2k
        }
310
4.76M
    }
311
312
4.79M
    if (_mutable_block->rows() >= _batch_size || eos ||
313
4.76M
        (_mutable_block->rows() > 0 && _mutable_block->allocated_bytes() > _buffer_mem_limit)) {
314
4.68M
        if (!_is_local) {
315
2.64M
            RETURN_IF_ERROR(_serialize_block(dest, num_receivers));
316
2.64M
        }
317
4.68M
        *serialized = true;
318
4.68M
    } else {
319
79.0k
        *serialized = false;
320
79.0k
    }
321
4.76M
    return Status::OK();
322
4.76M
}
323
324
2.63M
Status BlockSerializer::_serialize_block(PBlock* dest, size_t num_receivers) {
325
2.63M
    if (_mutable_block && _mutable_block->rows() > 0) {
326
20.0k
        auto block = _mutable_block->to_block();
327
20.0k
        RETURN_IF_ERROR(serialize_block(&block, dest, num_receivers));
328
20.0k
        if (_parent->state()->low_memory_mode()) {
329
0
            reset_block();
330
20.0k
        } else {
331
20.0k
            block.clear_column_data();
332
20.0k
            _mutable_block->set_mutable_columns(block.mutate_columns());
333
20.0k
        }
334
20.0k
    }
335
336
2.63M
    return Status::OK();
337
2.63M
}
338
339
63.5k
Status BlockSerializer::serialize_block(const Block* src, PBlock* dest, size_t num_receivers) {
340
63.5k
    SCOPED_TIMER(_parent->_serialize_batch_timer);
341
63.5k
    dest->Clear();
342
63.5k
    size_t uncompressed_bytes = 0, compressed_bytes = 0;
343
63.5k
    int64_t compress_time = 0;
344
63.5k
    RETURN_IF_ERROR(src->serialize(_parent->_state->be_exec_version(), dest, &uncompressed_bytes,
345
63.5k
                                   &compressed_bytes, &compress_time, _parent->compression_type(),
346
63.5k
                                   _parent->transfer_large_data_by_brpc()));
347
63.5k
    COUNTER_UPDATE(_parent->_bytes_sent_counter, compressed_bytes * num_receivers);
348
63.5k
    COUNTER_UPDATE(_parent->_uncompressed_bytes_counter, uncompressed_bytes * num_receivers);
349
63.5k
    COUNTER_UPDATE(_parent->_compress_timer, compress_time);
350
63.5k
#ifndef BE_TEST
351
63.5k
    _parent->state()->get_query_ctx()->resource_ctx()->io_context()->update_shuffle_send_bytes(
352
63.5k
            compressed_bytes * num_receivers);
353
63.5k
    _parent->state()->get_query_ctx()->resource_ctx()->io_context()->update_shuffle_send_rows(
354
63.5k
            src->rows() * num_receivers);
355
63.5k
#endif
356
357
63.5k
    return Status::OK();
358
63.5k
}
359
360
} // namespace doris