Coverage Report

Created: 2026-04-22 11:43

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
3.52M
Status Channel::init(RuntimeState* state) {
57
    // only enable_local_exchange() is true and the destination address is localhost, then the channel is local
58
3.52M
    _is_local &= state->enable_local_exchange();
59
60
3.52M
    if (_is_local) {
61
1.59M
        return Status::OK();
62
1.59M
    }
63
64
1.93M
    RETURN_IF_ERROR(_init_brpc_stub(state));
65
66
1.93M
    return Status::OK();
67
1.93M
}
68
69
1.93M
Status Channel::_init_brpc_stub(RuntimeState* state) {
70
1.93M
    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
1.93M
    auto network_address = _brpc_dest_addr;
77
1.94M
    if (_brpc_dest_addr.hostname == BackendOptions::get_localhost()) {
78
1.94M
        _brpc_stub = state->exec_env()->brpc_internal_client_cache()->get_client(
79
1.94M
                "127.0.0.1", _brpc_dest_addr.port);
80
1.94M
        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
1.93M
    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
1.95M
    if (config::enable_brpc_connection_check) {
93
1.95M
        state->get_query_ctx()->add_using_brpc_stub(_brpc_dest_addr, _brpc_stub);
94
1.95M
    }
95
1.93M
    return Status::OK();
96
1.93M
}
97
98
3.38M
Status Channel::open(RuntimeState* state) {
99
3.38M
    if (_is_local) {
100
1.65M
        RETURN_IF_ERROR(_find_local_recvr(state));
101
1.65M
    }
102
3.38M
    _be_number = state->be_number();
103
3.38M
    _brpc_timeout_ms = get_execution_rpc_timeout_ms(state->execution_timeout());
104
3.38M
    _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
3.38M
    _need_close = (_fragment_instance_id.hi != -1 && _fragment_instance_id.lo != -1);
110
111
3.38M
    return Status::OK();
112
3.38M
}
113
114
1.63M
Status Channel::_find_local_recvr(RuntimeState* state) {
115
1.63M
    auto st = _parent->state()->exec_env()->vstream_mgr()->find_recvr(_fragment_instance_id,
116
1.63M
                                                                      _dest_node_id, &_local_recvr);
117
1.63M
    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
5.86k
        LOG(INFO) << "Query: " << print_id(state->query_id())
124
5.86k
                  << " recvr is not found, maybe downstream task is finished. error st is: "
125
5.86k
                  << st.to_string();
126
5.86k
    }
127
1.63M
    return Status::OK();
128
1.63M
}
129
Status Channel::add_rows(Block* block, const uint32_t* data, const uint32_t offset,
130
3.02M
                         const uint32_t size, bool eos) {
131
3.02M
    if (_fragment_instance_id.lo == -1) {
132
190
        return Status::OK();
133
190
    }
134
135
3.02M
    bool serialized = false;
136
3.02M
    if (_pblock == nullptr) {
137
2.96M
        _pblock = std::make_unique<PBlock>();
138
2.96M
    }
139
3.02M
    RETURN_IF_ERROR(_serializer.next_serialized_block(block, _pblock.get(), 1, &serialized, eos,
140
3.02M
                                                      data, offset, size));
141
3.02M
    if (serialized) {
142
2.94M
        RETURN_IF_ERROR(_send_current_block(eos));
143
2.94M
    }
144
145
2.92M
    return Status::OK();
146
3.02M
}
147
148
1.64M
std::shared_ptr<Dependency> Channel::get_local_channel_dependency() {
149
1.64M
    if (!_local_recvr) {
150
5.87k
        return nullptr;
151
5.87k
    }
152
1.63M
    return _local_recvr->get_local_channel_dependency(_parent->sender_id());
153
1.64M
}
154
155
7.80M
int64_t Channel::mem_usage() const {
156
7.80M
    return _serializer.mem_usage();
157
7.80M
}
158
159
3.52M
Status Channel::send_remote_block(std::unique_ptr<PBlock>&& block, bool eos) {
160
3.52M
    COUNTER_UPDATE(_parent->blocks_sent_counter(), 1);
161
162
3.55M
    if (eos) {
163
3.55M
        if (_eos_send) {
164
1.74M
            return Status::OK();
165
1.80M
        } else {
166
1.80M
            _eos_send = true;
167
1.80M
        }
168
3.55M
    }
169
1.82M
    if (eos || block->column_metas_size()) {
170
1.82M
        RETURN_IF_ERROR(_buffer->add_block(this, {std::move(block), eos}));
171
1.82M
    }
172
1.69M
    return Status::OK();
173
1.77M
}
174
175
131k
Status Channel::send_broadcast_block(std::shared_ptr<BroadcastPBlockHolder>& block, bool eos) {
176
131k
    COUNTER_UPDATE(_parent->blocks_sent_counter(), 1);
177
131k
    if (eos) {
178
131k
        if (_eos_send) {
179
0
            return Status::OK();
180
0
        }
181
131k
        _eos_send = true;
182
131k
    }
183
132k
    if (eos || block->get_block()->column_metas_size()) {
184
132k
        RETURN_IF_ERROR(_buffer->add_block(this, {block, eos}));
185
132k
    }
186
131k
    return Status::OK();
187
131k
}
188
189
6.37M
Status Channel::_send_current_block(bool eos) {
190
6.37M
    if (is_local()) {
191
2.88M
        return _send_local_block(eos);
192
2.88M
    }
193
    // here _pblock maybe nullptr , but we must send the eos to the receiver
194
3.48M
    return send_remote_block(std::move(_pblock), eos);
195
6.37M
}
196
197
2.88M
Status Channel::_send_local_block(bool eos) {
198
2.88M
    Block block;
199
2.88M
    if (_serializer.get_block() != nullptr) {
200
2.50M
        block = _serializer.get_block()->to_block();
201
2.50M
        _serializer.get_block()->set_mutable_columns(block.clone_empty_columns());
202
2.50M
    }
203
204
2.89M
    if (!block.empty() || eos) { // if eos is true, we MUST to send an empty block
205
2.89M
        RETURN_IF_ERROR(send_local_block(&block, eos, true));
206
2.89M
    }
207
2.87M
    return Status::OK();
208
2.88M
}
209
210
3.26M
Status Channel::send_local_block(Block* block, bool eos, bool can_be_moved) {
211
3.26M
    SCOPED_TIMER(_parent->local_send_timer());
212
213
3.26M
    if (eos) {
214
3.17M
        if (_eos_send) {
215
1.52M
            return Status::OK();
216
1.64M
        } else {
217
1.64M
            _eos_send = true;
218
1.64M
        }
219
3.17M
    }
220
221
1.73M
    if (is_receiver_eof()) {
222
0
        return _receiver_status;
223
0
    }
224
1.73M
    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
1.73M
    TaskExecutionContextSPtr ctx_lock = nullptr;
229
1.73M
    if (receiver_status.ok() && _local_recvr != nullptr) {
230
1.73M
        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
1.73M
        if (ctx_lock == nullptr) {
235
0
            receiver_status = Status::EndOfFile("local data stream receiver is deconstructed");
236
0
        }
237
1.73M
    }
238
1.73M
    if (receiver_status.ok()) {
239
1.73M
        COUNTER_UPDATE(_parent->local_bytes_send_counter(), block->bytes());
240
1.73M
        COUNTER_UPDATE(_parent->local_sent_rows(), block->rows());
241
1.73M
        COUNTER_UPDATE(_parent->blocks_sent_counter(), 1);
242
243
1.73M
        const auto sender_id = _parent->sender_id();
244
1.73M
        if (!block->empty()) [[likely]] {
245
168k
            _local_recvr->add_block(block, sender_id, can_be_moved);
246
168k
        }
247
248
1.73M
        if (eos) [[unlikely]] {
249
1.64M
            _local_recvr->remove_sender(sender_id, _be_number, Status::OK());
250
1.64M
            _parent->on_channel_finished(_fragment_instance_id.lo);
251
1.64M
        }
252
1.73M
        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
1.73M
}
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
3.65M
Status Channel::close(RuntimeState* state) {
274
3.65M
    if (_closed) {
275
96.4k
        return Status::OK();
276
96.4k
    }
277
3.55M
    _closed = true;
278
3.55M
    if (!_need_close) {
279
6
        return Status::OK();
280
6
    }
281
282
3.55M
    if (is_receiver_eof()) {
283
95.5k
        _serializer.reset_block();
284
95.5k
        return Status::OK();
285
3.46M
    } else {
286
3.46M
        return _send_current_block(true);
287
3.46M
    }
288
3.55M
}
289
290
BlockSerializer::BlockSerializer(ExchangeSinkLocalState* parent, bool is_local)
291
4.11M
        : _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
3.16M
                                              const uint32_t offset, const uint32_t size) {
296
3.16M
    if (_mutable_block == nullptr) {
297
3.06M
        _mutable_block = MutableBlock::create_unique(block->clone_empty());
298
3.06M
    }
299
300
3.16M
    {
301
3.16M
        SCOPED_TIMER(_parent->merge_block_timer());
302
3.16M
        if (data) {
303
3.02M
            if (size > 0) {
304
63.7k
                RETURN_IF_ERROR(
305
63.7k
                        _mutable_block->add_rows(block, data + offset, data + offset + size));
306
63.7k
            }
307
3.02M
        } else if (!block->empty()) {
308
55.0k
            RETURN_IF_ERROR(_mutable_block->merge(*block));
309
55.0k
        }
310
3.16M
    }
311
312
3.19M
    if (_mutable_block->rows() >= _batch_size || eos ||
313
3.16M
        (_mutable_block->rows() > 0 && _mutable_block->allocated_bytes() > _buffer_mem_limit)) {
314
3.09M
        if (!_is_local) {
315
1.71M
            RETURN_IF_ERROR(_serialize_block(dest, num_receivers));
316
1.71M
        }
317
3.09M
        *serialized = true;
318
3.09M
    } else {
319
72.2k
        *serialized = false;
320
72.2k
    }
321
3.16M
    return Status::OK();
322
3.16M
}
323
324
1.70M
Status BlockSerializer::_serialize_block(PBlock* dest, size_t num_receivers) {
325
1.70M
    if (_mutable_block && _mutable_block->rows() > 0) {
326
18.5k
        auto block = _mutable_block->to_block();
327
18.5k
        RETURN_IF_ERROR(serialize_block(&block, dest, num_receivers));
328
18.5k
        if (_parent->state()->low_memory_mode()) {
329
0
            reset_block();
330
18.5k
        } else {
331
18.5k
            block.clear_column_data();
332
18.5k
            _mutable_block->set_mutable_columns(block.mutate_columns());
333
18.5k
        }
334
18.5k
    }
335
336
1.70M
    return Status::OK();
337
1.70M
}
338
339
63.8k
Status BlockSerializer::serialize_block(const Block* src, PBlock* dest, size_t num_receivers) {
340
63.8k
    SCOPED_TIMER(_parent->_serialize_batch_timer);
341
63.8k
    dest->Clear();
342
63.8k
    size_t uncompressed_bytes = 0, compressed_bytes = 0;
343
63.8k
    int64_t compress_time = 0;
344
63.8k
    RETURN_IF_ERROR(src->serialize(_parent->_state->be_exec_version(), dest, &uncompressed_bytes,
345
63.8k
                                   &compressed_bytes, &compress_time, _parent->compression_type(),
346
63.8k
                                   _parent->transfer_large_data_by_brpc()));
347
63.8k
    COUNTER_UPDATE(_parent->_bytes_sent_counter, compressed_bytes * num_receivers);
348
63.8k
    COUNTER_UPDATE(_parent->_uncompressed_bytes_counter, uncompressed_bytes * num_receivers);
349
63.8k
    COUNTER_UPDATE(_parent->_compress_timer, compress_time);
350
63.8k
#ifndef BE_TEST
351
63.8k
    _parent->state()->get_query_ctx()->resource_ctx()->io_context()->update_shuffle_send_bytes(
352
63.8k
            compressed_bytes * num_receivers);
353
63.8k
    _parent->state()->get_query_ctx()->resource_ctx()->io_context()->update_shuffle_send_rows(
354
63.8k
            src->rows() * num_receivers);
355
63.8k
#endif
356
357
63.8k
    return Status::OK();
358
63.8k
}
359
360
} // namespace doris