be/src/load/delta_writer/delta_writer.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 "load/delta_writer/delta_writer.h" |
19 | | |
20 | | #include <fmt/format.h> |
21 | | #include <gen_cpp/internal_service.pb.h> |
22 | | #include <gen_cpp/olap_file.pb.h> |
23 | | |
24 | | #include <ostream> |
25 | | #include <string> |
26 | | #include <utility> |
27 | | |
28 | | // IWYU pragma: no_include <opentelemetry/common/threadlocal.h> |
29 | | #include "common/compiler_util.h" // IWYU pragma: keep |
30 | | #include "common/config.h" |
31 | | #include "common/logging.h" |
32 | | #include "common/status.h" |
33 | | #include "core/block/block.h" |
34 | | #include "io/fs/file_writer.h" // IWYU pragma: keep |
35 | | #include "load/memtable/memtable_flush_executor.h" |
36 | | #include "load/memtable/memtable_memory_limiter.h" |
37 | | #include "runtime/exec_env.h" |
38 | | #include "runtime/thread_context.h" |
39 | | #include "runtime/workload_management/resource_context.h" |
40 | | #include "storage/olap_define.h" |
41 | | #include "storage/rowset/beta_rowset.h" |
42 | | #include "storage/rowset/beta_rowset_writer.h" |
43 | | #include "storage/rowset/rowset_meta.h" |
44 | | #include "storage/rowset_builder.h" |
45 | | #include "storage/schema_change/schema_change.h" |
46 | | #include "storage/storage_engine.h" |
47 | | #include "storage/tablet/tablet_manager.h" |
48 | | #include "storage/tablet_info.h" |
49 | | #include "util/mem_info.h" |
50 | | #include "util/stopwatch.hpp" |
51 | | #include "util/time.h" |
52 | | |
53 | | namespace doris { |
54 | | using namespace ErrorCode; |
55 | | |
56 | | BaseDeltaWriter::BaseDeltaWriter(const WriteRequest& req, RuntimeProfile* profile, |
57 | | const UniqueId& load_id) |
58 | 298k | : _req(req), _memtable_writer(new MemTableWriter(req)) { |
59 | 298k | if (profile != nullptr) { |
60 | 47 | _init_profile(profile); |
61 | 47 | } |
62 | 298k | } |
63 | | |
64 | | DeltaWriter::DeltaWriter(StorageEngine& engine, const WriteRequest& req, RuntimeProfile* profile, |
65 | | const UniqueId& load_id) |
66 | 29.4k | : BaseDeltaWriter(req, profile, load_id), _engine(engine) { |
67 | 29.4k | DCHECK(req.write_req_type == WriteRequestType::DATA); |
68 | 29.4k | _rowset_builder = std::make_unique<RowsetBuilder>(_engine, req, profile); |
69 | 29.4k | } |
70 | | |
71 | | DeltaWriter::DeltaWriter(StorageEngine& engine, const WriteRequest& group_build_req, |
72 | | const WriteRequest& sub_data_req, const WriteRequest& sub_row_binlog_req, |
73 | | RuntimeProfile* profile, const UniqueId& load_id) |
74 | 0 | : BaseDeltaWriter(group_build_req, profile, load_id), _engine(engine) { |
75 | 0 | DCHECK(group_build_req.write_req_type == WriteRequestType::GROUP && |
76 | 0 | sub_data_req.write_req_type == WriteRequestType::DATA && |
77 | 0 | sub_row_binlog_req.write_req_type == WriteRequestType::ROW_BINLOG); |
78 | 0 | _rowset_builder = std::make_unique<GroupRowsetBuilder>(_engine, group_build_req, sub_data_req, |
79 | 0 | sub_row_binlog_req, profile); |
80 | 0 | } |
81 | | |
82 | 47 | void BaseDeltaWriter::_init_profile(RuntimeProfile* profile) { |
83 | 47 | DCHECK(profile != nullptr); |
84 | 47 | _profile = profile->create_child(fmt::format("DeltaWriter {}", _req.tablet_id), true, true); |
85 | 47 | _close_wait_timer = ADD_TIMER(_profile, "CloseWaitTime"); |
86 | 47 | _wait_flush_limit_timer = ADD_TIMER(_profile, "WaitFlushLimitTime"); |
87 | 47 | } |
88 | | |
89 | 0 | void DeltaWriter::_init_profile(RuntimeProfile* profile) { |
90 | 0 | DCHECK(profile != nullptr); |
91 | 0 | BaseDeltaWriter::_init_profile(profile); |
92 | 0 | _commit_txn_timer = ADD_TIMER(_profile, "CommitTxnTime"); |
93 | 0 | } |
94 | | |
95 | 298k | BaseDeltaWriter::~BaseDeltaWriter() { |
96 | 298k | if (!_is_init) { |
97 | 215k | return; |
98 | 215k | } |
99 | | |
100 | | // cancel and wait all memtables in flush queue to be finished |
101 | 82.4k | static_cast<void>(_memtable_writer->cancel()); |
102 | | |
103 | 82.5k | if (_rowset_builder->tablet() != nullptr) { |
104 | 82.5k | const FlushStatistic& stat = _memtable_writer->get_flush_token_stats(); |
105 | 82.5k | _rowset_builder->tablet()->flush_bytes->increment(stat.flush_size_bytes); |
106 | 82.5k | _rowset_builder->tablet()->flush_finish_count->increment(stat.flush_finish_count); |
107 | 82.5k | } |
108 | 82.4k | } |
109 | | |
110 | | void BaseDeltaWriter::collect_tablet_load_rowset_num_info( |
111 | | BaseTablet* tablet, |
112 | 158k | google::protobuf::RepeatedPtrField<PTabletLoadRowsetInfo>* tablet_infos) { |
113 | 158k | if (tablet == nullptr) { |
114 | 0 | return; |
115 | 0 | } |
116 | 158k | auto max_version_config = tablet->max_version_config(); |
117 | 158k | if (auto version_cnt = tablet->tablet_meta()->version_count(); |
118 | 158k | UNLIKELY(version_cnt > |
119 | 158k | (max_version_config * config::load_back_pressure_version_threshold / 100))) { |
120 | 0 | auto* load_info = tablet_infos->Add(); |
121 | 0 | load_info->set_current_rowset_nums(static_cast<int32_t>(version_cnt)); |
122 | 0 | load_info->set_max_config_rowset_nums(max_version_config); |
123 | 0 | } |
124 | 158k | } |
125 | | |
126 | | void BaseDeltaWriter::set_tablet_load_rowset_num_info( |
127 | 130k | google::protobuf::RepeatedPtrField<PTabletLoadRowsetInfo>* tablet_infos) { |
128 | 130k | auto* tablet = _rowset_builder->tablet().get(); |
129 | 130k | collect_tablet_load_rowset_num_info(tablet, tablet_infos); |
130 | 130k | } |
131 | | |
132 | 4.37k | int64_t BaseDeltaWriter::table_id() const { |
133 | 4.37k | DORIS_CHECK(_req.table_schema_param != nullptr); |
134 | 4.37k | return _req.table_schema_param->table_id(); |
135 | 4.37k | } |
136 | | |
137 | | DeltaWriter::~DeltaWriter() = default; |
138 | | |
139 | 82.2k | Status BaseDeltaWriter::init() { |
140 | 82.2k | if (_is_init) { |
141 | 0 | return Status::OK(); |
142 | 0 | } |
143 | 82.2k | std::shared_ptr<WorkloadGroup> wg_sptr = nullptr; |
144 | 82.2k | if (doris::thread_context()->is_attach_task()) { |
145 | 82.2k | wg_sptr = doris::thread_context()->resource_ctx()->workload_group(); |
146 | 82.2k | } |
147 | 82.2k | RETURN_IF_ERROR(_rowset_builder->init()); |
148 | 82.2k | RETURN_IF_ERROR(_memtable_writer->init( |
149 | 82.2k | _rowset_builder->rowset_writer(), _rowset_builder->tablet_schema(), |
150 | 82.2k | _rowset_builder->get_partial_update_info(), wg_sptr, |
151 | 82.2k | _rowset_builder->tablet_sptr()->enable_unique_key_merge_on_write())); |
152 | 82.2k | ExecEnv::GetInstance()->memtable_memory_limiter()->register_writer(_memtable_writer); |
153 | 82.2k | _is_init = true; |
154 | 82.2k | return Status::OK(); |
155 | 82.2k | } |
156 | | |
157 | | Status DeltaWriter::write(const Block* block, const TabletAddRowsPayload& rows, |
158 | 15.3k | bool* memtable_flushed) { |
159 | 15.3k | if (memtable_flushed != nullptr) { |
160 | 15.2k | *memtable_flushed = false; |
161 | 15.2k | } |
162 | 15.3k | if (UNLIKELY(rows.row_idxs.empty())) { |
163 | 0 | return Status::OK(); |
164 | 0 | } |
165 | 15.3k | if (_req.enable_table_memtable_backpressure && !_req.is_high_priority) { |
166 | 0 | ExecEnv::GetInstance()->memtable_memory_limiter()->handle_table_memtable_backpressure( |
167 | 0 | [this]() { |
168 | 0 | std::lock_guard<std::mutex> l(_lock); |
169 | 0 | return _is_cancelled; |
170 | 0 | }, |
171 | 0 | table_id()); |
172 | 0 | } |
173 | 15.3k | _lock_watch.start(); |
174 | 15.3k | std::lock_guard<std::mutex> l(_lock); |
175 | 15.3k | _lock_watch.stop(); |
176 | 15.3k | if (!_is_init && !_is_cancelled) { |
177 | 15.3k | RETURN_IF_ERROR(init()); |
178 | 15.3k | } |
179 | 15.3k | { |
180 | 15.3k | SCOPED_TIMER(_wait_flush_limit_timer); |
181 | 15.3k | while (_memtable_writer->flush_running_count() >= |
182 | 15.3k | config::memtable_flush_running_count_limit) { |
183 | 0 | std::this_thread::sleep_for(std::chrono::milliseconds(10)); |
184 | 0 | } |
185 | 15.3k | } |
186 | 15.3k | return _memtable_writer->write(block, rows, memtable_flushed); |
187 | 15.3k | } |
188 | | |
189 | 27.9k | Status BaseDeltaWriter::wait_flush() { |
190 | 27.9k | return _memtable_writer->wait_flush(); |
191 | 27.9k | } |
192 | | |
193 | 0 | Status BaseDeltaWriter::flush_memtable_async() { |
194 | 0 | return _memtable_writer->flush_async(); |
195 | 0 | } |
196 | | |
197 | 0 | Status DeltaWriter::flush_memtable_async() { |
198 | 0 | _lock_watch.start(); |
199 | 0 | std::lock_guard<std::mutex> l(_lock); |
200 | 0 | _lock_watch.stop(); |
201 | 0 | return BaseDeltaWriter::flush_memtable_async(); |
202 | 0 | } |
203 | | |
204 | 27.9k | Status DeltaWriter::close() { |
205 | 27.9k | _lock_watch.start(); |
206 | 27.9k | std::lock_guard<std::mutex> l(_lock); |
207 | 27.9k | _lock_watch.stop(); |
208 | 27.9k | if (!_is_init && !_is_cancelled) { |
209 | | // if this delta writer is not initialized, but close() is called. |
210 | | // which means this tablet has no data loaded, but at least one tablet |
211 | | // in same partition has data loaded. |
212 | | // so we have to also init this DeltaWriter, so that it can create an empty rowset |
213 | | // for this tablet when being closed. |
214 | 12.6k | RETURN_IF_ERROR(init()); |
215 | 12.6k | } |
216 | 27.9k | return _memtable_writer->close(); |
217 | 27.9k | } |
218 | | |
219 | 82.5k | Status BaseDeltaWriter::build_rowset() { |
220 | 82.5k | SCOPED_TIMER(_close_wait_timer); |
221 | 82.5k | RETURN_IF_ERROR(_memtable_writer->close_wait(_profile)); |
222 | 82.4k | return _rowset_builder->build_rowset(); |
223 | 82.5k | } |
224 | | |
225 | 27.9k | Status DeltaWriter::build_rowset() { |
226 | 27.9k | std::lock_guard<std::mutex> l(_lock); |
227 | 27.9k | DCHECK(_is_init) |
228 | 0 | << "delta writer is supposed be to initialized before build_rowset() being called"; |
229 | 27.9k | return BaseDeltaWriter::build_rowset(); |
230 | 27.9k | } |
231 | | |
232 | 202k | Status BaseDeltaWriter::submit_calc_delete_bitmap_task() { |
233 | 202k | return _rowset_builder->submit_calc_delete_bitmap_task(); |
234 | 202k | } |
235 | | |
236 | 202k | Status BaseDeltaWriter::wait_calc_delete_bitmap() { |
237 | 202k | return _rowset_builder->wait_calc_delete_bitmap(); |
238 | 202k | } |
239 | | |
240 | 15 | Status DeltaWriter::commit_txn() { |
241 | 15 | std::lock_guard<std::mutex> l(_lock); |
242 | 15 | SCOPED_TIMER(_commit_txn_timer); |
243 | 15 | return _rowset_builder->commit_txn(); |
244 | 15 | } |
245 | | |
246 | 95.4k | Status BaseDeltaWriter::cancel() { |
247 | 95.4k | return cancel_with_status(Status::Cancelled("already cancelled")); |
248 | 95.4k | } |
249 | | |
250 | 95.4k | Status BaseDeltaWriter::cancel_with_status(const Status& st) { |
251 | 95.4k | if (_is_cancelled) { |
252 | 0 | return Status::OK(); |
253 | 0 | } |
254 | 95.4k | RETURN_IF_ERROR(_memtable_writer->cancel_with_status(st)); |
255 | 95.4k | _is_cancelled = true; |
256 | 95.4k | return Status::OK(); |
257 | 95.4k | } |
258 | | |
259 | 1.50k | Status DeltaWriter::cancel_with_status(const Status& st) { |
260 | 1.50k | std::lock_guard<std::mutex> l(_lock); |
261 | 1.50k | return BaseDeltaWriter::cancel_with_status(st); |
262 | 1.50k | } |
263 | | |
264 | 476 | int64_t BaseDeltaWriter::mem_consumption(MemType mem) { |
265 | 476 | return _memtable_writer->mem_consumption(mem); |
266 | 476 | } |
267 | | |
268 | 230k | int64_t BaseDeltaWriter::num_rows_filtered() const { |
269 | 230k | auto rowset_writer = _rowset_builder->rowset_writer(); |
270 | 230k | return rowset_writer == nullptr ? 0 : rowset_writer->num_rows_filtered(); |
271 | 230k | } |
272 | | |
273 | | } // namespace doris |