/root/doris/cloud/src/recycler/recycler.h
Line | Count | Source (jump to first uncovered line) |
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 | | #pragma once |
19 | | |
20 | | #include <gen_cpp/cloud.pb.h> |
21 | | |
22 | | #include <atomic> |
23 | | #include <condition_variable> |
24 | | #include <cstdint> |
25 | | #include <deque> |
26 | | #include <functional> |
27 | | #include <memory> |
28 | | #include <string> |
29 | | #include <string_view> |
30 | | #include <thread> |
31 | | #include <utility> |
32 | | |
33 | | #include "common/bvars.h" |
34 | | #include "meta-service/txn_lazy_committer.h" |
35 | | #include "recycler/storage_vault_accessor.h" |
36 | | #include "recycler/white_black_list.h" |
37 | | |
38 | | namespace brpc { |
39 | | class Server; |
40 | | } // namespace brpc |
41 | | |
42 | | namespace doris::cloud { |
43 | | class TxnKv; |
44 | | class InstanceRecycler; |
45 | | class StorageVaultAccessor; |
46 | | class Checker; |
47 | | class SimpleThreadPool; |
48 | | class RecyclerMetricsContext; |
49 | | struct RecyclerThreadPoolGroup { |
50 | 6 | RecyclerThreadPoolGroup() = default; |
51 | | RecyclerThreadPoolGroup(std::shared_ptr<SimpleThreadPool> s3_producer_pool, |
52 | | std::shared_ptr<SimpleThreadPool> recycle_tablet_pool, |
53 | | std::shared_ptr<SimpleThreadPool> group_recycle_function_pool) |
54 | | : s3_producer_pool(std::move(s3_producer_pool)), |
55 | | recycle_tablet_pool(std::move(recycle_tablet_pool)), |
56 | 8 | group_recycle_function_pool(std::move(group_recycle_function_pool)) {} |
57 | 160 | ~RecyclerThreadPoolGroup() = default; |
58 | 73 | RecyclerThreadPoolGroup(const RecyclerThreadPoolGroup&) = default; |
59 | | RecyclerThreadPoolGroup& operator=(RecyclerThreadPoolGroup& other) = default; |
60 | 8 | RecyclerThreadPoolGroup& operator=(RecyclerThreadPoolGroup&& other) = default; |
61 | 73 | RecyclerThreadPoolGroup(RecyclerThreadPoolGroup&&) = default; |
62 | | // used for accessor.delete_files, accessor.delete_directory |
63 | | std::shared_ptr<SimpleThreadPool> s3_producer_pool; |
64 | | // used for InstanceRecycler::recycle_tablet |
65 | | std::shared_ptr<SimpleThreadPool> recycle_tablet_pool; |
66 | | std::shared_ptr<SimpleThreadPool> group_recycle_function_pool; |
67 | | }; |
68 | | |
69 | | class Recycler { |
70 | | public: |
71 | | explicit Recycler(std::shared_ptr<TxnKv> txn_kv); |
72 | | ~Recycler(); |
73 | | |
74 | | // returns 0 for success otherwise error |
75 | | int start(brpc::Server* server); |
76 | | |
77 | | void stop(); |
78 | | |
79 | 293 | bool stopped() const { return stopped_.load(std::memory_order_acquire); } |
80 | | |
81 | | private: |
82 | | void recycle_callback(); |
83 | | |
84 | | void instance_scanner_callback(); |
85 | | |
86 | | void lease_recycle_jobs(); |
87 | | |
88 | | void check_recycle_tasks(); |
89 | | |
90 | | private: |
91 | | friend class RecyclerServiceImpl; |
92 | | |
93 | | std::shared_ptr<TxnKv> txn_kv_; |
94 | | std::atomic_bool stopped_ {false}; |
95 | | |
96 | | std::vector<std::thread> workers_; |
97 | | |
98 | | std::mutex mtx_; |
99 | | // notify recycle workers |
100 | | std::condition_variable pending_instance_cond_; |
101 | | std::deque<InstanceInfoPB> pending_instance_queue_; |
102 | | std::unordered_set<std::string> pending_instance_set_; |
103 | | std::unordered_map<std::string, std::shared_ptr<InstanceRecycler>> recycling_instance_map_; |
104 | | // notify instance scanner and lease thread |
105 | | std::condition_variable notifier_; |
106 | | |
107 | | std::string ip_port_; |
108 | | |
109 | | WhiteBlackList instance_filter_; |
110 | | std::unique_ptr<Checker> checker_; |
111 | | |
112 | | RecyclerThreadPoolGroup _thread_pool_group; |
113 | | |
114 | | std::shared_ptr<TxnLazyCommitter> txn_lazy_committer_; |
115 | | }; |
116 | | |
117 | | enum class RowsetRecyclingState { |
118 | | FORMAL_ROWSET, |
119 | | TMP_ROWSET, |
120 | | }; |
121 | | |
122 | | class InstanceRecycler { |
123 | | public: |
124 | | explicit InstanceRecycler(std::shared_ptr<TxnKv> txn_kv, const InstanceInfoPB& instance, |
125 | | RecyclerThreadPoolGroup thread_pool_group, |
126 | | std::shared_ptr<TxnLazyCommitter> txn_lazy_committer); |
127 | | ~InstanceRecycler(); |
128 | | |
129 | | // returns 0 for success otherwise error |
130 | | int init(); |
131 | | |
132 | 0 | void stop() { stopped_.store(true, std::memory_order_release); } |
133 | 32 | bool stopped() const { return stopped_.load(std::memory_order_acquire); } |
134 | | |
135 | | // returns 0 for success otherwise error |
136 | | int do_recycle(); |
137 | | |
138 | | // remove all kv and data in this instance, ONLY be called when instance has been deleted |
139 | | // returns 0 for success otherwise error |
140 | | int recycle_deleted_instance(); |
141 | | |
142 | | // scan and recycle expired indexes: |
143 | | // 1. dropped table, dropped mv |
144 | | // 2. half-successtable/index when create |
145 | | // returns 0 for success otherwise error |
146 | | int recycle_indexes(); |
147 | | |
148 | | // scan and recycle expired partitions: |
149 | | // 1. dropped parttion |
150 | | // 2. half-success partition when create |
151 | | // returns 0 for success otherwise error |
152 | | int recycle_partitions(); |
153 | | |
154 | | // scan and recycle expired rowsets: |
155 | | // 1. prepare_rowset will produce recycle_rowset before uploading data to remote storage (memo) |
156 | | // 2. compaction will change the input rowsets to recycle_rowset |
157 | | // returns 0 for success otherwise error |
158 | | int recycle_rowsets(); |
159 | | |
160 | | // scan and recycle expired tmp rowsets: |
161 | | // 1. commit_rowset will produce tmp_rowset when finish upload data (load or compaction) to remote storage |
162 | | // returns 0 for success otherwise error |
163 | | int recycle_tmp_rowsets(); |
164 | | |
165 | | /** |
166 | | * recycle all tablets belonging to the index specified by `index_id` |
167 | | * |
168 | | * @param partition_id if positive, only recycle tablets in this partition belonging to the specified index |
169 | | * @param is_empty_tablet indicates whether the tablet has object files, can skip delete objects if tablet is empty |
170 | | * @return 0 for success otherwise error |
171 | | */ |
172 | | int recycle_tablets(int64_t table_id, int64_t index_id, RecyclerMetricsContext& ctx, |
173 | | int64_t partition_id = -1, bool is_empty_tablet = false); |
174 | | |
175 | | /** |
176 | | * recycle all rowsets belonging to the tablet specified by `tablet_id` |
177 | | * |
178 | | * @return 0 for success otherwise error |
179 | | */ |
180 | | int recycle_tablet(int64_t tablet_id, RecyclerMetricsContext& metrics_context); |
181 | | |
182 | | // scan and recycle useless partition version kv |
183 | | int recycle_versions(); |
184 | | |
185 | | // scan and abort timeout txn label |
186 | | // returns 0 for success otherwise error |
187 | | int abort_timeout_txn(); |
188 | | |
189 | | //scan and recycle expire txn label |
190 | | // returns 0 for success otherwise error |
191 | | int recycle_expired_txn_label(); |
192 | | |
193 | | // scan and recycle finished or timeout copy jobs |
194 | | // returns 0 for success otherwise error |
195 | | int recycle_copy_jobs(); |
196 | | |
197 | | // scan and recycle dropped internal stage |
198 | | // returns 0 for success otherwise error |
199 | | int recycle_stage(); |
200 | | |
201 | | // scan and recycle expired stage objects |
202 | | // returns 0 for success otherwise error |
203 | | int recycle_expired_stage_objects(); |
204 | | |
205 | | bool check_recycle_tasks(); |
206 | | |
207 | | private: |
208 | | // returns 0 for success otherwise error |
209 | | int init_obj_store_accessors(); |
210 | | |
211 | | // returns 0 for success otherwise error |
212 | | int init_storage_vault_accessors(); |
213 | | |
214 | | /** |
215 | | * Scan key-value pairs between [`begin`, `end`), and perform `recycle_func` on each key-value pair. |
216 | | * |
217 | | * @param recycle_func defines how to recycle resources corresponding to a key-value pair. Returns 0 if the recycling is successful. |
218 | | * @param loop_done is called after `RangeGetIterator` has no next kv. Usually used to perform a batch recycling. Returns 0 if success. |
219 | | * @return 0 if all corresponding resources are recycled successfully, otherwise non-zero |
220 | | */ |
221 | | int scan_and_recycle(std::string begin, std::string_view end, |
222 | | std::function<int(std::string_view k, std::string_view v)> recycle_func, |
223 | | std::function<int()> loop_done = nullptr); |
224 | | |
225 | | int scan_for_recycle_and_statistics( |
226 | | std::string begin, std::string_view end, std::string task_name, |
227 | | RecyclerMetricsContext& metrics_context, |
228 | | std::function<int(std::string_view k, std::string_view v)> statistics_func, |
229 | | std::function<int(std::string_view k, std::string_view v)> recycle_func, |
230 | | std::function<int()> loop_done = nullptr); |
231 | | |
232 | | // return 0 for success otherwise error |
233 | | int delete_rowset_data(const doris::RowsetMetaCloudPB& rs_meta_pb); |
234 | | |
235 | | // return 0 for success otherwise error |
236 | | // NOTE: this function ONLY be called when the file paths cannot be calculated |
237 | | int delete_rowset_data(const std::string& resource_id, int64_t tablet_id, |
238 | | const std::string& rowset_id); |
239 | | |
240 | | // return 0 for success otherwise error |
241 | | int delete_rowset_data(const std::map<std::string, doris::RowsetMetaCloudPB>& rowsets, |
242 | | RowsetRecyclingState type, RecyclerMetricsContext& metrics_context); |
243 | | |
244 | | /** |
245 | | * Get stage storage info from instance and init StorageVaultAccessor |
246 | | * @return 0 if accessor is successfully inited, 1 if stage not found, negative for error |
247 | | */ |
248 | | int init_copy_job_accessor(const std::string& stage_id, const StagePB::StageType& stage_type, |
249 | | std::shared_ptr<StorageVaultAccessor>* accessor); |
250 | | |
251 | | void register_recycle_task(const std::string& task_name, int64_t start_time); |
252 | | |
253 | | void unregister_recycle_task(const std::string& task_name); |
254 | | |
255 | | // for scan all tablets and statistics metrics |
256 | | int scan_tablets_and_statistics(int64_t tablet_id, int64_t index_id, |
257 | | RecyclerMetricsContext& metrics_context, |
258 | | int64_t partition_id = -1, bool is_empty_tablet = false); |
259 | | |
260 | | // for scan all rs of tablet and statistics metrics |
261 | | int scan_tablet_and_statistics(int64_t tablet_id, RecyclerMetricsContext& metrics_context); |
262 | | |
263 | | private: |
264 | | std::atomic_bool stopped_ {false}; |
265 | | std::shared_ptr<TxnKv> txn_kv_; |
266 | | std::string instance_id_; |
267 | | InstanceInfoPB instance_info_; |
268 | | |
269 | | // TODO(plat1ko): Add new accessor to map in runtime for new created storage vaults |
270 | | std::unordered_map<std::string, std::shared_ptr<StorageVaultAccessor>> accessor_map_; |
271 | | using InvertedIndexInfo = |
272 | | std::pair<InvertedIndexStorageFormatPB, std::vector<std::pair<int64_t, std::string>>>; |
273 | | |
274 | | class InvertedIndexIdCache; |
275 | | std::unique_ptr<InvertedIndexIdCache> inverted_index_id_cache_; |
276 | | |
277 | | std::mutex recycled_tablets_mtx_; |
278 | | // Store recycled tablets, we can skip deleting rowset data of these tablets because these data has already been deleted. |
279 | | std::unordered_set<int64_t> recycled_tablets_; |
280 | | |
281 | | std::mutex recycle_tasks_mutex; |
282 | | // <task_name, start_time>> |
283 | | std::map<std::string, int64_t> running_recycle_tasks; |
284 | | |
285 | | RecyclerThreadPoolGroup _thread_pool_group; |
286 | | |
287 | | std::shared_ptr<TxnLazyCommitter> txn_lazy_committer_; |
288 | | }; |
289 | | |
290 | | class RecyclerMetricsContext { |
291 | | public: |
292 | 1 | RecyclerMetricsContext() = default; |
293 | | |
294 | | RecyclerMetricsContext(std::string instance_id, std::string operation_type) |
295 | 146 | : operation_type(std::move(operation_type)), instance_id(std::move(instance_id)) { |
296 | 146 | start(); |
297 | 146 | } |
298 | | |
299 | 143 | ~RecyclerMetricsContext() = default; |
300 | | |
301 | | int64_t total_need_recycle_data_size = 0; |
302 | | int64_t total_need_recycle_num = 0; |
303 | | |
304 | | std::atomic_long total_recycled_data_size {0}; |
305 | | std::atomic_long total_recycled_num {0}; |
306 | | |
307 | | std::string operation_type {}; |
308 | | std::string instance_id {}; |
309 | | |
310 | | double start_time = 0; |
311 | | |
312 | 147 | void start() { |
313 | 147 | start_time = duration_cast<std::chrono::milliseconds>( |
314 | 147 | std::chrono::system_clock::now().time_since_epoch()) |
315 | 147 | .count(); |
316 | 147 | } |
317 | | |
318 | 163 | double duration() const { |
319 | 163 | return duration_cast<std::chrono::milliseconds>( |
320 | 163 | std::chrono::system_clock::now().time_since_epoch()) |
321 | 163 | .count() - |
322 | 163 | start_time; |
323 | 163 | } |
324 | | |
325 | 20 | void reset() { |
326 | 20 | total_need_recycle_data_size = 0; |
327 | 20 | total_need_recycle_num = 0; |
328 | 20 | total_recycled_data_size.store(0); |
329 | 20 | total_recycled_num.store(0); |
330 | 20 | start_time = duration_cast<std::chrono::milliseconds>( |
331 | 20 | std::chrono::system_clock::now().time_since_epoch()) |
332 | 20 | .count(); |
333 | 20 | } |
334 | | |
335 | 163 | void finish_report() { |
336 | 163 | if (!operation_type.empty()) { |
337 | 163 | double cost = duration(); |
338 | 163 | g_bvar_recycler_instance_last_round_recycle_elpased_ts.put( |
339 | 163 | {instance_id, operation_type}, cost); |
340 | 163 | g_bvar_recycler_instance_recycle_round.put({instance_id, operation_type}, 1); |
341 | 163 | LOG(INFO) << "recycle instance: " << instance_id |
342 | 163 | << ", operation type: " << operation_type << ", cost: " << cost |
343 | 163 | << " ms, total recycled num: " << total_recycled_num.load() |
344 | 163 | << ", total recycled data size: " << total_recycled_data_size.load() |
345 | 163 | << " bytes"; |
346 | 163 | if (total_recycled_num.load()) { |
347 | 23 | g_bvar_recycler_instance_recycle_time_per_resource.put( |
348 | 23 | {instance_id, operation_type}, cost / total_recycled_num.load()); |
349 | 140 | } else { |
350 | 140 | g_bvar_recycler_instance_recycle_time_per_resource.put( |
351 | 140 | {instance_id, operation_type}, -1); |
352 | 140 | } |
353 | 163 | if (total_recycled_data_size.load()) { |
354 | 0 | g_bvar_recycler_instance_recycle_bytes_per_ms.put( |
355 | 0 | {instance_id, operation_type}, total_recycled_data_size.load() / cost); |
356 | 163 | } else { |
357 | 163 | g_bvar_recycler_instance_recycle_bytes_per_ms.put({instance_id, operation_type}, |
358 | 163 | -1); |
359 | 163 | } |
360 | 163 | } |
361 | 163 | } |
362 | | |
363 | | // `is_begin` is used to initialize total num of items need to be recycled |
364 | 26.0k | void report(bool is_begin = false) { |
365 | 26.0k | if (!operation_type.empty()) { |
366 | 25.5k | if (total_need_recycle_data_size > 0) { |
367 | | // is init |
368 | 0 | if (is_begin) { |
369 | 0 | g_bvar_recycler_instance_last_round_to_recycle_bytes.put( |
370 | 0 | {instance_id, operation_type}, total_need_recycle_data_size); |
371 | 0 | } else { |
372 | 0 | g_bvar_recycler_instance_last_round_recycled_bytes.put( |
373 | 0 | {instance_id, operation_type}, total_recycled_data_size.load()); |
374 | 0 | g_bvar_recycler_instance_recycle_total_bytes_since_started.put( |
375 | 0 | {instance_id, operation_type}, total_recycled_data_size.load()); |
376 | 0 | } |
377 | 0 | } |
378 | | |
379 | 25.5k | if (total_need_recycle_num > 0) { |
380 | | // is init |
381 | 0 | if (is_begin) { |
382 | 0 | g_bvar_recycler_instance_last_round_to_recycle_num.put( |
383 | 0 | {instance_id, operation_type}, total_need_recycle_num); |
384 | 0 | } else { |
385 | 0 | g_bvar_recycler_instance_last_round_recycled_num.put( |
386 | 0 | {instance_id, operation_type}, total_recycled_num.load()); |
387 | 0 | g_bvar_recycler_instance_recycle_total_num_since_started.put( |
388 | 0 | {instance_id, operation_type}, total_recycled_num.load()); |
389 | 0 | } |
390 | 0 | } |
391 | 25.5k | } |
392 | 26.0k | } |
393 | | }; |
394 | | |
395 | | } // namespace doris::cloud |