/Users/deen/code/yugabyte-db/src/yb/docdb/doc_write_batch.h
Line | Count | Source (jump to first uncovered line) |
1 | | // Copyright (c) YugaByte, Inc. |
2 | | // |
3 | | // Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except |
4 | | // in compliance with the License. You may obtain a copy of the License at |
5 | | // |
6 | | // http://www.apache.org/licenses/LICENSE-2.0 |
7 | | // |
8 | | // Unless required by applicable law or agreed to in writing, software distributed under the License |
9 | | // is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express |
10 | | // or implied. See the License for the specific language governing permissions and limitations |
11 | | // under the License. |
12 | | // |
13 | | |
14 | | #ifndef YB_DOCDB_DOC_WRITE_BATCH_H |
15 | | #define YB_DOCDB_DOC_WRITE_BATCH_H |
16 | | |
17 | | #include "yb/bfql/tserver_opcodes.h" |
18 | | |
19 | | #include "yb/common/constants.h" |
20 | | #include "yb/common/hybrid_time.h" |
21 | | #include "yb/common/read_hybrid_time.h" |
22 | | |
23 | | #include "yb/docdb/doc_write_batch_cache.h" |
24 | | #include "yb/docdb/docdb_types.h" |
25 | | #include "yb/docdb/intent_aware_iterator.h" |
26 | | #include "yb/docdb/key_bounds.h" |
27 | | #include "yb/docdb/value.h" |
28 | | |
29 | | #include "yb/rocksdb/cache.h" |
30 | | |
31 | | #include "yb/rocksutil/write_batch_formatter.h" |
32 | | |
33 | | #include "yb/util/enums.h" |
34 | | #include "yb/util/monotime.h" |
35 | | |
36 | | namespace rocksdb { |
37 | | class DB; |
38 | | } |
39 | | |
40 | | namespace yb { |
41 | | namespace docdb { |
42 | | |
43 | | class KeyValueWriteBatchPB; |
44 | | class IntentAwareIterator; |
45 | | |
46 | | struct LazyIterator { |
47 | | std::function<std::unique_ptr<IntentAwareIterator>()>* creator; |
48 | | std::unique_ptr<IntentAwareIterator> iterator; |
49 | | |
50 | | explicit LazyIterator(std::function<std::unique_ptr<IntentAwareIterator>()>* c) |
51 | 71.6M | : iterator(nullptr) { |
52 | 71.6M | creator = c; |
53 | 71.6M | } |
54 | | |
55 | 13 | explicit LazyIterator(std::unique_ptr<IntentAwareIterator> i) { |
56 | 13 | iterator = std::move(i); |
57 | 13 | } |
58 | | |
59 | 71.6M | ~LazyIterator() {} |
60 | | |
61 | 411k | IntentAwareIterator* Iterator() { |
62 | 411k | if (!iterator) |
63 | 270k | iterator = (*creator)(); |
64 | 411k | return iterator.get(); |
65 | 411k | } |
66 | | }; |
67 | | |
68 | | YB_DEFINE_ENUM(ValueRefType, (kPb)(kValueType)); |
69 | | |
70 | | // This class references value that should be inserted to DocWriteBatch. |
71 | | // Also it contains various options for this value. |
72 | | class ValueRef { |
73 | | public: |
74 | | explicit ValueRef(const QLValuePB& value_pb, |
75 | | SortingType sorting_type = SortingType::kNotSpecified, |
76 | | bfql::TSOpcode write_instruction = bfql::TSOpcode::kScalarInsert) |
77 | | : value_pb_(&value_pb), sorting_type_(sorting_type), write_instruction_(write_instruction), |
78 | 55.2M | list_extend_order_(ListExtendOrder::APPEND), value_type_(ValueType::kLowest) { |
79 | 55.2M | } |
80 | | |
81 | | explicit ValueRef(const QLValuePB& value_pb, |
82 | | const ValueRef& value_ref) |
83 | | : value_pb_(&value_pb), sorting_type_(value_ref.sorting_type_), |
84 | | write_instruction_(value_ref.write_instruction_), |
85 | 217k | list_extend_order_(value_ref.list_extend_order_), value_type_(ValueType::kLowest) { |
86 | 217k | } |
87 | | |
88 | | explicit ValueRef(const QLValuePB& value_pb, |
89 | | ListExtendOrder list_extend_order) |
90 | | : value_pb_(&value_pb), sorting_type_(SortingType::kNotSpecified), |
91 | | write_instruction_(bfql::TSOpcode::kScalarInsert), |
92 | | list_extend_order_(list_extend_order), |
93 | | value_type_(ValueType::kLowest) { |
94 | | } |
95 | | |
96 | | explicit ValueRef(ValueType value_type); |
97 | | |
98 | 71.7M | const QLValuePB& value_pb() const { |
99 | 71.7M | return *value_pb_; |
100 | 71.7M | } |
101 | | |
102 | 0 | void set_sorting_type(SortingType value) { |
103 | 0 | sorting_type_ = value; |
104 | 0 | } |
105 | | |
106 | 71.8M | SortingType sorting_type() const { |
107 | 71.8M | return sorting_type_; |
108 | 71.8M | } |
109 | | |
110 | 218k | ListExtendOrder list_extend_order() const { |
111 | 218k | return list_extend_order_; |
112 | 218k | } |
113 | | |
114 | 201 | void set_list_extend_order(ListExtendOrder value) { |
115 | 201 | list_extend_order_ = value; |
116 | 201 | } |
117 | | |
118 | 21.3k | void set_custom_value_type(ValueType value) { |
119 | 21.3k | value_type_ = value; |
120 | 21.3k | } |
121 | | |
122 | 160M | ValueType custom_value_type() const { |
123 | 160M | return value_type_; |
124 | 160M | } |
125 | | |
126 | 217k | bfql::TSOpcode write_instruction() const { |
127 | 217k | return write_instruction_; |
128 | 217k | } |
129 | | |
130 | 16 | void set_write_instruction(bfql::TSOpcode value) { |
131 | 16 | write_instruction_ = value; |
132 | 16 | } |
133 | | |
134 | | bool is_array() const; |
135 | | |
136 | | bool is_set() const; |
137 | | |
138 | | bool is_map() const; |
139 | | |
140 | | ValueType ContainerValueType() const; |
141 | | |
142 | | bool IsTombstoneOrPrimitive() const; |
143 | | |
144 | | std::string ToString() const; |
145 | | |
146 | | private: |
147 | | const QLValuePB* value_pb_; |
148 | | SortingType sorting_type_; |
149 | | bfql::TSOpcode write_instruction_; |
150 | | ListExtendOrder list_extend_order_; |
151 | | ValueType value_type_; |
152 | | }; |
153 | | |
154 | | // This controls whether "init markers" are required at all intermediate levels. |
155 | | YB_DEFINE_ENUM(InitMarkerBehavior, |
156 | | // This is used in Redis. We need to keep track of document types such as strings, |
157 | | // hashes, sets, because there is no schema and due to Redis's error checking. |
158 | | (kRequired) |
159 | | |
160 | | // This is used in CQL. Existence of "a.b.c" implies existence of "a" and "a.b", |
161 | | // unless there are delete markers / TTL expiration involved. |
162 | | (kOptional)); |
163 | | |
164 | | // The DocWriteBatch class is used to build a RocksDB write batch for a DocDB batch of operations |
165 | | // that may include a mix of write (set) or delete operations. It may read from RocksDB while |
166 | | // writing, and builds up an internal rocksdb::WriteBatch while handling the operations. |
167 | | // When all the operations are applied, the rocksdb::WriteBatch should be taken as output. |
168 | | // Take ownership of it using std::move if it needs to live longer than this DocWriteBatch. |
169 | | class DocWriteBatch { |
170 | | public: |
171 | | explicit DocWriteBatch(const DocDB& doc_db, |
172 | | InitMarkerBehavior init_marker_behavior, |
173 | | std::atomic<int64_t>* monotonic_counter = nullptr); |
174 | | |
175 | | Status SeekToKeyPrefix(LazyIterator* doc_iter, bool has_ancestor = false); |
176 | | Status SeekToKeyPrefix(IntentAwareIterator* doc_iter, bool has_ancestor); |
177 | | |
178 | | // Set the primitive at the given path to the given value. Intermediate subdocuments are created |
179 | | // if necessary and possible. |
180 | | CHECKED_STATUS SetPrimitive( |
181 | | const DocPath& doc_path, |
182 | | const ValueControlFields& control_fields, |
183 | | const ValueRef& value, |
184 | | LazyIterator* doc_iter); |
185 | | |
186 | | CHECKED_STATUS SetPrimitive( |
187 | | const DocPath& doc_path, |
188 | | const ValueControlFields& control_fields, |
189 | | const ValueRef& value, |
190 | | const ReadHybridTime& read_ht = ReadHybridTime::Max(), |
191 | | const CoarseTimePoint deadline = CoarseTimePoint::max(), |
192 | | rocksdb::QueryId query_id = rocksdb::kDefaultQueryId); |
193 | | |
194 | | CHECKED_STATUS SetPrimitive( |
195 | | const DocPath& doc_path, |
196 | | const ValueControlFields& control_fields, |
197 | | const ValueRef& value, |
198 | 13 | std::unique_ptr<IntentAwareIterator> intent_iter) { |
199 | 13 | LazyIterator iter(std::move(intent_iter)); |
200 | 13 | return SetPrimitive(doc_path, control_fields, value, &iter); |
201 | 13 | } |
202 | | |
203 | | |
204 | | CHECKED_STATUS SetPrimitive( |
205 | | const DocPath& doc_path, |
206 | | const ValueRef& value, |
207 | | const ReadHybridTime& read_ht = ReadHybridTime::Max(), |
208 | | const CoarseTimePoint deadline = CoarseTimePoint::max(), |
209 | | rocksdb::QueryId query_id = rocksdb::kDefaultQueryId, |
210 | 1.54M | UserTimeMicros user_timestamp = ValueControlFields::kInvalidUserTimestamp) { |
211 | 1.54M | return SetPrimitive( |
212 | 1.54M | doc_path, ValueControlFields { .user_timestamp = user_timestamp }, value, read_ht, |
213 | 1.54M | deadline, query_id); |
214 | 1.54M | } |
215 | | |
216 | | // Extend the SubDocument in the given key. We'll support List with Append and Prepend mode later. |
217 | | // TODO(akashnil): 03/20/17 ENG-1107 |
218 | | // In each SetPrimitive call, some common work is repeated. It may be made more |
219 | | // efficient by not calling SetPrimitive internally. |
220 | | CHECKED_STATUS ExtendSubDocument( |
221 | | const DocPath& doc_path, |
222 | | const ValueRef& value, |
223 | | const ReadHybridTime& read_ht = ReadHybridTime::Max(), |
224 | | const CoarseTimePoint deadline = CoarseTimePoint::max(), |
225 | | rocksdb::QueryId query_id = rocksdb::kDefaultQueryId, |
226 | | MonoDelta ttl = ValueControlFields::kMaxTtl, |
227 | | UserTimeMicros user_timestamp = ValueControlFields::kInvalidUserTimestamp); |
228 | | |
229 | | CHECKED_STATUS InsertSubDocument( |
230 | | const DocPath& doc_path, |
231 | | const ValueRef& value, |
232 | | const ReadHybridTime& read_ht = ReadHybridTime::Max(), |
233 | | const CoarseTimePoint deadline = CoarseTimePoint::max(), |
234 | | rocksdb::QueryId query_id = rocksdb::kDefaultQueryId, |
235 | | MonoDelta ttl = ValueControlFields::kMaxTtl, |
236 | | UserTimeMicros user_timestamp = ValueControlFields::kInvalidUserTimestamp, |
237 | | bool init_marker_ttl = true); |
238 | | |
239 | | CHECKED_STATUS ExtendList( |
240 | | const DocPath& doc_path, |
241 | | const ValueRef& value, |
242 | | const ReadHybridTime& read_ht = ReadHybridTime::Max(), |
243 | | const CoarseTimePoint deadline = CoarseTimePoint::max(), |
244 | | rocksdb::QueryId query_id = rocksdb::kDefaultQueryId, |
245 | | MonoDelta ttl = ValueControlFields::kMaxTtl, |
246 | | UserTimeMicros user_timestamp = ValueControlFields::kInvalidUserTimestamp); |
247 | | |
248 | | // 'indices' must be sorted. List indexes are not zero indexed, the first element is list[1]. |
249 | | CHECKED_STATUS ReplaceRedisInList( |
250 | | const DocPath& doc_path, |
251 | | int64_t index, |
252 | | const ValueRef& value, |
253 | | const ReadHybridTime& read_ht, |
254 | | const CoarseTimePoint deadline, |
255 | | const rocksdb::QueryId query_id, |
256 | | const Direction dir = Direction::kForward, |
257 | | const int64_t start_index = 0, |
258 | | std::vector<string>* results = nullptr, |
259 | | MonoDelta default_ttl = ValueControlFields::kMaxTtl, |
260 | | MonoDelta write_ttl = ValueControlFields::kMaxTtl); |
261 | | |
262 | | CHECKED_STATUS ReplaceCqlInList( |
263 | | const DocPath &doc_path, |
264 | | const int index, |
265 | | const ValueRef& value, |
266 | | const ReadHybridTime& read_ht, |
267 | | const CoarseTimePoint deadline, |
268 | | const rocksdb::QueryId query_id, |
269 | | MonoDelta default_ttl = ValueControlFields::kMaxTtl, |
270 | | MonoDelta write_ttl = ValueControlFields::kMaxTtl); |
271 | | |
272 | | CHECKED_STATUS DeleteSubDoc( |
273 | | const DocPath& doc_path, |
274 | | const ReadHybridTime& read_ht = ReadHybridTime::Max(), |
275 | | const CoarseTimePoint deadline = CoarseTimePoint::max(), |
276 | | rocksdb::QueryId query_id = rocksdb::kDefaultQueryId, |
277 | | UserTimeMicros user_timestamp = ValueControlFields::kInvalidUserTimestamp); |
278 | | |
279 | | void Clear(); |
280 | 762k | bool IsEmpty() const { return put_batch_.empty(); } |
281 | | |
282 | 0 | size_t size() const { return put_batch_.size(); } |
283 | | |
284 | 1.52M | const std::vector<std::pair<std::string, std::string>>& key_value_pairs() const { |
285 | 1.52M | return put_batch_; |
286 | 1.52M | } |
287 | | |
288 | | void MoveToWriteBatchPB(KeyValueWriteBatchPB *kv_pb); |
289 | | |
290 | | // This method has worse performance comparing to MoveToWriteBatchPB and intented to be used in |
291 | | // testing. Consider using MoveToWriteBatchPB in production code. |
292 | | void TEST_CopyToWriteBatchPB(KeyValueWriteBatchPB *kv_pb) const; |
293 | | |
294 | | // This is used in tests when measuring the number of seeks that a given update to this batch |
295 | | // performs. The internal seek count is reset. |
296 | | int GetAndResetNumRocksDBSeeks(); |
297 | | |
298 | 6.31M | const DocDB& doc_db() { return doc_db_; } |
299 | | |
300 | 41.3k | boost::optional<DocWriteBatchCache::Entry> LookupCache(const KeyBytes& encoded_key_prefix) { |
301 | 41.3k | return cache_.Get(encoded_key_prefix); |
302 | 41.3k | } |
303 | | |
304 | 474 | std::pair<std::string, std::string>& AddRaw() { |
305 | 474 | put_batch_.emplace_back(); |
306 | 474 | return put_batch_.back(); |
307 | 474 | } |
308 | | |
309 | | void UpdateMaxValueTtl(const MonoDelta& ttl); |
310 | | |
311 | 446 | int64_t ttl_ns() const { |
312 | 446 | return ttl_.ToNanoseconds(); |
313 | 446 | } |
314 | | |
315 | 3.11M | bool has_ttl() const { |
316 | 3.11M | return ttl_.Initialized(); |
317 | 3.11M | } |
318 | | |
319 | | private: |
320 | | // This member function performs the necessary operations to set a primitive value for a given |
321 | | // docpath assuming the appropriate operations have been taken care of for subkeys with index < |
322 | | // subkey_index. This method assumes responsibility of ensuring the proper DocDB structure |
323 | | // (e.g: init markers) is maintained for subdocuments starting at the given subkey_index. |
324 | | CHECKED_STATUS SetPrimitiveInternal( |
325 | | const DocPath& doc_path, |
326 | | const ValueControlFields& control_fields, |
327 | | const ValueRef& value, |
328 | | LazyIterator* doc_iter, |
329 | | bool is_deletion, |
330 | | size_t num_subkeys); |
331 | | |
332 | | // Handle the user provided timestamp during writes. |
333 | | Result<bool> SetPrimitiveInternalHandleUserTimestamp(const ValueControlFields& control_fields, |
334 | | LazyIterator* doc_iter); |
335 | | |
336 | 142M | bool required_init_markers() { |
337 | 142M | return init_marker_behavior_ == InitMarkerBehavior::kRequired; |
338 | 142M | } |
339 | | |
340 | 141M | bool optional_init_markers() { |
341 | 141M | return init_marker_behavior_ == InitMarkerBehavior::kOptional; |
342 | 141M | } |
343 | | |
344 | | DocWriteBatchCache cache_; |
345 | | |
346 | | DocDB doc_db_; |
347 | | |
348 | | InitMarkerBehavior init_marker_behavior_; |
349 | | std::atomic<int64_t>* monotonic_counter_; |
350 | | std::vector<std::pair<std::string, std::string>> put_batch_; |
351 | | |
352 | | // Taken from internal_doc_iterator |
353 | | KeyBytes key_prefix_; |
354 | | bool subdoc_exists_ = true; |
355 | | DocWriteBatchCache::Entry current_entry_; |
356 | | |
357 | | MonoDelta ttl_; |
358 | | }; |
359 | | |
360 | | // Converts a RocksDB WriteBatch to a string. |
361 | | // line_prefix is the prefix to be added to each line of the result. Could be used for indentation. |
362 | | Result<std::string> WriteBatchToString( |
363 | | const rocksdb::WriteBatch& write_batch, |
364 | | StorageDbType storage_db_type, |
365 | | BinaryOutputFormat binary_output_format, |
366 | | WriteBatchOutputFormat batch_output_format, |
367 | | const std::string& line_prefix); |
368 | | |
369 | | } // namespace docdb |
370 | | } // namespace yb |
371 | | |
372 | | #endif // YB_DOCDB_DOC_WRITE_BATCH_H |