]>
Commit | Line | Data |
---|---|---|
7c673cae | 1 | // Copyright (c) 2011-present, Facebook, Inc. All rights reserved. |
11fdf7f2 TL |
2 | // This source code is licensed under both the GPLv2 (found in the |
3 | // COPYING file in the root directory) and Apache 2.0 License | |
4 | // (found in the LICENSE.Apache file in the root directory). | |
7c673cae | 5 | |
11fdf7f2 | 6 | #pragma once |
7c673cae FG |
7 | |
8 | #include <deque> | |
9 | #include <memory> | |
10 | #include <string> | |
11 | #include <vector> | |
12 | ||
13 | #include "rocksdb/slice.h" | |
14 | ||
15 | namespace rocksdb { | |
16 | ||
17 | class Slice; | |
18 | class Logger; | |
19 | ||
20 | // The Merge Operator | |
21 | // | |
22 | // Essentially, a MergeOperator specifies the SEMANTICS of a merge, which only | |
23 | // client knows. It could be numeric addition, list append, string | |
24 | // concatenation, edit data structure, ... , anything. | |
25 | // The library, on the other hand, is concerned with the exercise of this | |
26 | // interface, at the right time (during get, iteration, compaction...) | |
27 | // | |
28 | // To use merge, the client needs to provide an object implementing one of | |
29 | // the following interfaces: | |
30 | // a) AssociativeMergeOperator - for most simple semantics (always take | |
31 | // two values, and merge them into one value, which is then put back | |
32 | // into rocksdb); numeric addition and string concatenation are examples; | |
33 | // | |
34 | // b) MergeOperator - the generic class for all the more abstract / complex | |
35 | // operations; one method (FullMergeV2) to merge a Put/Delete value with a | |
36 | // merge operand; and another method (PartialMerge) that merges multiple | |
37 | // operands together. this is especially useful if your key values have | |
38 | // complex structures but you would still like to support client-specific | |
39 | // incremental updates. | |
40 | // | |
41 | // AssociativeMergeOperator is simpler to implement. MergeOperator is simply | |
42 | // more powerful. | |
43 | // | |
44 | // Refer to rocksdb-merge wiki for more details and example implementations. | |
45 | // | |
46 | class MergeOperator { | |
47 | public: | |
48 | virtual ~MergeOperator() {} | |
49 | ||
50 | // Gives the client a way to express the read -> modify -> write semantics | |
51 | // key: (IN) The key that's associated with this merge operation. | |
52 | // Client could multiplex the merge operator based on it | |
53 | // if the key space is partitioned and different subspaces | |
54 | // refer to different types of data which have different | |
55 | // merge operation semantics | |
56 | // existing: (IN) null indicates that the key does not exist before this op | |
57 | // operand_list:(IN) the sequence of merge operations to apply, front() first. | |
58 | // new_value:(OUT) Client is responsible for filling the merge result here. | |
59 | // The string that new_value is pointing to will be empty. | |
60 | // logger: (IN) Client could use this to log errors during merge. | |
61 | // | |
62 | // Return true on success. | |
63 | // All values passed in will be client-specific values. So if this method | |
64 | // returns false, it is because client specified bad data or there was | |
65 | // internal corruption. This will be treated as an error by the library. | |
66 | // | |
67 | // Also make use of the *logger for error messages. | |
11fdf7f2 TL |
68 | virtual bool FullMerge(const Slice& /*key*/, const Slice* /*existing_value*/, |
69 | const std::deque<std::string>& /*operand_list*/, | |
70 | std::string* /*new_value*/, Logger* /*logger*/) const { | |
7c673cae FG |
71 | // deprecated, please use FullMergeV2() |
72 | assert(false); | |
73 | return false; | |
74 | } | |
75 | ||
76 | struct MergeOperationInput { | |
77 | explicit MergeOperationInput(const Slice& _key, | |
78 | const Slice* _existing_value, | |
79 | const std::vector<Slice>& _operand_list, | |
80 | Logger* _logger) | |
81 | : key(_key), | |
82 | existing_value(_existing_value), | |
83 | operand_list(_operand_list), | |
84 | logger(_logger) {} | |
85 | ||
86 | // The key associated with the merge operation. | |
87 | const Slice& key; | |
88 | // The existing value of the current key, nullptr means that the | |
11fdf7f2 | 89 | // value doesn't exist. |
7c673cae FG |
90 | const Slice* existing_value; |
91 | // A list of operands to apply. | |
92 | const std::vector<Slice>& operand_list; | |
93 | // Logger could be used by client to log any errors that happen during | |
94 | // the merge operation. | |
95 | Logger* logger; | |
96 | }; | |
97 | ||
98 | struct MergeOperationOutput { | |
99 | explicit MergeOperationOutput(std::string& _new_value, | |
100 | Slice& _existing_operand) | |
101 | : new_value(_new_value), existing_operand(_existing_operand) {} | |
102 | ||
103 | // Client is responsible for filling the merge result here. | |
104 | std::string& new_value; | |
105 | // If the merge result is one of the existing operands (or existing_value), | |
106 | // client can set this field to the operand (or existing_value) instead of | |
107 | // using new_value. | |
108 | Slice& existing_operand; | |
109 | }; | |
110 | ||
494da23a TL |
111 | // This function applies a stack of merge operands in chrionological order |
112 | // on top of an existing value. There are two ways in which this method is | |
113 | // being used: | |
114 | // a) During Get() operation, it used to calculate the final value of a key | |
115 | // b) During compaction, in order to collapse some operands with the based | |
116 | // value. | |
117 | // | |
118 | // Note: The name of the method is somewhat misleading, as both in the cases | |
119 | // of Get() or compaction it may be called on a subset of operands: | |
120 | // K: 0 +1 +2 +7 +4 +5 2 +1 +2 | |
121 | // ^ | |
122 | // | | |
123 | // snapshot | |
124 | // In the example above, Get(K) operation will call FullMerge with a base | |
125 | // value of 2 and operands [+1, +2]. Compaction process might decide to | |
126 | // collapse the beginning of the history up to the snapshot by performing | |
127 | // full Merge with base value of 0 and operands [+1, +2, +7, +3]. | |
7c673cae FG |
128 | virtual bool FullMergeV2(const MergeOperationInput& merge_in, |
129 | MergeOperationOutput* merge_out) const; | |
130 | ||
131 | // This function performs merge(left_op, right_op) | |
132 | // when both the operands are themselves merge operation types | |
133 | // that you would have passed to a DB::Merge() call in the same order | |
134 | // (i.e.: DB::Merge(key,left_op), followed by DB::Merge(key,right_op)). | |
135 | // | |
136 | // PartialMerge should combine them into a single merge operation that is | |
137 | // saved into *new_value, and then it should return true. | |
138 | // *new_value should be constructed such that a call to | |
139 | // DB::Merge(key, *new_value) would yield the same result as a call | |
140 | // to DB::Merge(key, left_op) followed by DB::Merge(key, right_op). | |
141 | // | |
142 | // The string that new_value is pointing to will be empty. | |
143 | // | |
144 | // The default implementation of PartialMergeMulti will use this function | |
145 | // as a helper, for backward compatibility. Any successor class of | |
146 | // MergeOperator should either implement PartialMerge or PartialMergeMulti, | |
147 | // although implementing PartialMergeMulti is suggested as it is in general | |
148 | // more effective to merge multiple operands at a time instead of two | |
149 | // operands at a time. | |
150 | // | |
151 | // If it is impossible or infeasible to combine the two operations, | |
152 | // leave new_value unchanged and return false. The library will | |
153 | // internally keep track of the operations, and apply them in the | |
154 | // correct order once a base-value (a Put/Delete/End-of-Database) is seen. | |
155 | // | |
156 | // TODO: Presently there is no way to differentiate between error/corruption | |
157 | // and simply "return false". For now, the client should simply return | |
158 | // false in any case it cannot perform partial-merge, regardless of reason. | |
159 | // If there is corruption in the data, handle it in the FullMergeV2() function | |
160 | // and return false there. The default implementation of PartialMerge will | |
161 | // always return false. | |
11fdf7f2 TL |
162 | virtual bool PartialMerge(const Slice& /*key*/, const Slice& /*left_operand*/, |
163 | const Slice& /*right_operand*/, | |
164 | std::string* /*new_value*/, | |
165 | Logger* /*logger*/) const { | |
7c673cae FG |
166 | return false; |
167 | } | |
168 | ||
169 | // This function performs merge when all the operands are themselves merge | |
170 | // operation types that you would have passed to a DB::Merge() call in the | |
171 | // same order (front() first) | |
172 | // (i.e. DB::Merge(key, operand_list[0]), followed by | |
173 | // DB::Merge(key, operand_list[1]), ...) | |
174 | // | |
175 | // PartialMergeMulti should combine them into a single merge operation that is | |
176 | // saved into *new_value, and then it should return true. *new_value should | |
177 | // be constructed such that a call to DB::Merge(key, *new_value) would yield | |
178 | // the same result as subquential individual calls to DB::Merge(key, operand) | |
179 | // for each operand in operand_list from front() to back(). | |
180 | // | |
181 | // The string that new_value is pointing to will be empty. | |
182 | // | |
183 | // The PartialMergeMulti function will be called when there are at least two | |
184 | // operands. | |
185 | // | |
186 | // In the default implementation, PartialMergeMulti will invoke PartialMerge | |
187 | // multiple times, where each time it only merges two operands. Developers | |
188 | // should either implement PartialMergeMulti, or implement PartialMerge which | |
189 | // is served as the helper function of the default PartialMergeMulti. | |
190 | virtual bool PartialMergeMulti(const Slice& key, | |
191 | const std::deque<Slice>& operand_list, | |
192 | std::string* new_value, Logger* logger) const; | |
193 | ||
194 | // The name of the MergeOperator. Used to check for MergeOperator | |
195 | // mismatches (i.e., a DB created with one MergeOperator is | |
196 | // accessed using a different MergeOperator) | |
197 | // TODO: the name is currently not stored persistently and thus | |
198 | // no checking is enforced. Client is responsible for providing | |
199 | // consistent MergeOperator between DB opens. | |
200 | virtual const char* Name() const = 0; | |
11fdf7f2 | 201 | |
494da23a | 202 | // Determines whether the PartialMerge can be called with just a single |
11fdf7f2 | 203 | // merge operand. |
494da23a TL |
204 | // Override and return true for allowing a single operand. PartialMerge |
205 | // and PartialMergeMulti should be overridden and implemented | |
206 | // correctly to properly handle a single operand. | |
11fdf7f2 TL |
207 | virtual bool AllowSingleOperand() const { return false; } |
208 | ||
209 | // Allows to control when to invoke a full merge during Get. | |
210 | // This could be used to limit the number of merge operands that are looked at | |
211 | // during a point lookup, thereby helping in limiting the number of levels to | |
212 | // read from. | |
213 | // Doesn't help with iterators. | |
214 | // | |
215 | // Note: the merge operands are passed to this function in the reversed order | |
216 | // relative to how they were merged (passed to FullMerge or FullMergeV2) | |
217 | // for performance reasons, see also: | |
218 | // https://github.com/facebook/rocksdb/issues/3865 | |
219 | virtual bool ShouldMerge(const std::vector<Slice>& /*operands*/) const { | |
220 | return false; | |
221 | } | |
7c673cae FG |
222 | }; |
223 | ||
224 | // The simpler, associative merge operator. | |
225 | class AssociativeMergeOperator : public MergeOperator { | |
226 | public: | |
11fdf7f2 | 227 | ~AssociativeMergeOperator() override {} |
7c673cae FG |
228 | |
229 | // Gives the client a way to express the read -> modify -> write semantics | |
230 | // key: (IN) The key that's associated with this merge operation. | |
231 | // existing_value:(IN) null indicates the key does not exist before this op | |
232 | // value: (IN) the value to update/merge the existing_value with | |
233 | // new_value: (OUT) Client is responsible for filling the merge result | |
234 | // here. The string that new_value is pointing to will be empty. | |
235 | // logger: (IN) Client could use this to log errors during merge. | |
236 | // | |
237 | // Return true on success. | |
238 | // All values passed in will be client-specific values. So if this method | |
239 | // returns false, it is because client specified bad data or there was | |
240 | // internal corruption. The client should assume that this will be treated | |
241 | // as an error by the library. | |
494da23a TL |
242 | virtual bool Merge(const Slice& key, const Slice* existing_value, |
243 | const Slice& value, std::string* new_value, | |
7c673cae FG |
244 | Logger* logger) const = 0; |
245 | ||
7c673cae FG |
246 | private: |
247 | // Default implementations of the MergeOperator functions | |
11fdf7f2 TL |
248 | bool FullMergeV2(const MergeOperationInput& merge_in, |
249 | MergeOperationOutput* merge_out) const override; | |
7c673cae | 250 | |
11fdf7f2 TL |
251 | bool PartialMerge(const Slice& key, const Slice& left_operand, |
252 | const Slice& right_operand, std::string* new_value, | |
253 | Logger* logger) const override; | |
7c673cae FG |
254 | }; |
255 | ||
256 | } // namespace rocksdb |