/Users/deen/code/yugabyte-db/src/yb/rocksdb/util/file_reader_writer.h
Line | Count | Source (jump to first uncovered line) |
1 | | // Copyright (c) 2011-present, Facebook, Inc. All rights reserved. |
2 | | // This source code is licensed under the BSD-style license found in the |
3 | | // LICENSE file in the root directory of this source tree. An additional grant |
4 | | // of patent rights can be found in the PATENTS file in the same directory. |
5 | | // |
6 | | // The following only applies to changes made to this file as part of YugaByte development. |
7 | | // |
8 | | // Portions Copyright (c) YugaByte, Inc. |
9 | | // |
10 | | // Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except |
11 | | // in compliance with the License. You may obtain a copy of the License at |
12 | | // |
13 | | // http://www.apache.org/licenses/LICENSE-2.0 |
14 | | // |
15 | | // Unless required by applicable law or agreed to in writing, software distributed under the License |
16 | | // is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express |
17 | | // or implied. See the License for the specific language governing permissions and limitations |
18 | | // under the License. |
19 | | // |
20 | | // Copyright (c) 2011 The LevelDB Authors. All rights reserved. |
21 | | // Use of this source code is governed by a BSD-style license that can be |
22 | | // found in the LICENSE file. See the AUTHORS file for names of contributors. |
23 | | |
24 | | #ifndef YB_ROCKSDB_UTIL_FILE_READER_WRITER_H |
25 | | #define YB_ROCKSDB_UTIL_FILE_READER_WRITER_H |
26 | | |
27 | | #include <string.h> |
28 | | |
29 | | #include <string> |
30 | | |
31 | | #include <gflags/gflags_declare.h> |
32 | | |
33 | | #include "yb/rocksdb/env.h" |
34 | | #include "yb/rocksdb/port/port.h" |
35 | | #include "yb/rocksdb/statistics.h" |
36 | | #include "yb/rocksdb/util/aligned_buffer.h" |
37 | | |
38 | | DECLARE_int32(rocksdb_file_starting_buffer_size); |
39 | | |
40 | | namespace yb { |
41 | | |
42 | | class PriorityThreadPoolSuspender; |
43 | | |
44 | | } |
45 | | |
46 | | namespace rocksdb { |
47 | | |
48 | | class Statistics; |
49 | | class HistogramImpl; |
50 | | |
51 | | std::unique_ptr<RandomAccessFile> NewReadaheadRandomAccessFile( |
52 | | std::unique_ptr<RandomAccessFile>&& file, size_t readahead_size); |
53 | | |
54 | | class SequentialFileReader { |
55 | | private: |
56 | | std::unique_ptr<SequentialFile> file_; |
57 | | |
58 | | public: |
59 | | explicit SequentialFileReader(std::unique_ptr<SequentialFile>&& _file) |
60 | 376k | : file_(std::move(_file)) {} |
61 | | |
62 | 0 | SequentialFileReader(SequentialFileReader&& o) ROCKSDB_NOEXCEPT { |
63 | 0 | *this = std::move(o); |
64 | 0 | } |
65 | | |
66 | 0 | SequentialFileReader& operator=(SequentialFileReader&& o) ROCKSDB_NOEXCEPT { |
67 | 0 | file_ = std::move(o.file_); |
68 | 0 | return *this; |
69 | 0 | } |
70 | | |
71 | | SequentialFileReader(const SequentialFileReader&) = delete; |
72 | | SequentialFileReader& operator=(const SequentialFileReader&) = delete; |
73 | | |
74 | | Status Read(size_t n, Slice* result, uint8_t* scratch); |
75 | | |
76 | | Status Skip(uint64_t n); |
77 | | |
78 | 26 | SequentialFile* file() { return file_.get(); } |
79 | | }; |
80 | | |
81 | | class RandomAccessFileReader { |
82 | | private: |
83 | | std::unique_ptr<RandomAccessFile> file_; |
84 | | Env* env_; |
85 | | Statistics* stats_; |
86 | | uint32_t hist_type_; |
87 | | HistogramImpl* file_read_hist_; |
88 | | |
89 | | public: |
90 | | explicit RandomAccessFileReader(std::unique_ptr<RandomAccessFile>&& raf, |
91 | | Env* env = nullptr, |
92 | | Statistics* stats = nullptr, |
93 | | Histograms hist_type = HISTOGRAM_ENUM_MAX, |
94 | | HistogramImpl* file_read_hist = nullptr) |
95 | | : file_(std::move(raf)), |
96 | | env_(env), |
97 | | stats_(stats), |
98 | | hist_type_(hist_type), |
99 | 169k | file_read_hist_(file_read_hist) {} |
100 | | |
101 | 0 | RandomAccessFileReader(RandomAccessFileReader&& o) ROCKSDB_NOEXCEPT { |
102 | 0 | *this = std::move(o); |
103 | 0 | } |
104 | | |
105 | 0 | RandomAccessFileReader& operator=(RandomAccessFileReader&& o) ROCKSDB_NOEXCEPT { |
106 | 0 | file_ = std::move(o.file_); |
107 | 0 | env_ = std::move(o.env_); |
108 | 0 | stats_ = std::move(o.stats_); |
109 | 0 | hist_type_ = std::move(o.hist_type_); |
110 | 0 | file_read_hist_ = std::move(o.file_read_hist_); |
111 | 0 | return *this; |
112 | 0 | } |
113 | | |
114 | | RandomAccessFileReader(const RandomAccessFileReader&) = delete; |
115 | | RandomAccessFileReader& operator=(const RandomAccessFileReader&) = delete; |
116 | | |
117 | | CHECKED_STATUS Read(uint64_t offset, size_t n, Slice* result, char* scratch) const; |
118 | | CHECKED_STATUS ReadAndValidate( |
119 | | uint64_t offset, size_t n, Slice* result, char* scratch, const yb::ReadValidator& validator); |
120 | | |
121 | 365k | RandomAccessFile* file() { return file_.get(); } |
122 | | }; |
123 | | |
124 | | // Use posix write to write data to a file. |
125 | | class WritableFileWriter { |
126 | | private: |
127 | | std::unique_ptr<WritableFile> writable_file_; |
128 | | AlignedBuffer buf_; |
129 | | size_t max_buffer_size_; |
130 | | // Actually written data size can be used for truncate |
131 | | // not counting padding data |
132 | | uint64_t filesize_; |
133 | | // This is necessary when we use unbuffered access |
134 | | // and writes must happen on aligned offsets |
135 | | // so we need to go back and write that page again |
136 | | uint64_t next_write_offset_; |
137 | | bool pending_sync_; |
138 | | bool pending_fsync_; |
139 | | const bool direct_io_; |
140 | | const bool use_os_buffer_; |
141 | | uint64_t last_sync_size_; |
142 | | uint64_t bytes_per_sync_; |
143 | | RateLimiter* rate_limiter_; |
144 | | // When the writer is used by the priority thread pool's task, this task could pass provided |
145 | | // suspender to the writer, so it will be used by writer to check whether task should be |
146 | | // paused after block is flushed. |
147 | | yb::PriorityThreadPoolSuspender* suspender_; |
148 | | |
149 | | public: |
150 | | WritableFileWriter(std::unique_ptr<WritableFile>&& file, |
151 | | const EnvOptions& options, |
152 | | yb::PriorityThreadPoolSuspender* suspender = nullptr) |
153 | | : writable_file_(std::move(file)), |
154 | | buf_(), |
155 | | max_buffer_size_(options.writable_file_max_buffer_size), |
156 | | filesize_(0), |
157 | | next_write_offset_(0), |
158 | | pending_sync_(false), |
159 | | pending_fsync_(false), |
160 | | direct_io_(writable_file_->UseDirectIO()), |
161 | | use_os_buffer_(writable_file_->UseOSBuffer()), |
162 | | last_sync_size_(0), |
163 | | bytes_per_sync_(options.bytes_per_sync), |
164 | | rate_limiter_(options.rate_limiter), |
165 | 1.09M | suspender_(suspender) { |
166 | | |
167 | 1.09M | buf_.Alignment(writable_file_->GetRequiredBufferAlignment()); |
168 | 1.09M | buf_.AllocateNewBuffer(FLAGS_rocksdb_file_starting_buffer_size); |
169 | 1.09M | } |
170 | | |
171 | | WritableFileWriter(const WritableFileWriter&) = delete; |
172 | | |
173 | | WritableFileWriter& operator=(const WritableFileWriter&) = delete; |
174 | | |
175 | | ~WritableFileWriter(); |
176 | | |
177 | | Status Append(const Slice& data); |
178 | | |
179 | | Status Flush(); |
180 | | |
181 | | Status Close(); |
182 | | |
183 | | Status Sync(bool use_fsync); |
184 | | |
185 | | // Sync only the data that was already Flush()ed. Safe to call concurrently |
186 | | // with Append() and Flush(). If !writable_file_->IsSyncThreadSafe(), |
187 | | // returns NotSupported status. |
188 | | Status SyncWithoutFlush(bool use_fsync); |
189 | | |
190 | 48.3M | uint64_t GetFileSize() { return filesize_; } |
191 | | |
192 | | Status InvalidateCache(size_t offset, size_t length); |
193 | | |
194 | 79.1k | WritableFile* writable_file() const { return writable_file_.get(); } |
195 | | |
196 | | private: |
197 | | // Used when os buffering is OFF and we are writing |
198 | | // DMA such as in Windows unbuffered mode |
199 | | Status WriteUnbuffered(); |
200 | | // Normal write |
201 | | Status WriteBuffered(const char* data, size_t size); |
202 | | Status RangeSync(uint64_t offset, uint64_t nbytes); |
203 | | size_t RequestToken(size_t bytes, bool align); |
204 | | Status SyncInternal(bool use_fsync); |
205 | | }; |
206 | | |
207 | | extern Status NewWritableFile(Env* env, const std::string& fname, |
208 | | unique_ptr<WritableFile>* result, |
209 | | const EnvOptions& options); |
210 | | } // namespace rocksdb |
211 | | |
212 | | #endif // YB_ROCKSDB_UTIL_FILE_READER_WRITER_H |