1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
|
/* Copyright 2015 The TensorFlow Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/
#include "tensorflow/core/lib/io/record_reader.h"
#include <limits.h>
#include "tensorflow/core/lib/core/coding.h"
#include "tensorflow/core/lib/core/errors.h"
#include "tensorflow/core/lib/hash/crc32c.h"
#include "tensorflow/core/lib/io/buffered_inputstream.h"
#include "tensorflow/core/lib/io/compression.h"
#include "tensorflow/core/lib/io/random_inputstream.h"
#include "tensorflow/core/platform/env.h"
namespace tensorflow {
namespace io {
RecordReaderOptions RecordReaderOptions::CreateRecordReaderOptions(
const string& compression_type) {
RecordReaderOptions options;
if (compression_type == "ZLIB") {
options.compression_type = io::RecordReaderOptions::ZLIB_COMPRESSION;
#if defined(IS_SLIM_BUILD)
LOG(ERROR) << "Compression is not supported but compression_type is set."
<< " No compression will be used.";
#else
options.zlib_options = io::ZlibCompressionOptions::DEFAULT();
#endif // IS_SLIM_BUILD
} else if (compression_type == compression::kGzip) {
options.compression_type = io::RecordReaderOptions::ZLIB_COMPRESSION;
#if defined(IS_SLIM_BUILD)
LOG(ERROR) << "Compression is not supported but compression_type is set."
<< " No compression will be used.";
#else
options.zlib_options = io::ZlibCompressionOptions::GZIP();
#endif // IS_SLIM_BUILD
} else if (compression_type != compression::kNone) {
LOG(ERROR) << "Unsupported compression_type:" << compression_type
<< ". No compression will be used.";
}
return options;
}
RecordReader::RecordReader(RandomAccessFile* file,
const RecordReaderOptions& options)
: options_(options),
input_stream_(new RandomAccessInputStream(file)),
last_read_failed_(false) {
if (options.buffer_size > 0) {
input_stream_.reset(new BufferedInputStream(input_stream_.release(),
options.buffer_size, true));
}
if (options.compression_type == RecordReaderOptions::ZLIB_COMPRESSION) {
// We don't have zlib available on all embedded platforms, so fail.
#if defined(IS_SLIM_BUILD)
LOG(FATAL) << "Zlib compression is unsupported on mobile platforms.";
#else // IS_SLIM_BUILD
input_stream_.reset(new ZlibInputStream(
input_stream_.release(), options.zlib_options.input_buffer_size,
options.zlib_options.output_buffer_size, options.zlib_options, true));
#endif // IS_SLIM_BUILD
} else if (options.compression_type == RecordReaderOptions::NONE) {
// Nothing to do.
} else {
LOG(FATAL) << "Unrecognized compression type :" << options.compression_type;
}
}
// Read n+4 bytes from file, verify that checksum of first n bytes is
// stored in the last 4 bytes and store the first n bytes in *result.
//
// offset corresponds to the user-provided value to ReadRecord()
// and is used only in error messages.
Status RecordReader::ReadChecksummed(uint64 offset, size_t n, string* result) {
if (n >= SIZE_MAX - sizeof(uint32)) {
return errors::DataLoss("record size too large");
}
const size_t expected = n + sizeof(uint32);
TF_RETURN_IF_ERROR(input_stream_->ReadNBytes(expected, result));
if (result->size() != expected) {
if (result->empty()) {
return errors::OutOfRange("eof");
} else {
return errors::DataLoss("truncated record at ", offset);
}
}
const uint32 masked_crc = core::DecodeFixed32(result->data() + n);
if (crc32c::Unmask(masked_crc) != crc32c::Value(result->data(), n)) {
return errors::DataLoss("corrupted record at ", offset);
}
result->resize(n);
return Status::OK();
}
Status RecordReader::GetMetadata(Metadata* md) {
if (!md) {
return errors::InvalidArgument(
"Metadata object call to GetMetadata() was null");
}
// Compute the metadata of the TFRecord file if not cached.
if (!cached_metadata_) {
TF_RETURN_IF_ERROR(input_stream_->Reset());
int64 data_size = 0;
int64 entries = 0;
// Within the loop, we always increment offset positively, so this
// loop should be guaranteed to either return after reaching EOF
// or encountering an error.
uint64 offset = 0;
string record;
while (true) {
// Read header, containing size of data.
Status s = ReadChecksummed(offset, sizeof(uint64), &record);
if (!s.ok()) {
if (errors::IsOutOfRange(s)) {
// We should reach out of range when the record file is complete.
break;
}
return s;
}
// Read the length of the data.
const uint64 length = core::DecodeFixed64(record.data());
// Skip reading the actual data since we just want the number
// of records and the size of the data.
TF_RETURN_IF_ERROR(input_stream_->SkipNBytes(length + kFooterSize));
offset += kHeaderSize + length + kFooterSize;
// Increment running stats.
data_size += length;
++entries;
}
cached_metadata_.reset(new Metadata());
cached_metadata_->stats.entries = entries;
cached_metadata_->stats.data_size = data_size;
cached_metadata_->stats.file_size =
data_size + (kHeaderSize + kFooterSize) * entries;
}
md->stats = cached_metadata_->stats;
return Status::OK();
}
Status RecordReader::ReadRecord(uint64* offset, string* record) {
// Position the input stream.
int64 curr_pos = input_stream_->Tell();
int64 desired_pos = static_cast<int64>(*offset);
if (curr_pos > desired_pos || curr_pos < 0 /* EOF */ ||
(curr_pos == desired_pos && last_read_failed_)) {
last_read_failed_ = false;
TF_RETURN_IF_ERROR(input_stream_->Reset());
TF_RETURN_IF_ERROR(input_stream_->SkipNBytes(desired_pos));
} else if (curr_pos < desired_pos) {
TF_RETURN_IF_ERROR(input_stream_->SkipNBytes(desired_pos - curr_pos));
}
DCHECK_EQ(desired_pos, input_stream_->Tell());
// Read header data.
Status s = ReadChecksummed(*offset, sizeof(uint64), record);
if (!s.ok()) {
last_read_failed_ = true;
return s;
}
const uint64 length = core::DecodeFixed64(record->data());
// Read data
s = ReadChecksummed(*offset + kHeaderSize, length, record);
if (!s.ok()) {
last_read_failed_ = true;
if (errors::IsOutOfRange(s)) {
s = errors::DataLoss("truncated record at ", *offset);
}
return s;
}
*offset += kHeaderSize + length + kFooterSize;
DCHECK_EQ(*offset, input_stream_->Tell());
return Status::OK();
}
SequentialRecordReader::SequentialRecordReader(
RandomAccessFile* file, const RecordReaderOptions& options)
: underlying_(file, options), offset_(0) {}
} // namespace io
} // namespace tensorflow
|
