2 * Copyright 2017 Facebook, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 #include <folly/io/Compression.h>
22 #include <unordered_map>
24 #include <boost/noncopyable.hpp>
25 #include <glog/logging.h>
27 #include <folly/Benchmark.h>
28 #include <folly/Hash.h>
29 #include <folly/Random.h>
30 #include <folly/Varint.h>
31 #include <folly/io/IOBufQueue.h>
32 #include <folly/portability/GTest.h>
34 namespace folly { namespace io { namespace test {
36 class DataHolder : private boost::noncopyable {
38 uint64_t hash(size_t size) const;
39 ByteRange data(size_t size) const;
42 explicit DataHolder(size_t sizeLog2);
44 std::unique_ptr<uint8_t[]> data_;
45 mutable std::unordered_map<uint64_t, uint64_t> hashCache_;
48 DataHolder::DataHolder(size_t sizeLog2)
49 : size_(size_t(1) << sizeLog2),
50 data_(new uint8_t[size_]) {
53 uint64_t DataHolder::hash(size_t size) const {
54 CHECK_LE(size, size_);
55 auto p = hashCache_.find(size);
56 if (p != hashCache_.end()) {
60 uint64_t h = folly::hash::fnv64_buf(data_.get(), size);
65 ByteRange DataHolder::data(size_t size) const {
66 CHECK_LE(size, size_);
67 return ByteRange(data_.get(), size);
70 uint64_t hashIOBuf(const IOBuf* buf) {
71 uint64_t h = folly::hash::FNV_64_HASH_START;
72 for (auto& range : *buf) {
73 h = folly::hash::fnv64_buf(range.data(), range.size(), h);
78 class RandomDataHolder : public DataHolder {
80 explicit RandomDataHolder(size_t sizeLog2);
83 RandomDataHolder::RandomDataHolder(size_t sizeLog2)
84 : DataHolder(sizeLog2) {
85 constexpr size_t numThreadsLog2 = 3;
86 constexpr size_t numThreads = size_t(1) << numThreadsLog2;
88 uint32_t seed = randomNumberSeed();
90 std::vector<std::thread> threads;
91 threads.reserve(numThreads);
92 for (size_t t = 0; t < numThreads; ++t) {
94 [this, seed, t, numThreadsLog2, sizeLog2] () {
95 std::mt19937 rng(seed + t);
96 size_t countLog2 = sizeLog2 - numThreadsLog2;
97 size_t start = size_t(t) << countLog2;
98 for (size_t i = 0; i < countLog2; ++i) {
99 this->data_[start + i] = rng();
104 for (auto& t : threads) {
109 class ConstantDataHolder : public DataHolder {
111 explicit ConstantDataHolder(size_t sizeLog2);
114 ConstantDataHolder::ConstantDataHolder(size_t sizeLog2)
115 : DataHolder(sizeLog2) {
116 memset(data_.get(), 'a', size_);
119 constexpr size_t dataSizeLog2 = 27; // 128MiB
120 RandomDataHolder randomDataHolder(dataSizeLog2);
121 ConstantDataHolder constantDataHolder(dataSizeLog2);
123 TEST(CompressionTestNeedsUncompressedLength, Simple) {
124 EXPECT_FALSE(getCodec(CodecType::NO_COMPRESSION)->needsUncompressedLength());
125 EXPECT_TRUE(getCodec(CodecType::LZ4)->needsUncompressedLength());
126 EXPECT_FALSE(getCodec(CodecType::SNAPPY)->needsUncompressedLength());
127 EXPECT_FALSE(getCodec(CodecType::ZLIB)->needsUncompressedLength());
128 EXPECT_FALSE(getCodec(CodecType::LZ4_VARINT_SIZE)->needsUncompressedLength());
129 EXPECT_TRUE(getCodec(CodecType::LZMA2)->needsUncompressedLength());
130 EXPECT_FALSE(getCodec(CodecType::LZMA2_VARINT_SIZE)
131 ->needsUncompressedLength());
132 EXPECT_FALSE(getCodec(CodecType::ZSTD)->needsUncompressedLength());
133 EXPECT_FALSE(getCodec(CodecType::GZIP)->needsUncompressedLength());
136 class CompressionTest
137 : public testing::TestWithParam<std::tr1::tuple<int, int, CodecType>> {
139 void SetUp() override {
140 auto tup = GetParam();
141 uncompressedLength_ = uint64_t(1) << std::tr1::get<0>(tup);
142 chunks_ = std::tr1::get<1>(tup);
143 codec_ = getCodec(std::tr1::get<2>(tup));
146 void runSimpleIOBufTest(const DataHolder& dh);
148 void runSimpleStringTest(const DataHolder& dh);
151 std::unique_ptr<IOBuf> split(std::unique_ptr<IOBuf> data) const;
153 uint64_t uncompressedLength_;
155 std::unique_ptr<Codec> codec_;
158 void CompressionTest::runSimpleIOBufTest(const DataHolder& dh) {
159 const auto original = split(IOBuf::wrapBuffer(dh.data(uncompressedLength_)));
160 const auto compressed = split(codec_->compress(original.get()));
161 if (!codec_->needsUncompressedLength()) {
162 auto uncompressed = codec_->uncompress(compressed.get());
163 EXPECT_EQ(uncompressedLength_, uncompressed->computeChainDataLength());
164 EXPECT_EQ(dh.hash(uncompressedLength_), hashIOBuf(uncompressed.get()));
167 auto uncompressed = codec_->uncompress(compressed.get(),
168 uncompressedLength_);
169 EXPECT_EQ(uncompressedLength_, uncompressed->computeChainDataLength());
170 EXPECT_EQ(dh.hash(uncompressedLength_), hashIOBuf(uncompressed.get()));
174 void CompressionTest::runSimpleStringTest(const DataHolder& dh) {
175 const auto original = std::string(
176 reinterpret_cast<const char*>(dh.data(uncompressedLength_).data()),
177 uncompressedLength_);
178 const auto compressed = codec_->compress(original);
179 if (!codec_->needsUncompressedLength()) {
180 auto uncompressed = codec_->uncompress(compressed);
181 EXPECT_EQ(uncompressedLength_, uncompressed.length());
182 EXPECT_EQ(uncompressed, original);
185 auto uncompressed = codec_->uncompress(compressed, uncompressedLength_);
186 EXPECT_EQ(uncompressedLength_, uncompressed.length());
187 EXPECT_EQ(uncompressed, original);
191 // Uniformly split data into (potentially empty) chunks.
192 std::unique_ptr<IOBuf> CompressionTest::split(
193 std::unique_ptr<IOBuf> data) const {
194 if (data->isChained()) {
198 const size_t size = data->computeChainDataLength();
200 std::multiset<size_t> splits;
201 for (size_t i = 1; i < chunks_; ++i) {
202 splits.insert(Random::rand64(size));
205 folly::IOBufQueue result;
208 for (size_t split : splits) {
209 result.append(IOBuf::copyBuffer(data->data() + offset, split - offset));
212 result.append(IOBuf::copyBuffer(data->data() + offset, size - offset));
214 return result.move();
217 TEST_P(CompressionTest, RandomData) {
218 runSimpleIOBufTest(randomDataHolder);
221 TEST_P(CompressionTest, ConstantData) {
222 runSimpleIOBufTest(constantDataHolder);
225 TEST_P(CompressionTest, RandomDataString) {
226 runSimpleStringTest(randomDataHolder);
229 TEST_P(CompressionTest, ConstantDataString) {
230 runSimpleStringTest(constantDataHolder);
233 INSTANTIATE_TEST_CASE_P(
237 testing::Values(0, 1, 12, 22, 25, 27),
238 testing::Values(1, 2, 3, 8, 65),
240 CodecType::NO_COMPRESSION,
244 CodecType::LZ4_VARINT_SIZE,
246 CodecType::LZMA2_VARINT_SIZE,
250 class CompressionVarintTest
251 : public testing::TestWithParam<std::tr1::tuple<int, CodecType>> {
253 void SetUp() override {
254 auto tup = GetParam();
255 uncompressedLength_ = uint64_t(1) << std::tr1::get<0>(tup);
256 codec_ = getCodec(std::tr1::get<1>(tup));
259 void runSimpleTest(const DataHolder& dh);
261 uint64_t uncompressedLength_;
262 std::unique_ptr<Codec> codec_;
265 inline uint64_t oneBasedMsbPos(uint64_t number) {
267 for (; number > 0; ++pos, number >>= 1) {
272 void CompressionVarintTest::runSimpleTest(const DataHolder& dh) {
273 auto original = IOBuf::wrapBuffer(dh.data(uncompressedLength_));
274 auto compressed = codec_->compress(original.get());
278 std::max(uint64_t(9), oneBasedMsbPos(uncompressedLength_)) / 9UL);
279 auto tinyBuf = IOBuf::copyBuffer(compressed->data(),
280 std::min(compressed->length(), breakPoint));
281 compressed->trimStart(breakPoint);
282 tinyBuf->prependChain(std::move(compressed));
283 compressed = std::move(tinyBuf);
285 auto uncompressed = codec_->uncompress(compressed.get());
287 EXPECT_EQ(uncompressedLength_, uncompressed->computeChainDataLength());
288 EXPECT_EQ(dh.hash(uncompressedLength_), hashIOBuf(uncompressed.get()));
291 TEST_P(CompressionVarintTest, RandomData) {
292 runSimpleTest(randomDataHolder);
295 TEST_P(CompressionVarintTest, ConstantData) {
296 runSimpleTest(constantDataHolder);
299 INSTANTIATE_TEST_CASE_P(
300 CompressionVarintTest,
301 CompressionVarintTest,
303 testing::Values(0, 1, 12, 22, 25, 27),
305 CodecType::LZ4_VARINT_SIZE,
306 CodecType::LZMA2_VARINT_SIZE)));
308 class CompressionCorruptionTest : public testing::TestWithParam<CodecType> {
310 void SetUp() override { codec_ = getCodec(GetParam()); }
312 void runSimpleTest(const DataHolder& dh);
314 std::unique_ptr<Codec> codec_;
317 void CompressionCorruptionTest::runSimpleTest(const DataHolder& dh) {
318 constexpr uint64_t uncompressedLength = 42;
319 auto original = IOBuf::wrapBuffer(dh.data(uncompressedLength));
320 auto compressed = codec_->compress(original.get());
322 if (!codec_->needsUncompressedLength()) {
323 auto uncompressed = codec_->uncompress(compressed.get());
324 EXPECT_EQ(uncompressedLength, uncompressed->computeChainDataLength());
325 EXPECT_EQ(dh.hash(uncompressedLength), hashIOBuf(uncompressed.get()));
328 auto uncompressed = codec_->uncompress(compressed.get(),
330 EXPECT_EQ(uncompressedLength, uncompressed->computeChainDataLength());
331 EXPECT_EQ(dh.hash(uncompressedLength), hashIOBuf(uncompressed.get()));
334 EXPECT_THROW(codec_->uncompress(compressed.get(), uncompressedLength + 1),
337 // Corrupt the first character
338 ++(compressed->writableData()[0]);
340 if (!codec_->needsUncompressedLength()) {
341 EXPECT_THROW(codec_->uncompress(compressed.get()),
345 EXPECT_THROW(codec_->uncompress(compressed.get(), uncompressedLength),
349 TEST_P(CompressionCorruptionTest, RandomData) {
350 runSimpleTest(randomDataHolder);
353 TEST_P(CompressionCorruptionTest, ConstantData) {
354 runSimpleTest(constantDataHolder);
357 INSTANTIATE_TEST_CASE_P(
358 CompressionCorruptionTest,
359 CompressionCorruptionTest,
361 // NO_COMPRESSION can't detect corruption
362 // LZ4 can't detect corruption reliably (sigh)
368 int main(int argc, char *argv[]) {
369 testing::InitGoogleTest(&argc, argv);
370 gflags::ParseCommandLineFlags(&argc, &argv, true);
372 auto ret = RUN_ALL_TESTS();
374 folly::runBenchmarksOnFlag();