Merge branch 'master' into performance

Conflicts:
	db/db_impl.cc
	util/options.cc
main
Dhruba Borthakur 12 years ago
commit 81f735d97c
  1. 7
      db/c.cc
  2. 1
      db/c_test.c
  3. 28
      db/db_bench.cc
  4. 20
      db/db_impl.cc
  5. 72
      db/db_test.cc
  6. 2
      db/table_cache.cc
  7. 24
      include/leveldb/options.h
  8. 18
      port/port_posix.h
  9. 9
      table/table_builder.cc
  10. 10
      table/table_test.cc
  11. 30
      util/options.cc

@ -502,6 +502,13 @@ void leveldb_options_set_compression(leveldb_options_t* opt, int t) {
opt->rep.compression = static_cast<CompressionType>(t);
}
void leveldb_options_set_compression_options(
leveldb_options_t* opt, int w_bits, int level, int strategy) {
opt->rep.compression_opts.window_bits = w_bits;
opt->rep.compression_opts.level = level;
opt->rep.compression_opts.strategy = strategy;
}
void leveldb_options_set_disable_data_sync(
leveldb_options_t* opt, bool disable_data_sync) {
opt->rep.disableDataSync = disable_data_sync;

@ -187,6 +187,7 @@ int main(int argc, char** argv) {
leveldb_options_set_block_size(options, 1024);
leveldb_options_set_block_restart_interval(options, 8);
leveldb_options_set_compression(options, leveldb_no_compression);
leveldb_options_set_compression_options(options, -14, -1, 0);
roptions = leveldb_readoptions_create();
leveldb_readoptions_set_verify_checksums(roptions, 1);

@ -193,6 +193,8 @@ static enum leveldb::CompressionType FLAGS_compression_type =
// other levels are compressed
static int FLAGS_min_level_to_compress = -1;
static int FLAGS_table_cache_numshardbits = 4;
// posix or hdfs environment
static leveldb::Env* FLAGS_env = leveldb::Env::Default();
@ -518,15 +520,18 @@ class Benchmark {
switch (FLAGS_compression_type) {
case kSnappyCompression:
result = port::Snappy_Compress(text, strlen(text), &compressed);
result = port::Snappy_Compress(Options().compression_opts, text,
strlen(text), &compressed);
name = "Snappy";
break;
case kZlibCompression:
result = port::Zlib_Compress(text, strlen(text), &compressed);
result = port::Zlib_Compress(Options().compression_opts, text,
strlen(text), &compressed);
name = "Zlib";
break;
case kBZip2Compression:
result = port::BZip2_Compress(text, strlen(text), &compressed);
result = port::BZip2_Compress(Options().compression_opts, text,
strlen(text), &compressed);
name = "BZip2";
break;
}
@ -864,7 +869,8 @@ class Benchmark {
bool ok = true;
std::string compressed;
while (ok && bytes < 1024 * 1048576) { // Compress 1G
ok = port::Snappy_Compress(input.data(), input.size(), &compressed);
ok = port::Snappy_Compress(Options().compression_opts, input.data(),
input.size(), &compressed);
produced += compressed.size();
bytes += input.size();
thread->stats.FinishedSingleOp(NULL);
@ -885,7 +891,8 @@ class Benchmark {
RandomGenerator gen;
Slice input = gen.Generate(Options().block_size);
std::string compressed;
bool ok = port::Snappy_Compress(input.data(), input.size(), &compressed);
bool ok = port::Snappy_Compress(Options().compression_opts, input.data(),
input.size(), &compressed);
int64_t bytes = 0;
char* uncompressed = new char[input.size()];
while (ok && bytes < 1024 * 1048576) { // Compress 1G
@ -908,6 +915,9 @@ class Benchmark {
Options options;
options.create_if_missing = !FLAGS_use_existing_db;
options.block_cache = cache_;
if (cache_ == NULL) {
options.no_block_cache = true;
}
options.write_buffer_size = FLAGS_write_buffer_size;
options.max_write_buffer_number = FLAGS_max_write_buffer_number;
options.max_background_compactions = FLAGS_max_background_compactions;
@ -945,6 +955,7 @@ class Benchmark {
options.delete_obsolete_files_period_micros =
FLAGS_delete_obsolete_files_period_micros;
options.rate_limit = FLAGS_rate_limit;
options.table_cache_numshardbits = FLAGS_table_cache_numshardbits;
Status s = DB::Open(options, FLAGS_db, &db_);
if (!s.ok()) {
fprintf(stderr, "open error: %s\n", s.ToString().c_str());
@ -1276,6 +1287,13 @@ int main(int argc, char** argv) {
fprintf(stderr, "The cache cannot be sharded into 2**%d pieces\n", n);
exit(1);
}
} else if (sscanf(argv[i], "--table_cache_numshardbits=%d%c",
&n, &junk) == 1) {
if (n <= 0 || n > 20) {
fprintf(stderr, "The cache cannot be sharded into 2**%d pieces\n", n);
exit(1);
}
FLAGS_table_cache_numshardbits = n;
} else if (sscanf(argv[i], "--bloom_bits=%d%c", &n, &junk) == 1) {
FLAGS_bloom_bits = n;
} else if (sscanf(argv[i], "--open_files=%d%c", &n, &junk) == 1) {

@ -155,7 +155,7 @@ Options SanitizeOptions(const std::string& dbname,
result.info_log = NULL;
}
}
if (result.block_cache == NULL) {
if (result.block_cache == NULL && !result.no_block_cache) {
result.block_cache = NewLRUCache(8 << 20);
}
if (src.compression_per_level != NULL) {
@ -1707,9 +1707,14 @@ Status DBImpl::MakeRoomForWrite(bool force) {
// this delay hands over some CPU to the compaction thread in
// case it is sharing the same core as the writer.
mutex_.Unlock();
uint64_t t1 = env_->NowMicros();
env_->SleepForMicroseconds(1000);
stall_level0_slowdown_ += 1000;
uint64_t delayed = env_->NowMicros() - t1;
stall_level0_slowdown_ += delayed;
allow_delay = false; // Do not delay a single write more than once
//Log(options_.info_log,
// "delaying write %llu usecs for level0_slowdown_writes_trigger\n",
// delayed);
mutex_.Lock();
delayed_writes_++;
} else if (!force &&
@ -1741,11 +1746,14 @@ Status DBImpl::MakeRoomForWrite(bool force) {
(score = versions_->MaxCompactionScore()) > options_.rate_limit) {
// Delay a write when the compaction score for any level is too large.
mutex_.Unlock();
uint64_t t1 = env_->NowMicros();
env_->SleepForMicroseconds(1000);
stall_leveln_slowdown_ += 1000;
uint64_t delayed = env_->NowMicros() - t1;
stall_leveln_slowdown_ += delayed;
allow_delay = false; // Do not delay a single write more than once
Log(options_.info_log,
"delaying write for rate limits with max score %.2f\n", score);
"delaying write %llu usecs for rate limits with max score %.2f\n",
delayed, score);
mutex_.Lock();
} else {
// Attempt to switch to a new memtable and trigger compaction of old
@ -1930,6 +1938,10 @@ Status DB::Open(const Options& options, const std::string& dbname,
DB** dbptr) {
*dbptr = NULL;
if (options.block_cache != NULL && options.no_block_cache) {
return Status::InvalidArgument(
"no_block_cache is true while block_cache is not NULL");
}
DBImpl* impl = new DBImpl(options, dbname);
impl->mutex_.Lock();
VersionEdit edit(impl->NumberLevels());

@ -20,22 +20,22 @@
namespace leveldb {
static bool SnappyCompressionSupported() {
static bool SnappyCompressionSupported(const CompressionOptions& options) {
std::string out;
Slice in = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
return port::Snappy_Compress(in.data(), in.size(), &out);
return port::Snappy_Compress(options, in.data(), in.size(), &out);
}
static bool ZlibCompressionSupported() {
static bool ZlibCompressionSupported(const CompressionOptions& options) {
std::string out;
Slice in = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
return port::Zlib_Compress(in.data(), in.size(), &out);
return port::Zlib_Compress(options, in.data(), in.size(), &out);
}
static bool BZip2CompressionSupported() {
static bool BZip2CompressionSupported(const CompressionOptions& options) {
std::string out;
Slice in = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
return port::BZip2_Compress(in.data(), in.size(), &out);
return port::BZip2_Compress(options, in.data(), in.size(), &out);
}
static std::string RandomString(Random* rnd, int len) {
@ -1105,22 +1105,24 @@ void MinLevelHelper(DBTest* self, Options& options) {
ASSERT_EQ(self->NumTableFilesAtLevel(1), 1);
}
TEST(DBTest, MinLevelToCompress) {
Options options = CurrentOptions();
void MinLevelToCompress(CompressionType& type, Options& options, int wbits,
int lev, int strategy) {
fprintf(stderr, "Test with compression options : window_bits = %d, level = %d, strategy = %d}\n", wbits, lev, strategy);
options.write_buffer_size = 100<<10; //100KB
options.num_levels = 3;
options.max_mem_compaction_level = 0;
options.level0_file_num_compaction_trigger = 3;
options.create_if_missing = true;
CompressionType type;
if (SnappyCompressionSupported()) {
if (SnappyCompressionSupported(CompressionOptions(wbits, lev, strategy))) {
type = kSnappyCompression;
fprintf(stderr, "using snappy\n");
} else if (ZlibCompressionSupported()) {
} else if (ZlibCompressionSupported(
CompressionOptions(wbits, lev, strategy))) {
type = kZlibCompression;
fprintf(stderr, "using zlib\n");
} else if (BZip2CompressionSupported()) {
} else if (BZip2CompressionSupported(
CompressionOptions(wbits, lev, strategy))) {
type = kBZip2Compression;
fprintf(stderr, "using bzip2\n");
} else {
@ -1136,6 +1138,29 @@ TEST(DBTest, MinLevelToCompress) {
for (int i = 1; i < options.num_levels; i++) {
options.compression_per_level[i] = type;
}
}
TEST(DBTest, MinLevelToCompress1) {
Options options = CurrentOptions();
CompressionType type;
MinLevelToCompress(type, options, -14, -1, 0);
Reopen(&options);
MinLevelHelper(this, options);
// do not compress L0 and L1
for (int i = 0; i < 2; i++) {
options.compression_per_level[i] = kNoCompression;
}
for (int i = 2; i < options.num_levels; i++) {
options.compression_per_level[i] = type;
}
DestroyAndReopen(&options);
MinLevelHelper(this, options);
}
TEST(DBTest, MinLevelToCompress2) {
Options options = CurrentOptions();
CompressionType type;
MinLevelToCompress(type, options, 15, -1, 0);
Reopen(&options);
MinLevelHelper(this, options);
@ -1682,6 +1707,29 @@ TEST(DBTest, DBOpen_Options) {
db = NULL;
}
TEST(DBTest, DBOpen_Change_NumLevels) {
std::string dbname = test::TmpDir() + "/db_change_num_levels";
DestroyDB(dbname, Options());
Options opts;
Status s;
DB* db = NULL;
opts.create_if_missing = true;
s = DB::Open(opts, dbname, &db);
ASSERT_OK(s);
ASSERT_TRUE(db != NULL);
db->Put(WriteOptions(), "a", "123");
db->Put(WriteOptions(), "b", "234");
db->CompactRange(NULL, NULL);
delete db;
db = NULL;
opts.create_if_missing = false;
opts.num_levels = 2;
s = DB::Open(opts, dbname, &db);
ASSERT_TRUE(strstr(s.ToString().c_str(), "Corruption") != NULL);
ASSERT_TRUE(db == NULL);
}
// Check that number of files does not grow when we are out of space
TEST(DBTest, NoSpace) {
Options options = CurrentOptions();

@ -39,7 +39,7 @@ TableCache::TableCache(const std::string& dbname,
: env_(options->env),
dbname_(dbname),
options_(options),
cache_(NewLRUCache(entries)) {
cache_(NewLRUCache(entries, options->table_cache_numshardbits)) {
dbstatistics = (DBStatistics*)options->statistics;
}

@ -32,6 +32,19 @@ enum CompressionType {
kBZip2Compression = 0x3
};
// Compression options for different compression algorithms like Zlib
struct CompressionOptions {
int window_bits;
int level;
int strategy;
CompressionOptions():window_bits(-14),
level(-1),
strategy(0){}
CompressionOptions(int wbits, int lev, int strategy):window_bits(wbits),
level(lev),
strategy(strategy){}
};
// Options to control the behavior of a database (passed to DB::Open)
struct Options {
// -------------------
@ -155,6 +168,9 @@ struct Options {
// java/C api hard to construct.
CompressionType* compression_per_level;
//different options for compression algorithms
CompressionOptions compression_opts;
// If non-NULL, use the specified filter policy to reduce disk reads.
// Many applications will benefit from passing the result of
// NewBloomFilterPolicy() here.
@ -282,6 +298,14 @@ struct Options {
// exceeds rate_limit. This is ignored when <= 1.0.
double rate_limit;
// Disable block cache. If this is set to false,
// then no block cache should be used, and the block_cache should
// point to a NULL object.
bool no_block_cache;
// Number of shards used for table cache.
int table_cache_numshardbits;
// Create an Options object with default values for all fields.
Options();

@ -44,6 +44,7 @@
#include <stdint.h>
#include <string>
#include <string.h>
#include "leveldb/options.h"
#include "port/atomic_pointer.h"
#ifndef PLATFORM_IS_LITTLE_ENDIAN
@ -131,8 +132,8 @@ typedef pthread_once_t OnceType;
#define LEVELDB_ONCE_INIT PTHREAD_ONCE_INIT
extern void InitOnce(OnceType* once, void (*initializer)());
inline bool Snappy_Compress(const char* input, size_t length,
::std::string* output) {
inline bool Snappy_Compress(const CompressionOptions& opts, const char* input,
size_t length, ::std::string* output) {
#ifdef SNAPPY
output->resize(snappy::MaxCompressedLength(length));
size_t outlen;
@ -162,9 +163,8 @@ inline bool Snappy_Uncompress(const char* input, size_t length,
#endif
}
inline bool Zlib_Compress(const char* input, size_t length,
::std::string* output, int windowBits = -14, int level = -1,
int strategy = 0) {
inline bool Zlib_Compress(const CompressionOptions& opts, const char* input,
size_t length, ::std::string* output) {
#ifdef ZLIB
// The memLevel parameter specifies how much memory should be allocated for
// the internal compression state.
@ -174,8 +174,8 @@ inline bool Zlib_Compress(const char* input, size_t length,
static const int memLevel = 8;
z_stream _stream;
memset(&_stream, 0, sizeof(z_stream));
int st = deflateInit2(&_stream, level, Z_DEFLATED, windowBits,
memLevel, strategy);
int st = deflateInit2(&_stream, opts.level, Z_DEFLATED, opts.window_bits,
memLevel, opts.strategy);
if (st != Z_OK) {
return false;
}
@ -284,8 +284,8 @@ inline char* Zlib_Uncompress(const char* input_data, size_t input_length,
return NULL;
}
inline bool BZip2_Compress(const char* input, size_t length,
::std::string* output) {
inline bool BZip2_Compress(const CompressionOptions& opts, const char* input,
size_t length, ::std::string* output) {
#ifdef BZIP2
bz_stream _stream;
memset(&_stream, 0, sizeof(bz_stream));

@ -175,7 +175,8 @@ void TableBuilder::WriteBlock(BlockBuilder* block, BlockHandle* handle) {
case kSnappyCompression: {
std::string* compressed = &r->compressed_output;
if (port::Snappy_Compress(raw.data(), raw.size(), compressed) &&
if (port::Snappy_Compress(r->options.compression_opts, raw.data(),
raw.size(), compressed) &&
GoodCompressionRatio(compressed->size(), raw.size())) {
block_contents = *compressed;
} else {
@ -187,7 +188,8 @@ void TableBuilder::WriteBlock(BlockBuilder* block, BlockHandle* handle) {
break;
}
case kZlibCompression:
if (port::Zlib_Compress(raw.data(), raw.size(), compressed) &&
if (port::Zlib_Compress(r->options.compression_opts, raw.data(),
raw.size(), compressed) &&
GoodCompressionRatio(compressed->size(), raw.size())) {
block_contents = *compressed;
} else {
@ -198,7 +200,8 @@ void TableBuilder::WriteBlock(BlockBuilder* block, BlockHandle* handle) {
}
break;
case kBZip2Compression:
if (port::BZip2_Compress(raw.data(), raw.size(), compressed) &&
if (port::BZip2_Compress(r->options.compression_opts, raw.data(),
raw.size(), compressed) &&
GoodCompressionRatio(compressed->size(), raw.size())) {
block_contents = *compressed;
} else {

@ -247,6 +247,7 @@ class TableConstructor: public Constructor {
source_ = new StringSource(sink.contents());
Options table_options;
table_options.comparator = options.comparator;
table_options.compression_opts = options.compression_opts;
return Table::Open(table_options, source_, sink.contents().size(), &table_);
}
@ -399,19 +400,22 @@ class DBConstructor: public Constructor {
static bool SnappyCompressionSupported() {
std::string out;
Slice in = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
return port::Snappy_Compress(in.data(), in.size(), &out);
return port::Snappy_Compress(Options().compression_opts, in.data(), in.size(),
&out);
}
static bool ZlibCompressionSupported() {
std::string out;
Slice in = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
return port::Zlib_Compress(in.data(), in.size(), &out);
return port::Zlib_Compress(Options().compression_opts, in.data(), in.size(),
&out);
}
static bool BZip2CompressionSupported() {
std::string out;
Slice in = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
return port::BZip2_Compress(in.data(), in.size(), &out);
return port::BZip2_Compress(Options().compression_opts, in.data(), in.size(),
&out);
}
enum TestType {

@ -44,10 +44,12 @@ Options::Options()
db_stats_log_interval(1800),
db_log_dir(""),
disable_seek_compaction(false),
delete_obsolete_files_period_micros(0),
max_background_compactions(1),
no_block_cache(false),
table_cache_numshardbits(4),
max_log_file_size(0),
rate_limit(0.0) {
delete_obsolete_files_period_micros(0),
rate_limit(0.0),
max_background_compactions(1) {
}
void
@ -64,15 +66,12 @@ Options::Dump(
Log(log," Options.max_write_buffer_number: %zd", max_write_buffer_number);
Log(log," Options.max_open_files: %d", max_open_files);
Log(log," Options.block_cache: %p", block_cache);
Log(log," Options.block_cache_size: %zd", block_cache->GetCapacity());
if (block_cache) {
Log(log," Options.block_cache_size: %zd",
block_cache->GetCapacity());
}
Log(log," Options.block_size: %zd", block_size);
Log(log," Options.block_restart_interval: %d", block_restart_interval);
Log(log," Options.compression: %d", compression);
Log(log," Options.filter_policy: %s",
filter_policy == NULL ? "NULL" : filter_policy->Name());
Log(log," Options.num_levels: %d", num_levels);
Log(log," Options.disableDataSync: %d", disableDataSync);
Log(log," Options.use_fsync: %d", use_fsync);
if (compression_per_level != NULL) {
for (unsigned int i = 0; i < num_levels; i++){
Log(log," Options.compression[%d]: %d",
@ -89,6 +88,12 @@ Options::Dump(
Log(log," Options.max_log_file_size: %d", max_log_file_size);
Log(log," Options.db_stats_log_interval: %d",
db_stats_log_interval);
Log(log," Options.compression_opts.window_bits: %d",
compression_opts.window_bits);
Log(log," Options.compression_opts.level: %d",
compression_opts.level);
Log(log," Options.compression_opts.strategy: %d",
compression_opts.strategy);
Log(log," Options.level0_file_num_compaction_trigger: %d",
level0_file_num_compaction_trigger);
Log(log," Options.level0_slowdown_writes_trigger: %d",
@ -113,6 +118,10 @@ Options::Dump(
db_log_dir.c_str());
Log(log," Options.disable_seek_compaction: %d",
disable_seek_compaction);
Log(log," Options.no_block_cache: %d",
no_block_cache);
Log(log," Options.table_cache_numshardbits: %d",
table_cache_numshardbits);
Log(log," Options.delete_obsolete_files_period_micros: %ld",
delete_obsolete_files_period_micros);
Log(log," Options.max_background_compactions: %d",
@ -121,5 +130,4 @@ Options::Dump(
rate_limit);
} // Options::Dump
} // namespace leveldb

Loading…
Cancel
Save