36a5f8ed7f
- Replace raw slice comparison with a call to user comparator.
Added test for custom comparators.
- Fix end of namespace comments.
- Fixed bug in picking inputs for a level-0 compaction.
When finding overlapping files, the covered range may expand
as files are added to the input set. We now correctly expand
the range when this happens instead of continuing to use the
old range. For example, suppose L0 contains files with the
following ranges:
F1: a .. d
F2: c .. g
F3: f .. j
and the initial compaction target is F3. We used to search
for range f..j which yielded {F2,F3}. However we now expand
the range as soon as another file is added. In this case,
when F2 is added, we expand the range to c..j and restart the
search. That picks up file F1 as well.
This change fixes a bug related to deleted keys showing up
incorrectly after a compaction as described in Issue 44.
(Sync with upstream @25072954)
507 lines
15 KiB
C++
507 lines
15 KiB
C++
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
|
|
// Use of this source code is governed by a BSD-style license that can be
|
|
// found in the LICENSE file. See the AUTHORS file for names of contributors.
|
|
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
#include <kcpolydb.h>
|
|
#include "util/histogram.h"
|
|
#include "util/random.h"
|
|
#include "util/testutil.h"
|
|
|
|
// Comma-separated list of operations to run in the specified order
|
|
// Actual benchmarks:
|
|
//
|
|
// fillseq -- write N values in sequential key order in async mode
|
|
// fillrandom -- write N values in random key order in async mode
|
|
// overwrite -- overwrite N values in random key order in async mode
|
|
// fillseqsync -- write N/100 values in sequential key order in sync mode
|
|
// fillrandsync -- write N/100 values in random key order in sync mode
|
|
// fillrand100K -- write N/1000 100K values in random order in async mode
|
|
// fillseq100K -- write N/1000 100K values in seq order in async mode
|
|
// readseq -- read N times sequentially
|
|
// readseq100K -- read N/1000 100K values in sequential order in async mode
|
|
// readrand100K -- read N/1000 100K values in sequential order in async mode
|
|
// readrandom -- read N times in random order
|
|
static const char* FLAGS_benchmarks =
|
|
"fillseq,"
|
|
"fillseqsync,"
|
|
"fillrandsync,"
|
|
"fillrandom,"
|
|
"overwrite,"
|
|
"readrandom,"
|
|
"readseq,"
|
|
"fillrand100K,"
|
|
"fillseq100K,"
|
|
"readseq100K,"
|
|
"readrand100K,"
|
|
;
|
|
|
|
// Number of key/values to place in database
|
|
static int FLAGS_num = 1000000;
|
|
|
|
// Number of read operations to do. If negative, do FLAGS_num reads.
|
|
static int FLAGS_reads = -1;
|
|
|
|
// Size of each value
|
|
static int FLAGS_value_size = 100;
|
|
|
|
// Arrange to generate values that shrink to this fraction of
|
|
// their original size after compression
|
|
static double FLAGS_compression_ratio = 0.5;
|
|
|
|
// Print histogram of operation timings
|
|
static bool FLAGS_histogram = false;
|
|
|
|
// Cache size. Default 4 MB
|
|
static int FLAGS_cache_size = 4194304;
|
|
|
|
// Page size. Default 1 KB
|
|
static int FLAGS_page_size = 1024;
|
|
|
|
// If true, do not destroy the existing database. If you set this
|
|
// flag and also specify a benchmark that wants a fresh database, that
|
|
// benchmark will fail.
|
|
static bool FLAGS_use_existing_db = false;
|
|
|
|
// Compression flag. If true, compression is on. If false, compression
|
|
// is off.
|
|
static bool FLAGS_compression = true;
|
|
|
|
inline
|
|
static void DBSynchronize(kyotocabinet::TreeDB* db_)
|
|
{
|
|
// Synchronize will flush writes to disk
|
|
if (!db_->synchronize()) {
|
|
fprintf(stderr, "synchronize error: %s\n", db_->error().name());
|
|
}
|
|
}
|
|
|
|
namespace leveldb {
|
|
|
|
// Helper for quickly generating random data.
|
|
namespace {
|
|
class RandomGenerator {
|
|
private:
|
|
std::string data_;
|
|
int pos_;
|
|
|
|
public:
|
|
RandomGenerator() {
|
|
// We use a limited amount of data over and over again and ensure
|
|
// that it is larger than the compression window (32KB), and also
|
|
// large enough to serve all typical value sizes we want to write.
|
|
Random rnd(301);
|
|
std::string piece;
|
|
while (data_.size() < 1048576) {
|
|
// Add a short fragment that is as compressible as specified
|
|
// by FLAGS_compression_ratio.
|
|
test::CompressibleString(&rnd, FLAGS_compression_ratio, 100, &piece);
|
|
data_.append(piece);
|
|
}
|
|
pos_ = 0;
|
|
}
|
|
|
|
Slice Generate(int len) {
|
|
if (pos_ + len > data_.size()) {
|
|
pos_ = 0;
|
|
assert(len < data_.size());
|
|
}
|
|
pos_ += len;
|
|
return Slice(data_.data() + pos_ - len, len);
|
|
}
|
|
};
|
|
|
|
static Slice TrimSpace(Slice s) {
|
|
int start = 0;
|
|
while (start < s.size() && isspace(s[start])) {
|
|
start++;
|
|
}
|
|
int limit = s.size();
|
|
while (limit > start && isspace(s[limit-1])) {
|
|
limit--;
|
|
}
|
|
return Slice(s.data() + start, limit - start);
|
|
}
|
|
|
|
} // namespace
|
|
|
|
class Benchmark {
|
|
private:
|
|
kyotocabinet::TreeDB* db_;
|
|
int db_num_;
|
|
int num_;
|
|
int reads_;
|
|
double start_;
|
|
double last_op_finish_;
|
|
int64_t bytes_;
|
|
std::string message_;
|
|
Histogram hist_;
|
|
RandomGenerator gen_;
|
|
Random rand_;
|
|
kyotocabinet::LZOCompressor<kyotocabinet::LZO::RAW> comp_;
|
|
|
|
// State kept for progress messages
|
|
int done_;
|
|
int next_report_; // When to report next
|
|
|
|
void PrintHeader() {
|
|
const int kKeySize = 16;
|
|
PrintEnvironment();
|
|
fprintf(stdout, "Keys: %d bytes each\n", kKeySize);
|
|
fprintf(stdout, "Values: %d bytes each (%d bytes after compression)\n",
|
|
FLAGS_value_size,
|
|
static_cast<int>(FLAGS_value_size * FLAGS_compression_ratio + 0.5));
|
|
fprintf(stdout, "Entries: %d\n", num_);
|
|
fprintf(stdout, "RawSize: %.1f MB (estimated)\n",
|
|
((static_cast<int64_t>(kKeySize + FLAGS_value_size) * num_)
|
|
/ 1048576.0));
|
|
fprintf(stdout, "FileSize: %.1f MB (estimated)\n",
|
|
(((kKeySize + FLAGS_value_size * FLAGS_compression_ratio) * num_)
|
|
/ 1048576.0));
|
|
PrintWarnings();
|
|
fprintf(stdout, "------------------------------------------------\n");
|
|
}
|
|
|
|
void PrintWarnings() {
|
|
#if defined(__GNUC__) && !defined(__OPTIMIZE__)
|
|
fprintf(stdout,
|
|
"WARNING: Optimization is disabled: benchmarks unnecessarily slow\n"
|
|
);
|
|
#endif
|
|
#ifndef NDEBUG
|
|
fprintf(stdout,
|
|
"WARNING: Assertions are enabled; benchmarks unnecessarily slow\n");
|
|
#endif
|
|
}
|
|
|
|
void PrintEnvironment() {
|
|
fprintf(stderr, "Kyoto Cabinet: version %s, lib ver %d, lib rev %d\n",
|
|
kyotocabinet::VERSION, kyotocabinet::LIBVER, kyotocabinet::LIBREV);
|
|
|
|
#if defined(__linux)
|
|
time_t now = time(NULL);
|
|
fprintf(stderr, "Date: %s", ctime(&now)); // ctime() adds newline
|
|
|
|
FILE* cpuinfo = fopen("/proc/cpuinfo", "r");
|
|
if (cpuinfo != NULL) {
|
|
char line[1000];
|
|
int num_cpus = 0;
|
|
std::string cpu_type;
|
|
std::string cache_size;
|
|
while (fgets(line, sizeof(line), cpuinfo) != NULL) {
|
|
const char* sep = strchr(line, ':');
|
|
if (sep == NULL) {
|
|
continue;
|
|
}
|
|
Slice key = TrimSpace(Slice(line, sep - 1 - line));
|
|
Slice val = TrimSpace(Slice(sep + 1));
|
|
if (key == "model name") {
|
|
++num_cpus;
|
|
cpu_type = val.ToString();
|
|
} else if (key == "cache size") {
|
|
cache_size = val.ToString();
|
|
}
|
|
}
|
|
fclose(cpuinfo);
|
|
fprintf(stderr, "CPU: %d * %s\n", num_cpus, cpu_type.c_str());
|
|
fprintf(stderr, "CPUCache: %s\n", cache_size.c_str());
|
|
}
|
|
#endif
|
|
}
|
|
|
|
void Start() {
|
|
start_ = Env::Default()->NowMicros() * 1e-6;
|
|
bytes_ = 0;
|
|
message_.clear();
|
|
last_op_finish_ = start_;
|
|
hist_.Clear();
|
|
done_ = 0;
|
|
next_report_ = 100;
|
|
}
|
|
|
|
void FinishedSingleOp() {
|
|
if (FLAGS_histogram) {
|
|
double now = Env::Default()->NowMicros() * 1e-6;
|
|
double micros = (now - last_op_finish_) * 1e6;
|
|
hist_.Add(micros);
|
|
if (micros > 20000) {
|
|
fprintf(stderr, "long op: %.1f micros%30s\r", micros, "");
|
|
fflush(stderr);
|
|
}
|
|
last_op_finish_ = now;
|
|
}
|
|
|
|
done_++;
|
|
if (done_ >= next_report_) {
|
|
if (next_report_ < 1000) next_report_ += 100;
|
|
else if (next_report_ < 5000) next_report_ += 500;
|
|
else if (next_report_ < 10000) next_report_ += 1000;
|
|
else if (next_report_ < 50000) next_report_ += 5000;
|
|
else if (next_report_ < 100000) next_report_ += 10000;
|
|
else if (next_report_ < 500000) next_report_ += 50000;
|
|
else next_report_ += 100000;
|
|
fprintf(stderr, "... finished %d ops%30s\r", done_, "");
|
|
fflush(stderr);
|
|
}
|
|
}
|
|
|
|
void Stop(const Slice& name) {
|
|
double finish = Env::Default()->NowMicros() * 1e-6;
|
|
|
|
// Pretend at least one op was done in case we are running a benchmark
|
|
// that does not call FinishedSingleOp().
|
|
if (done_ < 1) done_ = 1;
|
|
|
|
if (bytes_ > 0) {
|
|
char rate[100];
|
|
snprintf(rate, sizeof(rate), "%6.1f MB/s",
|
|
(bytes_ / 1048576.0) / (finish - start_));
|
|
if (!message_.empty()) {
|
|
message_ = std::string(rate) + " " + message_;
|
|
} else {
|
|
message_ = rate;
|
|
}
|
|
}
|
|
|
|
fprintf(stdout, "%-12s : %11.3f micros/op;%s%s\n",
|
|
name.ToString().c_str(),
|
|
(finish - start_) * 1e6 / done_,
|
|
(message_.empty() ? "" : " "),
|
|
message_.c_str());
|
|
if (FLAGS_histogram) {
|
|
fprintf(stdout, "Microseconds per op:\n%s\n", hist_.ToString().c_str());
|
|
}
|
|
fflush(stdout);
|
|
}
|
|
|
|
public:
|
|
enum Order {
|
|
SEQUENTIAL,
|
|
RANDOM
|
|
};
|
|
enum DBState {
|
|
FRESH,
|
|
EXISTING
|
|
};
|
|
|
|
Benchmark()
|
|
: db_(NULL),
|
|
num_(FLAGS_num),
|
|
reads_(FLAGS_reads < 0 ? FLAGS_num : FLAGS_reads),
|
|
bytes_(0),
|
|
rand_(301) {
|
|
std::vector<std::string> files;
|
|
Env::Default()->GetChildren("/tmp", &files);
|
|
if (!FLAGS_use_existing_db) {
|
|
for (int i = 0; i < files.size(); i++) {
|
|
if (Slice(files[i]).starts_with("dbbench_polyDB")) {
|
|
Env::Default()->DeleteFile("/tmp/" + files[i]);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
~Benchmark() {
|
|
if (!db_->close()) {
|
|
fprintf(stderr, "close error: %s\n", db_->error().name());
|
|
}
|
|
}
|
|
|
|
void Run() {
|
|
PrintHeader();
|
|
Open(false);
|
|
|
|
const char* benchmarks = FLAGS_benchmarks;
|
|
while (benchmarks != NULL) {
|
|
const char* sep = strchr(benchmarks, ',');
|
|
Slice name;
|
|
if (sep == NULL) {
|
|
name = benchmarks;
|
|
benchmarks = NULL;
|
|
} else {
|
|
name = Slice(benchmarks, sep - benchmarks);
|
|
benchmarks = sep + 1;
|
|
}
|
|
|
|
Start();
|
|
|
|
bool known = true;
|
|
bool write_sync = false;
|
|
if (name == Slice("fillseq")) {
|
|
Write(write_sync, SEQUENTIAL, FRESH, num_, FLAGS_value_size, 1);
|
|
|
|
} else if (name == Slice("fillrandom")) {
|
|
Write(write_sync, RANDOM, FRESH, num_, FLAGS_value_size, 1);
|
|
DBSynchronize(db_);
|
|
} else if (name == Slice("overwrite")) {
|
|
Write(write_sync, RANDOM, EXISTING, num_, FLAGS_value_size, 1);
|
|
DBSynchronize(db_);
|
|
} else if (name == Slice("fillrandsync")) {
|
|
write_sync = true;
|
|
Write(write_sync, RANDOM, FRESH, num_ / 100, FLAGS_value_size, 1);
|
|
DBSynchronize(db_);
|
|
} else if (name == Slice("fillseqsync")) {
|
|
write_sync = true;
|
|
Write(write_sync, SEQUENTIAL, FRESH, num_ / 100, FLAGS_value_size, 1);
|
|
DBSynchronize(db_);
|
|
} else if (name == Slice("fillrand100K")) {
|
|
Write(write_sync, RANDOM, FRESH, num_ / 1000, 100 * 1000, 1);
|
|
DBSynchronize(db_);
|
|
} else if (name == Slice("fillseq100K")) {
|
|
Write(write_sync, SEQUENTIAL, FRESH, num_ / 1000, 100 * 1000, 1);
|
|
DBSynchronize(db_);
|
|
} else if (name == Slice("readseq")) {
|
|
ReadSequential();
|
|
} else if (name == Slice("readrandom")) {
|
|
ReadRandom();
|
|
} else if (name == Slice("readrand100K")) {
|
|
int n = reads_;
|
|
reads_ /= 1000;
|
|
ReadRandom();
|
|
reads_ = n;
|
|
} else if (name == Slice("readseq100K")) {
|
|
int n = reads_;
|
|
reads_ /= 1000;
|
|
ReadSequential();
|
|
reads_ = n;
|
|
} else {
|
|
known = false;
|
|
if (name != Slice()) { // No error message for empty name
|
|
fprintf(stderr, "unknown benchmark '%s'\n", name.ToString().c_str());
|
|
}
|
|
}
|
|
if (known) {
|
|
Stop(name);
|
|
}
|
|
}
|
|
}
|
|
|
|
private:
|
|
void Open(bool sync) {
|
|
assert(db_ == NULL);
|
|
|
|
// Initialize db_
|
|
db_ = new kyotocabinet::TreeDB();
|
|
char file_name[100];
|
|
db_num_++;
|
|
snprintf(file_name, sizeof(file_name), "/tmp/dbbench_polyDB-%d.kct",
|
|
db_num_);
|
|
|
|
// Create tuning options and open the database
|
|
int open_options = kyotocabinet::PolyDB::OWRITER |
|
|
kyotocabinet::PolyDB::OCREATE;
|
|
int tune_options = kyotocabinet::TreeDB::TSMALL |
|
|
kyotocabinet::TreeDB::TLINEAR;
|
|
if (FLAGS_compression) {
|
|
tune_options |= kyotocabinet::TreeDB::TCOMPRESS;
|
|
db_->tune_compressor(&comp_);
|
|
}
|
|
db_->tune_options(tune_options);
|
|
db_->tune_page_cache(FLAGS_cache_size);
|
|
db_->tune_page(FLAGS_page_size);
|
|
db_->tune_map(256LL<<20);
|
|
if (sync) {
|
|
open_options |= kyotocabinet::PolyDB::OAUTOSYNC;
|
|
}
|
|
if (!db_->open(file_name, open_options)) {
|
|
fprintf(stderr, "open error: %s\n", db_->error().name());
|
|
}
|
|
}
|
|
|
|
void Write(bool sync, Order order, DBState state,
|
|
int num_entries, int value_size, int entries_per_batch) {
|
|
// Create new database if state == FRESH
|
|
if (state == FRESH) {
|
|
if (FLAGS_use_existing_db) {
|
|
message_ = "skipping (--use_existing_db is true)";
|
|
return;
|
|
}
|
|
delete db_;
|
|
db_ = NULL;
|
|
Open(sync);
|
|
Start(); // Do not count time taken to destroy/open
|
|
}
|
|
|
|
if (num_entries != num_) {
|
|
char msg[100];
|
|
snprintf(msg, sizeof(msg), "(%d ops)", num_entries);
|
|
message_ = msg;
|
|
}
|
|
|
|
// Write to database
|
|
for (int i = 0; i < num_entries; i++)
|
|
{
|
|
const int k = (order == SEQUENTIAL) ? i : (rand_.Next() % num_entries);
|
|
char key[100];
|
|
snprintf(key, sizeof(key), "%016d", k);
|
|
bytes_ += value_size + strlen(key);
|
|
std::string cpp_key = key;
|
|
if (!db_->set(cpp_key, gen_.Generate(value_size).ToString())) {
|
|
fprintf(stderr, "set error: %s\n", db_->error().name());
|
|
}
|
|
FinishedSingleOp();
|
|
}
|
|
}
|
|
|
|
void ReadSequential() {
|
|
kyotocabinet::DB::Cursor* cur = db_->cursor();
|
|
cur->jump();
|
|
std::string ckey, cvalue;
|
|
while (cur->get(&ckey, &cvalue, true)) {
|
|
bytes_ += ckey.size() + cvalue.size();
|
|
FinishedSingleOp();
|
|
}
|
|
delete cur;
|
|
}
|
|
|
|
void ReadRandom() {
|
|
std::string value;
|
|
for (int i = 0; i < reads_; i++) {
|
|
char key[100];
|
|
const int k = rand_.Next() % reads_;
|
|
snprintf(key, sizeof(key), "%016d", k);
|
|
db_->get(key, &value);
|
|
FinishedSingleOp();
|
|
}
|
|
}
|
|
};
|
|
|
|
} // namespace leveldb
|
|
|
|
int main(int argc, char** argv) {
|
|
for (int i = 1; i < argc; i++) {
|
|
double d;
|
|
int n;
|
|
char junk;
|
|
if (leveldb::Slice(argv[i]).starts_with("--benchmarks=")) {
|
|
FLAGS_benchmarks = argv[i] + strlen("--benchmarks=");
|
|
} else if (sscanf(argv[i], "--compression_ratio=%lf%c", &d, &junk) == 1) {
|
|
FLAGS_compression_ratio = d;
|
|
} else if (sscanf(argv[i], "--histogram=%d%c", &n, &junk) == 1 &&
|
|
(n == 0 || n == 1)) {
|
|
FLAGS_histogram = n;
|
|
} else if (sscanf(argv[i], "--num=%d%c", &n, &junk) == 1) {
|
|
FLAGS_num = n;
|
|
} else if (sscanf(argv[i], "--reads=%d%c", &n, &junk) == 1) {
|
|
FLAGS_reads = n;
|
|
} else if (sscanf(argv[i], "--value_size=%d%c", &n, &junk) == 1) {
|
|
FLAGS_value_size = n;
|
|
} else if (sscanf(argv[i], "--cache_size=%d%c", &n, &junk) == 1) {
|
|
FLAGS_cache_size = n;
|
|
} else if (sscanf(argv[i], "--page_size=%d%c", &n, &junk) == 1) {
|
|
FLAGS_page_size = n;
|
|
} else if (sscanf(argv[i], "--compression=%d%c", &n, &junk) == 1 &&
|
|
(n == 0 || n == 1)) {
|
|
FLAGS_compression = (n == 1) ? true : false;
|
|
} else {
|
|
fprintf(stderr, "Invalid flag '%s'\n", argv[i]);
|
|
exit(1);
|
|
}
|
|
}
|
|
|
|
leveldb::Benchmark benchmark;
|
|
benchmark.Run();
|
|
return 0;
|
|
}
|