diff options
Diffstat (limited to 'src/leveldb/benchmarks')
-rw-r--r-- | src/leveldb/benchmarks/db_bench.cc | 981 | ||||
-rw-r--r-- | src/leveldb/benchmarks/db_bench_sqlite3.cc | 714 | ||||
-rw-r--r-- | src/leveldb/benchmarks/db_bench_tree_db.cc | 522 |
3 files changed, 2217 insertions, 0 deletions
diff --git a/src/leveldb/benchmarks/db_bench.cc b/src/leveldb/benchmarks/db_bench.cc new file mode 100644 index 0000000000..3696023b70 --- /dev/null +++ b/src/leveldb/benchmarks/db_bench.cc @@ -0,0 +1,981 @@ +// 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 <sys/types.h> + +#include "leveldb/cache.h" +#include "leveldb/db.h" +#include "leveldb/env.h" +#include "leveldb/filter_policy.h" +#include "leveldb/write_batch.h" +#include "port/port.h" +#include "util/crc32c.h" +#include "util/histogram.h" +#include "util/mutexlock.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 +// fillsync -- write N/100 values in random key order in sync mode +// fill100K -- write N/1000 100K values in random order in async mode +// deleteseq -- delete N keys in sequential order +// deleterandom -- delete N keys in random order +// readseq -- read N times sequentially +// readreverse -- read N times in reverse order +// readrandom -- read N times in random order +// readmissing -- read N missing keys in random order +// readhot -- read N times in random order from 1% section of DB +// seekrandom -- N random seeks +// open -- cost of opening a DB +// crc32c -- repeated crc32c of 4K of data +// Meta operations: +// compact -- Compact the entire DB +// stats -- Print DB stats +// sstables -- Print sstable info +// heapprofile -- Dump a heap profile (if supported by this port) +static const char* FLAGS_benchmarks = + "fillseq," + "fillsync," + "fillrandom," + "overwrite," + "readrandom," + "readrandom," // Extra run to allow previous compactions to quiesce + "readseq," + "readreverse," + "compact," + "readrandom," + "readseq," + "readreverse," + "fill100K," + "crc32c," + "snappycomp," + "snappyuncomp,"; + +// 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; + +// Number of concurrent threads to run. +static int FLAGS_threads = 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; + +// Number of bytes to buffer in memtable before compacting +// (initialized to default value by "main") +static int FLAGS_write_buffer_size = 0; + +// Number of bytes written to each file. +// (initialized to default value by "main") +static int FLAGS_max_file_size = 0; + +// Approximate size of user data packed per block (before compression. +// (initialized to default value by "main") +static int FLAGS_block_size = 0; + +// Number of bytes to use as a cache of uncompressed data. +// Negative means use default settings. +static int FLAGS_cache_size = -1; + +// Maximum number of files to keep open at the same time (use default if == 0) +static int FLAGS_open_files = 0; + +// Bloom filter bits per key. +// Negative means use default settings. +static int FLAGS_bloom_bits = -1; + +// 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; + +// If true, reuse existing log/MANIFEST files when re-opening a database. +static bool FLAGS_reuse_logs = false; + +// Use the db with the following name. +static const char* FLAGS_db = nullptr; + +namespace leveldb { + +namespace { +leveldb::Env* g_env = nullptr; + +// Helper for quickly generating random data. +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(size_t len) { + if (pos_ + len > data_.size()) { + pos_ = 0; + assert(len < data_.size()); + } + pos_ += len; + return Slice(data_.data() + pos_ - len, len); + } +}; + +#if defined(__linux) +static Slice TrimSpace(Slice s) { + size_t start = 0; + while (start < s.size() && isspace(s[start])) { + start++; + } + size_t limit = s.size(); + while (limit > start && isspace(s[limit - 1])) { + limit--; + } + return Slice(s.data() + start, limit - start); +} +#endif + +static void AppendWithSpace(std::string* str, Slice msg) { + if (msg.empty()) return; + if (!str->empty()) { + str->push_back(' '); + } + str->append(msg.data(), msg.size()); +} + +class Stats { + private: + double start_; + double finish_; + double seconds_; + int done_; + int next_report_; + int64_t bytes_; + double last_op_finish_; + Histogram hist_; + std::string message_; + + public: + Stats() { Start(); } + + void Start() { + next_report_ = 100; + hist_.Clear(); + done_ = 0; + bytes_ = 0; + seconds_ = 0; + message_.clear(); + start_ = finish_ = last_op_finish_ = g_env->NowMicros(); + } + + void Merge(const Stats& other) { + hist_.Merge(other.hist_); + done_ += other.done_; + bytes_ += other.bytes_; + seconds_ += other.seconds_; + if (other.start_ < start_) start_ = other.start_; + if (other.finish_ > finish_) finish_ = other.finish_; + + // Just keep the messages from one thread + if (message_.empty()) message_ = other.message_; + } + + void Stop() { + finish_ = g_env->NowMicros(); + seconds_ = (finish_ - start_) * 1e-6; + } + + void AddMessage(Slice msg) { AppendWithSpace(&message_, msg); } + + void FinishedSingleOp() { + if (FLAGS_histogram) { + double now = g_env->NowMicros(); + double micros = now - last_op_finish_; + 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 AddBytes(int64_t n) { bytes_ += n; } + + void Report(const Slice& name) { + // Pretend at least one op was done in case we are running a benchmark + // that does not call FinishedSingleOp(). + if (done_ < 1) done_ = 1; + + std::string extra; + if (bytes_ > 0) { + // Rate is computed on actual elapsed time, not the sum of per-thread + // elapsed times. + double elapsed = (finish_ - start_) * 1e-6; + char rate[100]; + snprintf(rate, sizeof(rate), "%6.1f MB/s", + (bytes_ / 1048576.0) / elapsed); + extra = rate; + } + AppendWithSpace(&extra, message_); + + fprintf(stdout, "%-12s : %11.3f micros/op;%s%s\n", name.ToString().c_str(), + seconds_ * 1e6 / done_, (extra.empty() ? "" : " "), extra.c_str()); + if (FLAGS_histogram) { + fprintf(stdout, "Microseconds per op:\n%s\n", hist_.ToString().c_str()); + } + fflush(stdout); + } +}; + +// State shared by all concurrent executions of the same benchmark. +struct SharedState { + port::Mutex mu; + port::CondVar cv GUARDED_BY(mu); + int total GUARDED_BY(mu); + + // Each thread goes through the following states: + // (1) initializing + // (2) waiting for others to be initialized + // (3) running + // (4) done + + int num_initialized GUARDED_BY(mu); + int num_done GUARDED_BY(mu); + bool start GUARDED_BY(mu); + + SharedState(int total) + : cv(&mu), total(total), num_initialized(0), num_done(0), start(false) {} +}; + +// Per-thread state for concurrent executions of the same benchmark. +struct ThreadState { + int tid; // 0..n-1 when running in n threads + Random rand; // Has different seeds for different threads + Stats stats; + SharedState* shared; + + ThreadState(int index) : tid(index), rand(1000 + index), shared(nullptr) {} +}; + +} // namespace + +class Benchmark { + private: + Cache* cache_; + const FilterPolicy* filter_policy_; + DB* db_; + int num_; + int value_size_; + int entries_per_batch_; + WriteOptions write_options_; + int reads_; + int heap_counter_; + + 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 + + // See if snappy is working by attempting to compress a compressible string + const char text[] = "yyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyy"; + std::string compressed; + if (!port::Snappy_Compress(text, sizeof(text), &compressed)) { + fprintf(stdout, "WARNING: Snappy compression is not enabled\n"); + } else if (compressed.size() >= sizeof(text)) { + fprintf(stdout, "WARNING: Snappy compression is not effective\n"); + } + } + + void PrintEnvironment() { + fprintf(stderr, "LevelDB: version %d.%d\n", kMajorVersion, + kMinorVersion); + +#if defined(__linux) + time_t now = time(nullptr); + fprintf(stderr, "Date: %s", ctime(&now)); // ctime() adds newline + + FILE* cpuinfo = fopen("/proc/cpuinfo", "r"); + if (cpuinfo != nullptr) { + char line[1000]; + int num_cpus = 0; + std::string cpu_type; + std::string cache_size; + while (fgets(line, sizeof(line), cpuinfo) != nullptr) { + const char* sep = strchr(line, ':'); + if (sep == nullptr) { + 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 + } + + public: + Benchmark() + : cache_(FLAGS_cache_size >= 0 ? NewLRUCache(FLAGS_cache_size) : nullptr), + filter_policy_(FLAGS_bloom_bits >= 0 + ? NewBloomFilterPolicy(FLAGS_bloom_bits) + : nullptr), + db_(nullptr), + num_(FLAGS_num), + value_size_(FLAGS_value_size), + entries_per_batch_(1), + reads_(FLAGS_reads < 0 ? FLAGS_num : FLAGS_reads), + heap_counter_(0) { + std::vector<std::string> files; + g_env->GetChildren(FLAGS_db, &files); + for (size_t i = 0; i < files.size(); i++) { + if (Slice(files[i]).starts_with("heap-")) { + g_env->DeleteFile(std::string(FLAGS_db) + "/" + files[i]); + } + } + if (!FLAGS_use_existing_db) { + DestroyDB(FLAGS_db, Options()); + } + } + + ~Benchmark() { + delete db_; + delete cache_; + delete filter_policy_; + } + + void Run() { + PrintHeader(); + Open(); + + const char* benchmarks = FLAGS_benchmarks; + while (benchmarks != nullptr) { + const char* sep = strchr(benchmarks, ','); + Slice name; + if (sep == nullptr) { + name = benchmarks; + benchmarks = nullptr; + } else { + name = Slice(benchmarks, sep - benchmarks); + benchmarks = sep + 1; + } + + // Reset parameters that may be overridden below + num_ = FLAGS_num; + reads_ = (FLAGS_reads < 0 ? FLAGS_num : FLAGS_reads); + value_size_ = FLAGS_value_size; + entries_per_batch_ = 1; + write_options_ = WriteOptions(); + + void (Benchmark::*method)(ThreadState*) = nullptr; + bool fresh_db = false; + int num_threads = FLAGS_threads; + + if (name == Slice("open")) { + method = &Benchmark::OpenBench; + num_ /= 10000; + if (num_ < 1) num_ = 1; + } else if (name == Slice("fillseq")) { + fresh_db = true; + method = &Benchmark::WriteSeq; + } else if (name == Slice("fillbatch")) { + fresh_db = true; + entries_per_batch_ = 1000; + method = &Benchmark::WriteSeq; + } else if (name == Slice("fillrandom")) { + fresh_db = true; + method = &Benchmark::WriteRandom; + } else if (name == Slice("overwrite")) { + fresh_db = false; + method = &Benchmark::WriteRandom; + } else if (name == Slice("fillsync")) { + fresh_db = true; + num_ /= 1000; + write_options_.sync = true; + method = &Benchmark::WriteRandom; + } else if (name == Slice("fill100K")) { + fresh_db = true; + num_ /= 1000; + value_size_ = 100 * 1000; + method = &Benchmark::WriteRandom; + } else if (name == Slice("readseq")) { + method = &Benchmark::ReadSequential; + } else if (name == Slice("readreverse")) { + method = &Benchmark::ReadReverse; + } else if (name == Slice("readrandom")) { + method = &Benchmark::ReadRandom; + } else if (name == Slice("readmissing")) { + method = &Benchmark::ReadMissing; + } else if (name == Slice("seekrandom")) { + method = &Benchmark::SeekRandom; + } else if (name == Slice("readhot")) { + method = &Benchmark::ReadHot; + } else if (name == Slice("readrandomsmall")) { + reads_ /= 1000; + method = &Benchmark::ReadRandom; + } else if (name == Slice("deleteseq")) { + method = &Benchmark::DeleteSeq; + } else if (name == Slice("deleterandom")) { + method = &Benchmark::DeleteRandom; + } else if (name == Slice("readwhilewriting")) { + num_threads++; // Add extra thread for writing + method = &Benchmark::ReadWhileWriting; + } else if (name == Slice("compact")) { + method = &Benchmark::Compact; + } else if (name == Slice("crc32c")) { + method = &Benchmark::Crc32c; + } else if (name == Slice("snappycomp")) { + method = &Benchmark::SnappyCompress; + } else if (name == Slice("snappyuncomp")) { + method = &Benchmark::SnappyUncompress; + } else if (name == Slice("heapprofile")) { + HeapProfile(); + } else if (name == Slice("stats")) { + PrintStats("leveldb.stats"); + } else if (name == Slice("sstables")) { + PrintStats("leveldb.sstables"); + } else { + if (!name.empty()) { // No error message for empty name + fprintf(stderr, "unknown benchmark '%s'\n", name.ToString().c_str()); + } + } + + if (fresh_db) { + if (FLAGS_use_existing_db) { + fprintf(stdout, "%-12s : skipped (--use_existing_db is true)\n", + name.ToString().c_str()); + method = nullptr; + } else { + delete db_; + db_ = nullptr; + DestroyDB(FLAGS_db, Options()); + Open(); + } + } + + if (method != nullptr) { + RunBenchmark(num_threads, name, method); + } + } + } + + private: + struct ThreadArg { + Benchmark* bm; + SharedState* shared; + ThreadState* thread; + void (Benchmark::*method)(ThreadState*); + }; + + static void ThreadBody(void* v) { + ThreadArg* arg = reinterpret_cast<ThreadArg*>(v); + SharedState* shared = arg->shared; + ThreadState* thread = arg->thread; + { + MutexLock l(&shared->mu); + shared->num_initialized++; + if (shared->num_initialized >= shared->total) { + shared->cv.SignalAll(); + } + while (!shared->start) { + shared->cv.Wait(); + } + } + + thread->stats.Start(); + (arg->bm->*(arg->method))(thread); + thread->stats.Stop(); + + { + MutexLock l(&shared->mu); + shared->num_done++; + if (shared->num_done >= shared->total) { + shared->cv.SignalAll(); + } + } + } + + void RunBenchmark(int n, Slice name, + void (Benchmark::*method)(ThreadState*)) { + SharedState shared(n); + + ThreadArg* arg = new ThreadArg[n]; + for (int i = 0; i < n; i++) { + arg[i].bm = this; + arg[i].method = method; + arg[i].shared = &shared; + arg[i].thread = new ThreadState(i); + arg[i].thread->shared = &shared; + g_env->StartThread(ThreadBody, &arg[i]); + } + + shared.mu.Lock(); + while (shared.num_initialized < n) { + shared.cv.Wait(); + } + + shared.start = true; + shared.cv.SignalAll(); + while (shared.num_done < n) { + shared.cv.Wait(); + } + shared.mu.Unlock(); + + for (int i = 1; i < n; i++) { + arg[0].thread->stats.Merge(arg[i].thread->stats); + } + arg[0].thread->stats.Report(name); + + for (int i = 0; i < n; i++) { + delete arg[i].thread; + } + delete[] arg; + } + + void Crc32c(ThreadState* thread) { + // Checksum about 500MB of data total + const int size = 4096; + const char* label = "(4K per op)"; + std::string data(size, 'x'); + int64_t bytes = 0; + uint32_t crc = 0; + while (bytes < 500 * 1048576) { + crc = crc32c::Value(data.data(), size); + thread->stats.FinishedSingleOp(); + bytes += size; + } + // Print so result is not dead + fprintf(stderr, "... crc=0x%x\r", static_cast<unsigned int>(crc)); + + thread->stats.AddBytes(bytes); + thread->stats.AddMessage(label); + } + + void SnappyCompress(ThreadState* thread) { + RandomGenerator gen; + Slice input = gen.Generate(Options().block_size); + int64_t bytes = 0; + int64_t produced = 0; + bool ok = true; + std::string compressed; + while (ok && bytes < 1024 * 1048576) { // Compress 1G + ok = port::Snappy_Compress(input.data(), input.size(), &compressed); + produced += compressed.size(); + bytes += input.size(); + thread->stats.FinishedSingleOp(); + } + + if (!ok) { + thread->stats.AddMessage("(snappy failure)"); + } else { + char buf[100]; + snprintf(buf, sizeof(buf), "(output: %.1f%%)", + (produced * 100.0) / bytes); + thread->stats.AddMessage(buf); + thread->stats.AddBytes(bytes); + } + } + + void SnappyUncompress(ThreadState* thread) { + RandomGenerator gen; + Slice input = gen.Generate(Options().block_size); + std::string compressed; + bool ok = port::Snappy_Compress(input.data(), input.size(), &compressed); + int64_t bytes = 0; + char* uncompressed = new char[input.size()]; + while (ok && bytes < 1024 * 1048576) { // Compress 1G + ok = port::Snappy_Uncompress(compressed.data(), compressed.size(), + uncompressed); + bytes += input.size(); + thread->stats.FinishedSingleOp(); + } + delete[] uncompressed; + + if (!ok) { + thread->stats.AddMessage("(snappy failure)"); + } else { + thread->stats.AddBytes(bytes); + } + } + + void Open() { + assert(db_ == nullptr); + Options options; + options.env = g_env; + options.create_if_missing = !FLAGS_use_existing_db; + options.block_cache = cache_; + options.write_buffer_size = FLAGS_write_buffer_size; + options.max_file_size = FLAGS_max_file_size; + options.block_size = FLAGS_block_size; + options.max_open_files = FLAGS_open_files; + options.filter_policy = filter_policy_; + options.reuse_logs = FLAGS_reuse_logs; + Status s = DB::Open(options, FLAGS_db, &db_); + if (!s.ok()) { + fprintf(stderr, "open error: %s\n", s.ToString().c_str()); + exit(1); + } + } + + void OpenBench(ThreadState* thread) { + for (int i = 0; i < num_; i++) { + delete db_; + Open(); + thread->stats.FinishedSingleOp(); + } + } + + void WriteSeq(ThreadState* thread) { DoWrite(thread, true); } + + void WriteRandom(ThreadState* thread) { DoWrite(thread, false); } + + void DoWrite(ThreadState* thread, bool seq) { + if (num_ != FLAGS_num) { + char msg[100]; + snprintf(msg, sizeof(msg), "(%d ops)", num_); + thread->stats.AddMessage(msg); + } + + RandomGenerator gen; + WriteBatch batch; + Status s; + int64_t bytes = 0; + for (int i = 0; i < num_; i += entries_per_batch_) { + batch.Clear(); + for (int j = 0; j < entries_per_batch_; j++) { + const int k = seq ? i + j : (thread->rand.Next() % FLAGS_num); + char key[100]; + snprintf(key, sizeof(key), "%016d", k); + batch.Put(key, gen.Generate(value_size_)); + bytes += value_size_ + strlen(key); + thread->stats.FinishedSingleOp(); + } + s = db_->Write(write_options_, &batch); + if (!s.ok()) { + fprintf(stderr, "put error: %s\n", s.ToString().c_str()); + exit(1); + } + } + thread->stats.AddBytes(bytes); + } + + void ReadSequential(ThreadState* thread) { + Iterator* iter = db_->NewIterator(ReadOptions()); + int i = 0; + int64_t bytes = 0; + for (iter->SeekToFirst(); i < reads_ && iter->Valid(); iter->Next()) { + bytes += iter->key().size() + iter->value().size(); + thread->stats.FinishedSingleOp(); + ++i; + } + delete iter; + thread->stats.AddBytes(bytes); + } + + void ReadReverse(ThreadState* thread) { + Iterator* iter = db_->NewIterator(ReadOptions()); + int i = 0; + int64_t bytes = 0; + for (iter->SeekToLast(); i < reads_ && iter->Valid(); iter->Prev()) { + bytes += iter->key().size() + iter->value().size(); + thread->stats.FinishedSingleOp(); + ++i; + } + delete iter; + thread->stats.AddBytes(bytes); + } + + void ReadRandom(ThreadState* thread) { + ReadOptions options; + std::string value; + int found = 0; + for (int i = 0; i < reads_; i++) { + char key[100]; + const int k = thread->rand.Next() % FLAGS_num; + snprintf(key, sizeof(key), "%016d", k); + if (db_->Get(options, key, &value).ok()) { + found++; + } + thread->stats.FinishedSingleOp(); + } + char msg[100]; + snprintf(msg, sizeof(msg), "(%d of %d found)", found, num_); + thread->stats.AddMessage(msg); + } + + void ReadMissing(ThreadState* thread) { + ReadOptions options; + std::string value; + for (int i = 0; i < reads_; i++) { + char key[100]; + const int k = thread->rand.Next() % FLAGS_num; + snprintf(key, sizeof(key), "%016d.", k); + db_->Get(options, key, &value); + thread->stats.FinishedSingleOp(); + } + } + + void ReadHot(ThreadState* thread) { + ReadOptions options; + std::string value; + const int range = (FLAGS_num + 99) / 100; + for (int i = 0; i < reads_; i++) { + char key[100]; + const int k = thread->rand.Next() % range; + snprintf(key, sizeof(key), "%016d", k); + db_->Get(options, key, &value); + thread->stats.FinishedSingleOp(); + } + } + + void SeekRandom(ThreadState* thread) { + ReadOptions options; + int found = 0; + for (int i = 0; i < reads_; i++) { + Iterator* iter = db_->NewIterator(options); + char key[100]; + const int k = thread->rand.Next() % FLAGS_num; + snprintf(key, sizeof(key), "%016d", k); + iter->Seek(key); + if (iter->Valid() && iter->key() == key) found++; + delete iter; + thread->stats.FinishedSingleOp(); + } + char msg[100]; + snprintf(msg, sizeof(msg), "(%d of %d found)", found, num_); + thread->stats.AddMessage(msg); + } + + void DoDelete(ThreadState* thread, bool seq) { + RandomGenerator gen; + WriteBatch batch; + Status s; + for (int i = 0; i < num_; i += entries_per_batch_) { + batch.Clear(); + for (int j = 0; j < entries_per_batch_; j++) { + const int k = seq ? i + j : (thread->rand.Next() % FLAGS_num); + char key[100]; + snprintf(key, sizeof(key), "%016d", k); + batch.Delete(key); + thread->stats.FinishedSingleOp(); + } + s = db_->Write(write_options_, &batch); + if (!s.ok()) { + fprintf(stderr, "del error: %s\n", s.ToString().c_str()); + exit(1); + } + } + } + + void DeleteSeq(ThreadState* thread) { DoDelete(thread, true); } + + void DeleteRandom(ThreadState* thread) { DoDelete(thread, false); } + + void ReadWhileWriting(ThreadState* thread) { + if (thread->tid > 0) { + ReadRandom(thread); + } else { + // Special thread that keeps writing until other threads are done. + RandomGenerator gen; + while (true) { + { + MutexLock l(&thread->shared->mu); + if (thread->shared->num_done + 1 >= thread->shared->num_initialized) { + // Other threads have finished + break; + } + } + + const int k = thread->rand.Next() % FLAGS_num; + char key[100]; + snprintf(key, sizeof(key), "%016d", k); + Status s = db_->Put(write_options_, key, gen.Generate(value_size_)); + if (!s.ok()) { + fprintf(stderr, "put error: %s\n", s.ToString().c_str()); + exit(1); + } + } + + // Do not count any of the preceding work/delay in stats. + thread->stats.Start(); + } + } + + void Compact(ThreadState* thread) { db_->CompactRange(nullptr, nullptr); } + + void PrintStats(const char* key) { + std::string stats; + if (!db_->GetProperty(key, &stats)) { + stats = "(failed)"; + } + fprintf(stdout, "\n%s\n", stats.c_str()); + } + + static void WriteToFile(void* arg, const char* buf, int n) { + reinterpret_cast<WritableFile*>(arg)->Append(Slice(buf, n)); + } + + void HeapProfile() { + char fname[100]; + snprintf(fname, sizeof(fname), "%s/heap-%04d", FLAGS_db, ++heap_counter_); + WritableFile* file; + Status s = g_env->NewWritableFile(fname, &file); + if (!s.ok()) { + fprintf(stderr, "%s\n", s.ToString().c_str()); + return; + } + bool ok = port::GetHeapProfile(WriteToFile, file); + delete file; + if (!ok) { + fprintf(stderr, "heap profiling not supported\n"); + g_env->DeleteFile(fname); + } + } +}; + +} // namespace leveldb + +int main(int argc, char** argv) { + FLAGS_write_buffer_size = leveldb::Options().write_buffer_size; + FLAGS_max_file_size = leveldb::Options().max_file_size; + FLAGS_block_size = leveldb::Options().block_size; + FLAGS_open_files = leveldb::Options().max_open_files; + std::string default_db_path; + + 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], "--use_existing_db=%d%c", &n, &junk) == 1 && + (n == 0 || n == 1)) { + FLAGS_use_existing_db = n; + } else if (sscanf(argv[i], "--reuse_logs=%d%c", &n, &junk) == 1 && + (n == 0 || n == 1)) { + FLAGS_reuse_logs = 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], "--threads=%d%c", &n, &junk) == 1) { + FLAGS_threads = n; + } else if (sscanf(argv[i], "--value_size=%d%c", &n, &junk) == 1) { + FLAGS_value_size = n; + } else if (sscanf(argv[i], "--write_buffer_size=%d%c", &n, &junk) == 1) { + FLAGS_write_buffer_size = n; + } else if (sscanf(argv[i], "--max_file_size=%d%c", &n, &junk) == 1) { + FLAGS_max_file_size = n; + } else if (sscanf(argv[i], "--block_size=%d%c", &n, &junk) == 1) { + FLAGS_block_size = n; + } else if (sscanf(argv[i], "--cache_size=%d%c", &n, &junk) == 1) { + FLAGS_cache_size = 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) { + FLAGS_open_files = n; + } else if (strncmp(argv[i], "--db=", 5) == 0) { + FLAGS_db = argv[i] + 5; + } else { + fprintf(stderr, "Invalid flag '%s'\n", argv[i]); + exit(1); + } + } + + leveldb::g_env = leveldb::Env::Default(); + + // Choose a location for the test database if none given with --db=<path> + if (FLAGS_db == nullptr) { + leveldb::g_env->GetTestDirectory(&default_db_path); + default_db_path += "/dbbench"; + FLAGS_db = default_db_path.c_str(); + } + + leveldb::Benchmark benchmark; + benchmark.Run(); + return 0; +} diff --git a/src/leveldb/benchmarks/db_bench_sqlite3.cc b/src/leveldb/benchmarks/db_bench_sqlite3.cc new file mode 100644 index 0000000000..f183f4fcfd --- /dev/null +++ b/src/leveldb/benchmarks/db_bench_sqlite3.cc @@ -0,0 +1,714 @@ +// 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 <sqlite3.h> +#include <stdio.h> +#include <stdlib.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 +// fillseqsync -- write N/100 values in sequential key order in sync mode +// fillseqbatch -- batch write N values in sequential key order in async mode +// fillrandom -- write N values in random key order in async mode +// fillrandsync -- write N/100 values in random key order in sync mode +// fillrandbatch -- batch write N values in sequential key order in async mode +// overwrite -- overwrite N values in random key order in async mode +// fillrand100K -- write N/1000 100K values in random order in async mode +// fillseq100K -- write N/1000 100K values in sequential order in async mode +// readseq -- read N times sequentially +// readrandom -- read N times in random order +// readrand100K -- read N/1000 100K values in sequential order in async mode +static const char* FLAGS_benchmarks = + "fillseq," + "fillseqsync," + "fillseqbatch," + "fillrandom," + "fillrandsync," + "fillrandbatch," + "overwrite," + "overwritebatch," + "readrandom," + "readseq," + "fillrand100K," + "fillseq100K," + "readseq," + "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; + +// Print histogram of operation timings +static bool FLAGS_histogram = false; + +// Arrange to generate values that shrink to this fraction of +// their original size after compression +static double FLAGS_compression_ratio = 0.5; + +// Page size. Default 1 KB. +static int FLAGS_page_size = 1024; + +// Number of pages. +// Default cache size = FLAGS_page_size * FLAGS_num_pages = 4 MB. +static int FLAGS_num_pages = 4096; + +// 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; + +// If true, we allow batch writes to occur +static bool FLAGS_transaction = true; + +// If true, we enable Write-Ahead Logging +static bool FLAGS_WAL_enabled = true; + +// Use the db with the following name. +static const char* FLAGS_db = nullptr; + +inline static void ExecErrorCheck(int status, char* err_msg) { + if (status != SQLITE_OK) { + fprintf(stderr, "SQL error: %s\n", err_msg); + sqlite3_free(err_msg); + exit(1); + } +} + +inline static void StepErrorCheck(int status) { + if (status != SQLITE_DONE) { + fprintf(stderr, "SQL step error: status = %d\n", status); + exit(1); + } +} + +inline static void ErrorCheck(int status) { + if (status != SQLITE_OK) { + fprintf(stderr, "sqlite3 error: status = %d\n", status); + exit(1); + } +} + +inline static void WalCheckpoint(sqlite3* db_) { + // Flush all writes to disk + if (FLAGS_WAL_enabled) { + sqlite3_wal_checkpoint_v2(db_, nullptr, SQLITE_CHECKPOINT_FULL, nullptr, + nullptr); + } +} + +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: + sqlite3* 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_; + + // 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\n", FLAGS_value_size); + fprintf(stdout, "Entries: %d\n", num_); + fprintf(stdout, "RawSize: %.1f MB (estimated)\n", + ((static_cast<int64_t>(kKeySize + FLAGS_value_size) * 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, "SQLite: version %s\n", SQLITE_VERSION); + +#if defined(__linux) + time_t now = time(nullptr); + fprintf(stderr, "Date: %s", ctime(&now)); // ctime() adds newline + + FILE* cpuinfo = fopen("/proc/cpuinfo", "r"); + if (cpuinfo != nullptr) { + char line[1000]; + int num_cpus = 0; + std::string cpu_type; + std::string cache_size; + while (fgets(line, sizeof(line), cpuinfo) != nullptr) { + const char* sep = strchr(line, ':'); + if (sep == nullptr) { + 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_(nullptr), + db_num_(0), + num_(FLAGS_num), + reads_(FLAGS_reads < 0 ? FLAGS_num : FLAGS_reads), + bytes_(0), + rand_(301) { + std::vector<std::string> files; + std::string test_dir; + Env::Default()->GetTestDirectory(&test_dir); + Env::Default()->GetChildren(test_dir, &files); + if (!FLAGS_use_existing_db) { + for (int i = 0; i < files.size(); i++) { + if (Slice(files[i]).starts_with("dbbench_sqlite3")) { + std::string file_name(test_dir); + file_name += "/"; + file_name += files[i]; + Env::Default()->DeleteFile(file_name.c_str()); + } + } + } + } + + ~Benchmark() { + int status = sqlite3_close(db_); + ErrorCheck(status); + } + + void Run() { + PrintHeader(); + Open(); + + const char* benchmarks = FLAGS_benchmarks; + while (benchmarks != nullptr) { + const char* sep = strchr(benchmarks, ','); + Slice name; + if (sep == nullptr) { + name = benchmarks; + benchmarks = nullptr; + } else { + name = Slice(benchmarks, sep - benchmarks); + benchmarks = sep + 1; + } + + bytes_ = 0; + Start(); + + bool known = true; + bool write_sync = false; + if (name == Slice("fillseq")) { + Write(write_sync, SEQUENTIAL, FRESH, num_, FLAGS_value_size, 1); + WalCheckpoint(db_); + } else if (name == Slice("fillseqbatch")) { + Write(write_sync, SEQUENTIAL, FRESH, num_, FLAGS_value_size, 1000); + WalCheckpoint(db_); + } else if (name == Slice("fillrandom")) { + Write(write_sync, RANDOM, FRESH, num_, FLAGS_value_size, 1); + WalCheckpoint(db_); + } else if (name == Slice("fillrandbatch")) { + Write(write_sync, RANDOM, FRESH, num_, FLAGS_value_size, 1000); + WalCheckpoint(db_); + } else if (name == Slice("overwrite")) { + Write(write_sync, RANDOM, EXISTING, num_, FLAGS_value_size, 1); + WalCheckpoint(db_); + } else if (name == Slice("overwritebatch")) { + Write(write_sync, RANDOM, EXISTING, num_, FLAGS_value_size, 1000); + WalCheckpoint(db_); + } else if (name == Slice("fillrandsync")) { + write_sync = true; + Write(write_sync, RANDOM, FRESH, num_ / 100, FLAGS_value_size, 1); + WalCheckpoint(db_); + } else if (name == Slice("fillseqsync")) { + write_sync = true; + Write(write_sync, SEQUENTIAL, FRESH, num_ / 100, FLAGS_value_size, 1); + WalCheckpoint(db_); + } else if (name == Slice("fillrand100K")) { + Write(write_sync, RANDOM, FRESH, num_ / 1000, 100 * 1000, 1); + WalCheckpoint(db_); + } else if (name == Slice("fillseq100K")) { + Write(write_sync, SEQUENTIAL, FRESH, num_ / 1000, 100 * 1000, 1); + WalCheckpoint(db_); + } else if (name == Slice("readseq")) { + ReadSequential(); + } else if (name == Slice("readrandom")) { + Read(RANDOM, 1); + } else if (name == Slice("readrand100K")) { + int n = reads_; + reads_ /= 1000; + Read(RANDOM, 1); + 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); + } + } + } + + void Open() { + assert(db_ == nullptr); + + int status; + char file_name[100]; + char* err_msg = nullptr; + db_num_++; + + // Open database + std::string tmp_dir; + Env::Default()->GetTestDirectory(&tmp_dir); + snprintf(file_name, sizeof(file_name), "%s/dbbench_sqlite3-%d.db", + tmp_dir.c_str(), db_num_); + status = sqlite3_open(file_name, &db_); + if (status) { + fprintf(stderr, "open error: %s\n", sqlite3_errmsg(db_)); + exit(1); + } + + // Change SQLite cache size + char cache_size[100]; + snprintf(cache_size, sizeof(cache_size), "PRAGMA cache_size = %d", + FLAGS_num_pages); + status = sqlite3_exec(db_, cache_size, nullptr, nullptr, &err_msg); + ExecErrorCheck(status, err_msg); + + // FLAGS_page_size is defaulted to 1024 + if (FLAGS_page_size != 1024) { + char page_size[100]; + snprintf(page_size, sizeof(page_size), "PRAGMA page_size = %d", + FLAGS_page_size); + status = sqlite3_exec(db_, page_size, nullptr, nullptr, &err_msg); + ExecErrorCheck(status, err_msg); + } + + // Change journal mode to WAL if WAL enabled flag is on + if (FLAGS_WAL_enabled) { + std::string WAL_stmt = "PRAGMA journal_mode = WAL"; + + // LevelDB's default cache size is a combined 4 MB + std::string WAL_checkpoint = "PRAGMA wal_autocheckpoint = 4096"; + status = sqlite3_exec(db_, WAL_stmt.c_str(), nullptr, nullptr, &err_msg); + ExecErrorCheck(status, err_msg); + status = + sqlite3_exec(db_, WAL_checkpoint.c_str(), nullptr, nullptr, &err_msg); + ExecErrorCheck(status, err_msg); + } + + // Change locking mode to exclusive and create tables/index for database + std::string locking_stmt = "PRAGMA locking_mode = EXCLUSIVE"; + std::string create_stmt = + "CREATE TABLE test (key blob, value blob, PRIMARY KEY(key))"; + std::string stmt_array[] = {locking_stmt, create_stmt}; + int stmt_array_length = sizeof(stmt_array) / sizeof(std::string); + for (int i = 0; i < stmt_array_length; i++) { + status = + sqlite3_exec(db_, stmt_array[i].c_str(), nullptr, nullptr, &err_msg); + ExecErrorCheck(status, err_msg); + } + } + + void Write(bool write_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; + } + sqlite3_close(db_); + db_ = nullptr; + Open(); + Start(); + } + + if (num_entries != num_) { + char msg[100]; + snprintf(msg, sizeof(msg), "(%d ops)", num_entries); + message_ = msg; + } + + char* err_msg = nullptr; + int status; + + sqlite3_stmt *replace_stmt, *begin_trans_stmt, *end_trans_stmt; + std::string replace_str = "REPLACE INTO test (key, value) VALUES (?, ?)"; + std::string begin_trans_str = "BEGIN TRANSACTION;"; + std::string end_trans_str = "END TRANSACTION;"; + + // Check for synchronous flag in options + std::string sync_stmt = + (write_sync) ? "PRAGMA synchronous = FULL" : "PRAGMA synchronous = OFF"; + status = sqlite3_exec(db_, sync_stmt.c_str(), nullptr, nullptr, &err_msg); + ExecErrorCheck(status, err_msg); + + // Preparing sqlite3 statements + status = sqlite3_prepare_v2(db_, replace_str.c_str(), -1, &replace_stmt, + nullptr); + ErrorCheck(status); + status = sqlite3_prepare_v2(db_, begin_trans_str.c_str(), -1, + &begin_trans_stmt, nullptr); + ErrorCheck(status); + status = sqlite3_prepare_v2(db_, end_trans_str.c_str(), -1, &end_trans_stmt, + nullptr); + ErrorCheck(status); + + bool transaction = (entries_per_batch > 1); + for (int i = 0; i < num_entries; i += entries_per_batch) { + // Begin write transaction + if (FLAGS_transaction && transaction) { + status = sqlite3_step(begin_trans_stmt); + StepErrorCheck(status); + status = sqlite3_reset(begin_trans_stmt); + ErrorCheck(status); + } + + // Create and execute SQL statements + for (int j = 0; j < entries_per_batch; j++) { + const char* value = gen_.Generate(value_size).data(); + + // Create values for key-value pair + const int k = + (order == SEQUENTIAL) ? i + j : (rand_.Next() % num_entries); + char key[100]; + snprintf(key, sizeof(key), "%016d", k); + + // Bind KV values into replace_stmt + status = sqlite3_bind_blob(replace_stmt, 1, key, 16, SQLITE_STATIC); + ErrorCheck(status); + status = sqlite3_bind_blob(replace_stmt, 2, value, value_size, + SQLITE_STATIC); + ErrorCheck(status); + + // Execute replace_stmt + bytes_ += value_size + strlen(key); + status = sqlite3_step(replace_stmt); + StepErrorCheck(status); + + // Reset SQLite statement for another use + status = sqlite3_clear_bindings(replace_stmt); + ErrorCheck(status); + status = sqlite3_reset(replace_stmt); + ErrorCheck(status); + + FinishedSingleOp(); + } + + // End write transaction + if (FLAGS_transaction && transaction) { + status = sqlite3_step(end_trans_stmt); + StepErrorCheck(status); + status = sqlite3_reset(end_trans_stmt); + ErrorCheck(status); + } + } + + status = sqlite3_finalize(replace_stmt); + ErrorCheck(status); + status = sqlite3_finalize(begin_trans_stmt); + ErrorCheck(status); + status = sqlite3_finalize(end_trans_stmt); + ErrorCheck(status); + } + + void Read(Order order, int entries_per_batch) { + int status; + sqlite3_stmt *read_stmt, *begin_trans_stmt, *end_trans_stmt; + + std::string read_str = "SELECT * FROM test WHERE key = ?"; + std::string begin_trans_str = "BEGIN TRANSACTION;"; + std::string end_trans_str = "END TRANSACTION;"; + + // Preparing sqlite3 statements + status = sqlite3_prepare_v2(db_, begin_trans_str.c_str(), -1, + &begin_trans_stmt, nullptr); + ErrorCheck(status); + status = sqlite3_prepare_v2(db_, end_trans_str.c_str(), -1, &end_trans_stmt, + nullptr); + ErrorCheck(status); + status = sqlite3_prepare_v2(db_, read_str.c_str(), -1, &read_stmt, nullptr); + ErrorCheck(status); + + bool transaction = (entries_per_batch > 1); + for (int i = 0; i < reads_; i += entries_per_batch) { + // Begin read transaction + if (FLAGS_transaction && transaction) { + status = sqlite3_step(begin_trans_stmt); + StepErrorCheck(status); + status = sqlite3_reset(begin_trans_stmt); + ErrorCheck(status); + } + + // Create and execute SQL statements + for (int j = 0; j < entries_per_batch; j++) { + // Create key value + char key[100]; + int k = (order == SEQUENTIAL) ? i + j : (rand_.Next() % reads_); + snprintf(key, sizeof(key), "%016d", k); + + // Bind key value into read_stmt + status = sqlite3_bind_blob(read_stmt, 1, key, 16, SQLITE_STATIC); + ErrorCheck(status); + + // Execute read statement + while ((status = sqlite3_step(read_stmt)) == SQLITE_ROW) { + } + StepErrorCheck(status); + + // Reset SQLite statement for another use + status = sqlite3_clear_bindings(read_stmt); + ErrorCheck(status); + status = sqlite3_reset(read_stmt); + ErrorCheck(status); + FinishedSingleOp(); + } + + // End read transaction + if (FLAGS_transaction && transaction) { + status = sqlite3_step(end_trans_stmt); + StepErrorCheck(status); + status = sqlite3_reset(end_trans_stmt); + ErrorCheck(status); + } + } + + status = sqlite3_finalize(read_stmt); + ErrorCheck(status); + status = sqlite3_finalize(begin_trans_stmt); + ErrorCheck(status); + status = sqlite3_finalize(end_trans_stmt); + ErrorCheck(status); + } + + void ReadSequential() { + int status; + sqlite3_stmt* pStmt; + std::string read_str = "SELECT * FROM test ORDER BY key"; + + status = sqlite3_prepare_v2(db_, read_str.c_str(), -1, &pStmt, nullptr); + ErrorCheck(status); + for (int i = 0; i < reads_ && SQLITE_ROW == sqlite3_step(pStmt); i++) { + bytes_ += sqlite3_column_bytes(pStmt, 1) + sqlite3_column_bytes(pStmt, 2); + FinishedSingleOp(); + } + + status = sqlite3_finalize(pStmt); + ErrorCheck(status); + } +}; + +} // namespace leveldb + +int main(int argc, char** argv) { + std::string default_db_path; + 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], "--histogram=%d%c", &n, &junk) == 1 && + (n == 0 || n == 1)) { + FLAGS_histogram = n; + } else if (sscanf(argv[i], "--compression_ratio=%lf%c", &d, &junk) == 1) { + FLAGS_compression_ratio = d; + } else if (sscanf(argv[i], "--use_existing_db=%d%c", &n, &junk) == 1 && + (n == 0 || n == 1)) { + FLAGS_use_existing_db = 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 (leveldb::Slice(argv[i]) == leveldb::Slice("--no_transaction")) { + FLAGS_transaction = false; + } else if (sscanf(argv[i], "--page_size=%d%c", &n, &junk) == 1) { + FLAGS_page_size = n; + } else if (sscanf(argv[i], "--num_pages=%d%c", &n, &junk) == 1) { + FLAGS_num_pages = n; + } else if (sscanf(argv[i], "--WAL_enabled=%d%c", &n, &junk) == 1 && + (n == 0 || n == 1)) { + FLAGS_WAL_enabled = n; + } else if (strncmp(argv[i], "--db=", 5) == 0) { + FLAGS_db = argv[i] + 5; + } else { + fprintf(stderr, "Invalid flag '%s'\n", argv[i]); + exit(1); + } + } + + // Choose a location for the test database if none given with --db=<path> + if (FLAGS_db == nullptr) { + leveldb::Env::Default()->GetTestDirectory(&default_db_path); + default_db_path += "/dbbench"; + FLAGS_db = default_db_path.c_str(); + } + + leveldb::Benchmark benchmark; + benchmark.Run(); + return 0; +} diff --git a/src/leveldb/benchmarks/db_bench_tree_db.cc b/src/leveldb/benchmarks/db_bench_tree_db.cc new file mode 100644 index 0000000000..b2f6646d89 --- /dev/null +++ b/src/leveldb/benchmarks/db_bench_tree_db.cc @@ -0,0 +1,522 @@ +// 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 <kcpolydb.h> +#include <stdio.h> +#include <stdlib.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; + +// Use the db with the following name. +static const char* FLAGS_db = nullptr; + +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(nullptr); + fprintf(stderr, "Date: %s", ctime(&now)); // ctime() adds newline + + FILE* cpuinfo = fopen("/proc/cpuinfo", "r"); + if (cpuinfo != nullptr) { + char line[1000]; + int num_cpus = 0; + std::string cpu_type; + std::string cache_size; + while (fgets(line, sizeof(line), cpuinfo) != nullptr) { + const char* sep = strchr(line, ':'); + if (sep == nullptr) { + 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_(nullptr), + num_(FLAGS_num), + reads_(FLAGS_reads < 0 ? FLAGS_num : FLAGS_reads), + bytes_(0), + rand_(301) { + std::vector<std::string> files; + std::string test_dir; + Env::Default()->GetTestDirectory(&test_dir); + Env::Default()->GetChildren(test_dir.c_str(), &files); + if (!FLAGS_use_existing_db) { + for (int i = 0; i < files.size(); i++) { + if (Slice(files[i]).starts_with("dbbench_polyDB")) { + std::string file_name(test_dir); + file_name += "/"; + file_name += files[i]; + Env::Default()->DeleteFile(file_name.c_str()); + } + } + } + } + + ~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 != nullptr) { + const char* sep = strchr(benchmarks, ','); + Slice name; + if (sep == nullptr) { + name = benchmarks; + benchmarks = nullptr; + } 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); + DBSynchronize(db_); + } 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_ == nullptr); + + // Initialize db_ + db_ = new kyotocabinet::TreeDB(); + char file_name[100]; + db_num_++; + std::string test_dir; + Env::Default()->GetTestDirectory(&test_dir); + snprintf(file_name, sizeof(file_name), "%s/dbbench_polyDB-%d.kct", + test_dir.c_str(), 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_ = nullptr; + 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) { + std::string default_db_path; + 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 if (strncmp(argv[i], "--db=", 5) == 0) { + FLAGS_db = argv[i] + 5; + } else { + fprintf(stderr, "Invalid flag '%s'\n", argv[i]); + exit(1); + } + } + + // Choose a location for the test database if none given with --db=<path> + if (FLAGS_db == nullptr) { + leveldb::Env::Default()->GetTestDirectory(&default_db_path); + default_db_path += "/dbbench"; + FLAGS_db = default_db_path.c_str(); + } + + leveldb::Benchmark benchmark; + benchmark.Run(); + return 0; +} |