rocksdb/utilities/persistent_cache/hash_table_bench.cc

311 lines
8.1 KiB
C++
Raw Normal View History

// Copyright (c) 2013, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
//
#if !defined(OS_WIN)
#ifndef GFLAGS
#include <cstdio>
int main() { fprintf(stderr, "Please install gflags to run tools\n"); }
#else
#include <sys/time.h>
#include <unistd.h>
#include <atomic>
#include <functional>
#include <string>
#include <unordered_map>
#include "port/port_posix.h"
#include "port/sys_time.h"
#include "rocksdb/env.h"
#include "util/gflags_compat.h"
#include "util/mutexlock.h"
Fix build for linux Summary: * Include `unistd.h` for `sleep(3)` * Include `sys/time.h` for `gettimeofday(3)` * Include `utils/random.h` for `Random64` Error messages: utilities/persistent_cache/hash_table_bench.cc: In constructor ‘rocksdb::HashTableBenchmark::HashTableBenchmark(rocksdb::HashTableImpl<long unsigned int, std::__cxx11::basic_string<char> >*, size_t, size_t, size_t, size_t)’: utilities/persistent_cache/hash_table_bench.cc:76:28: error: ‘sleep’ was not declared in this scope /* sleep override */ sleep(1); ^~~~~ utilities/persistent_cache/hash_table_bench.cc:76:28: note: suggested alternative: ‘strsep’ /* sleep override */ sleep(1); ^~~~~ strsep utilities/persistent_cache/hash_table_bench.cc: In member function ‘void rocksdb::HashTableBenchmark::RunRead()’: utilities/persistent_cache/hash_table_bench.cc:107:5: error: ‘Random64’ was not declared in this scope Random64 rgen(time(nullptr)); ^~~~~~~~ utilities/persistent_cache/hash_table_bench.cc:107:5: note: suggested alternative: ‘random_r’ Random64 rgen(time(nullptr)); ^~~~~~~~ random_r utilities/persistent_cache/hash_table_bench.cc:110:18: error: ‘rgen’ was not declared in this scope size_t k = rgen.Next() % max_prepop_key; ^~~~ utilities/persistent_cache/hash_table_bench.cc: In static member function ‘static uint64_t rocksdb::HashTableBenchmark::NowInMillSec()’: utilities/persistent_cache/hash_table_bench.cc:153:5: error: ‘gettimeofday’ was not declared in this scope gettimeofday(&tv, /*tz=*/nullptr); ^~~~~~~~~~~~ make[2]: *** [CMakeFiles/hash_table_bench.dir/build.make:63: CMakeFiles/hash_table_bench.dir/utilities/persistent_cache/hash_table_bench.cc.o] Error 1 make[1]: *** [CMakeFiles/Makefile2:3346: CMakeFiles/hash_table_bench.dir/all] Error 2 make[1]: *** Waiting for unfinished jobs.... Closes https://github.com/facebook/rocksdb/pull/3283 Differential Revision: D6594850 Pulled By: ajkr fbshipit-source-id: fd83957338c210cdfd253763347aafd39476824f
2017-12-18 20:23:40 +00:00
#include "util/random.h"
#include "utilities/persistent_cache/hash_table.h"
using std::string;
DEFINE_int32(nsec, 10, "nsec");
DEFINE_int32(nthread_write, 1, "insert %");
DEFINE_int32(nthread_read, 1, "lookup %");
DEFINE_int32(nthread_erase, 1, "erase %");
namespace ROCKSDB_NAMESPACE {
//
// HashTableImpl interface
//
// Abstraction of a hash table implementation
template <class Key, class Value>
class HashTableImpl {
public:
virtual ~HashTableImpl() {}
virtual bool Insert(const Key& key, const Value& val) = 0;
virtual bool Erase(const Key& key) = 0;
virtual bool Lookup(const Key& key, Value* val) = 0;
};
// HashTableBenchmark
//
// Abstraction to test a given hash table implementation. The test mostly
// focus on insert, lookup and erase. The test can operate in test mode and
// benchmark mode.
class HashTableBenchmark {
public:
explicit HashTableBenchmark(HashTableImpl<size_t, std::string>* impl,
const size_t sec = 10,
const size_t nthread_write = 1,
const size_t nthread_read = 1,
const size_t nthread_erase = 1)
: impl_(impl),
sec_(sec),
ninserts_(0),
nreads_(0),
nerases_(0),
nerases_failed_(0),
quit_(false) {
Prepop();
StartThreads(nthread_write, WriteMain);
StartThreads(nthread_read, ReadMain);
StartThreads(nthread_erase, EraseMain);
uint64_t start = NowInMillSec();
while (!quit_) {
quit_ = NowInMillSec() - start > sec_ * 1000;
/* sleep override */ sleep(1);
}
Env* env = Env::Default();
env->WaitForJoin();
if (sec_) {
printf("Result \n");
printf("====== \n");
printf("insert/sec = %f \n", ninserts_ / static_cast<double>(sec_));
printf("read/sec = %f \n", nreads_ / static_cast<double>(sec_));
printf("erases/sec = %f \n", nerases_ / static_cast<double>(sec_));
const uint64_t ops = ninserts_ + nreads_ + nerases_;
printf("ops/sec = %f \n", ops / static_cast<double>(sec_));
printf("erase fail = %d (%f%%)\n", static_cast<int>(nerases_failed_),
static_cast<float>(nerases_failed_ / nerases_ * 100));
printf("====== \n");
}
}
void RunWrite() {
while (!quit_) {
size_t k = insert_key_++;
std::string tmp(1000, k % 255);
bool status = impl_->Insert(k, tmp);
assert(status);
ninserts_++;
}
}
void RunRead() {
Random64 rgen(time(nullptr));
while (!quit_) {
std::string s;
size_t k = rgen.Next() % max_prepop_key;
bool status = impl_->Lookup(k, &s);
assert(status);
assert(s == std::string(1000, k % 255));
nreads_++;
}
}
void RunErase() {
while (!quit_) {
size_t k = erase_key_++;
bool status = impl_->Erase(k);
nerases_failed_ += !status;
nerases_++;
}
}
private:
// Start threads for a given function
void StartThreads(const size_t n, void (*fn)(void*)) {
Env* env = Env::Default();
for (size_t i = 0; i < n; ++i) {
env->StartThread(fn, this);
}
}
// Prepop the hash table with 1M keys
void Prepop() {
for (size_t i = 0; i < max_prepop_key; ++i) {
bool status = impl_->Insert(i, std::string(1000, i % 255));
assert(status);
}
erase_key_ = insert_key_ = max_prepop_key;
for (size_t i = 0; i < 10 * max_prepop_key; ++i) {
bool status = impl_->Insert(insert_key_++, std::string(1000, 'x'));
assert(status);
}
}
static uint64_t NowInMillSec() {
port::TimeVal tv;
port::GetTimeOfDay(&tv, /*tz=*/nullptr);
return tv.tv_sec * 1000 + tv.tv_usec / 1000;
}
//
// Wrapper functions for thread entry
//
static void WriteMain(void* args) {
Prefer static_cast in place of most reinterpret_cast (#12308) Summary: The following are risks associated with pointer-to-pointer reinterpret_cast: * Can produce the "wrong result" (crash or memory corruption). IIRC, in theory this can happen for any up-cast or down-cast for a non-standard-layout type, though in practice would only happen for multiple inheritance cases (where the base class pointer might be "inside" the derived object). We don't use multiple inheritance a lot, but we do. * Can mask useful compiler errors upon code change, including converting between unrelated pointer types that you are expecting to be related, and converting between pointer and scalar types unintentionally. I can only think of some obscure cases where static_cast could be troublesome when it compiles as a replacement: * Going through `void*` could plausibly cause unnecessary or broken pointer arithmetic. Suppose we have `struct Derived: public Base1, public Base2`. If we have `Derived*` -> `void*` -> `Base2*` -> `Derived*` through reinterpret casts, this could plausibly work (though technical UB) assuming the `Base2*` is not dereferenced. Changing to static cast could introduce breaking pointer arithmetic. * Unnecessary (but safe) pointer arithmetic could arise in a case like `Derived*` -> `Base2*` -> `Derived*` where before the Base2 pointer might not have been dereferenced. This could potentially affect performance. With some light scripting, I tried replacing pointer-to-pointer reinterpret_casts with static_cast and kept the cases that still compile. Most occurrences of reinterpret_cast have successfully been changed (except for java/ and third-party/). 294 changed, 257 remain. A couple of related interventions included here: * Previously Cache::Handle was not actually derived from in the implementations and just used as a `void*` stand-in with reinterpret_cast. Now there is a relationship to allow static_cast. In theory, this could introduce pointer arithmetic (as described above) but is unlikely without multiple inheritance AND non-empty Cache::Handle. * Remove some unnecessary casts to void* as this is allowed to be implicit (for better or worse). Most of the remaining reinterpret_casts are for converting to/from raw bytes of objects. We could consider better idioms for these patterns in follow-up work. I wish there were a way to implement a template variant of static_cast that would only compile if no pointer arithmetic is generated, but best I can tell, this is not possible. AFAIK the best you could do is a dynamic check that the void* conversion after the static cast is unchanged. Pull Request resolved: https://github.com/facebook/rocksdb/pull/12308 Test Plan: existing tests, CI Reviewed By: ltamasi Differential Revision: D53204947 Pulled By: pdillinger fbshipit-source-id: 9de23e618263b0d5b9820f4e15966876888a16e2
2024-02-07 18:44:11 +00:00
static_cast<HashTableBenchmark*>(args)->RunWrite();
}
static void ReadMain(void* args) {
Prefer static_cast in place of most reinterpret_cast (#12308) Summary: The following are risks associated with pointer-to-pointer reinterpret_cast: * Can produce the "wrong result" (crash or memory corruption). IIRC, in theory this can happen for any up-cast or down-cast for a non-standard-layout type, though in practice would only happen for multiple inheritance cases (where the base class pointer might be "inside" the derived object). We don't use multiple inheritance a lot, but we do. * Can mask useful compiler errors upon code change, including converting between unrelated pointer types that you are expecting to be related, and converting between pointer and scalar types unintentionally. I can only think of some obscure cases where static_cast could be troublesome when it compiles as a replacement: * Going through `void*` could plausibly cause unnecessary or broken pointer arithmetic. Suppose we have `struct Derived: public Base1, public Base2`. If we have `Derived*` -> `void*` -> `Base2*` -> `Derived*` through reinterpret casts, this could plausibly work (though technical UB) assuming the `Base2*` is not dereferenced. Changing to static cast could introduce breaking pointer arithmetic. * Unnecessary (but safe) pointer arithmetic could arise in a case like `Derived*` -> `Base2*` -> `Derived*` where before the Base2 pointer might not have been dereferenced. This could potentially affect performance. With some light scripting, I tried replacing pointer-to-pointer reinterpret_casts with static_cast and kept the cases that still compile. Most occurrences of reinterpret_cast have successfully been changed (except for java/ and third-party/). 294 changed, 257 remain. A couple of related interventions included here: * Previously Cache::Handle was not actually derived from in the implementations and just used as a `void*` stand-in with reinterpret_cast. Now there is a relationship to allow static_cast. In theory, this could introduce pointer arithmetic (as described above) but is unlikely without multiple inheritance AND non-empty Cache::Handle. * Remove some unnecessary casts to void* as this is allowed to be implicit (for better or worse). Most of the remaining reinterpret_casts are for converting to/from raw bytes of objects. We could consider better idioms for these patterns in follow-up work. I wish there were a way to implement a template variant of static_cast that would only compile if no pointer arithmetic is generated, but best I can tell, this is not possible. AFAIK the best you could do is a dynamic check that the void* conversion after the static cast is unchanged. Pull Request resolved: https://github.com/facebook/rocksdb/pull/12308 Test Plan: existing tests, CI Reviewed By: ltamasi Differential Revision: D53204947 Pulled By: pdillinger fbshipit-source-id: 9de23e618263b0d5b9820f4e15966876888a16e2
2024-02-07 18:44:11 +00:00
static_cast<HashTableBenchmark*>(args)->RunRead();
}
static void EraseMain(void* args) {
Prefer static_cast in place of most reinterpret_cast (#12308) Summary: The following are risks associated with pointer-to-pointer reinterpret_cast: * Can produce the "wrong result" (crash or memory corruption). IIRC, in theory this can happen for any up-cast or down-cast for a non-standard-layout type, though in practice would only happen for multiple inheritance cases (where the base class pointer might be "inside" the derived object). We don't use multiple inheritance a lot, but we do. * Can mask useful compiler errors upon code change, including converting between unrelated pointer types that you are expecting to be related, and converting between pointer and scalar types unintentionally. I can only think of some obscure cases where static_cast could be troublesome when it compiles as a replacement: * Going through `void*` could plausibly cause unnecessary or broken pointer arithmetic. Suppose we have `struct Derived: public Base1, public Base2`. If we have `Derived*` -> `void*` -> `Base2*` -> `Derived*` through reinterpret casts, this could plausibly work (though technical UB) assuming the `Base2*` is not dereferenced. Changing to static cast could introduce breaking pointer arithmetic. * Unnecessary (but safe) pointer arithmetic could arise in a case like `Derived*` -> `Base2*` -> `Derived*` where before the Base2 pointer might not have been dereferenced. This could potentially affect performance. With some light scripting, I tried replacing pointer-to-pointer reinterpret_casts with static_cast and kept the cases that still compile. Most occurrences of reinterpret_cast have successfully been changed (except for java/ and third-party/). 294 changed, 257 remain. A couple of related interventions included here: * Previously Cache::Handle was not actually derived from in the implementations and just used as a `void*` stand-in with reinterpret_cast. Now there is a relationship to allow static_cast. In theory, this could introduce pointer arithmetic (as described above) but is unlikely without multiple inheritance AND non-empty Cache::Handle. * Remove some unnecessary casts to void* as this is allowed to be implicit (for better or worse). Most of the remaining reinterpret_casts are for converting to/from raw bytes of objects. We could consider better idioms for these patterns in follow-up work. I wish there were a way to implement a template variant of static_cast that would only compile if no pointer arithmetic is generated, but best I can tell, this is not possible. AFAIK the best you could do is a dynamic check that the void* conversion after the static cast is unchanged. Pull Request resolved: https://github.com/facebook/rocksdb/pull/12308 Test Plan: existing tests, CI Reviewed By: ltamasi Differential Revision: D53204947 Pulled By: pdillinger fbshipit-source-id: 9de23e618263b0d5b9820f4e15966876888a16e2
2024-02-07 18:44:11 +00:00
static_cast<HashTableBenchmark*>(args)->RunErase();
}
HashTableImpl<size_t, std::string>* impl_; // Implementation to test
const size_t sec_; // Test time
const size_t max_prepop_key = 1ULL * 1024 * 1024; // Max prepop key
std::atomic<size_t> insert_key_; // Last inserted key
std::atomic<size_t> erase_key_; // Erase key
std::atomic<size_t> ninserts_; // Number of inserts
std::atomic<size_t> nreads_; // Number of reads
std::atomic<size_t> nerases_; // Number of erases
std::atomic<size_t> nerases_failed_; // Number of erases failed
bool quit_; // Should the threads quit ?
};
//
// SimpleImpl
// Lock safe unordered_map implementation
class SimpleImpl : public HashTableImpl<size_t, string> {
public:
bool Insert(const size_t& key, const string& val) override {
WriteLock _(&rwlock_);
map_.insert(make_pair(key, val));
return true;
}
bool Erase(const size_t& key) override {
WriteLock _(&rwlock_);
auto it = map_.find(key);
if (it == map_.end()) {
return false;
}
map_.erase(it);
return true;
}
bool Lookup(const size_t& key, string* val) override {
ReadLock _(&rwlock_);
auto it = map_.find(key);
if (it != map_.end()) {
*val = it->second;
}
return it != map_.end();
}
private:
port::RWMutex rwlock_;
std::unordered_map<size_t, string> map_;
};
//
// GranularLockImpl
// Thread safe custom RocksDB implementation of hash table with granular
// locking
class GranularLockImpl : public HashTableImpl<size_t, string> {
public:
bool Insert(const size_t& key, const string& val) override {
Node n(key, val);
return impl_.Insert(n);
}
bool Erase(const size_t& key) override {
Node n(key, string());
return impl_.Erase(n, nullptr);
}
bool Lookup(const size_t& key, string* val) override {
Node n(key, string());
port::RWMutex* rlock;
bool status = impl_.Find(n, &n, &rlock);
if (status) {
ReadUnlock _(rlock);
*val = n.val_;
}
return status;
}
private:
struct Node {
explicit Node(const size_t key, const string& val) : key_(key), val_(val) {}
size_t key_ = 0;
string val_;
};
struct Hash {
uint64_t operator()(const Node& node) {
return std::hash<uint64_t>()(node.key_);
}
};
struct Equal {
bool operator()(const Node& lhs, const Node& rhs) {
return lhs.key_ == rhs.key_;
}
};
HashTable<Node, Hash, Equal> impl_;
};
} // namespace ROCKSDB_NAMESPACE
//
// main
//
int main(int argc, char** argv) {
GFLAGS_NAMESPACE::SetUsageMessage(std::string("\nUSAGE:\n") +
std::string(argv[0]) + " [OPTIONS]...");
GFLAGS_NAMESPACE::ParseCommandLineFlags(&argc, &argv, false);
//
// Micro benchmark unordered_map
//
printf("Micro benchmarking std::unordered_map \n");
{
ROCKSDB_NAMESPACE::SimpleImpl impl;
ROCKSDB_NAMESPACE::HashTableBenchmark _(
&impl, FLAGS_nsec, FLAGS_nthread_write, FLAGS_nthread_read,
FLAGS_nthread_erase);
}
//
// Micro benchmark scalable hash table
//
printf("Micro benchmarking scalable hash map \n");
{
ROCKSDB_NAMESPACE::GranularLockImpl impl;
ROCKSDB_NAMESPACE::HashTableBenchmark _(
&impl, FLAGS_nsec, FLAGS_nthread_write, FLAGS_nthread_read,
FLAGS_nthread_erase);
}
return 0;
}
#endif // #ifndef GFLAGS
#else
int main(int /*argc*/, char** /*argv*/) { return 0; }
#endif