You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
rocksdb/cache/fast_lru_cache.h

312 lines
10 KiB

// Copyright (c) 2011-present, 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).
//
// 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.
#pragma once
#include <memory>
#include <string>
#include "cache/sharded_cache.h"
#include "port/lang.h"
#include "port/malloc.h"
#include "port/port.h"
#include "rocksdb/secondary_cache.h"
#include "util/autovector.h"
Use optimized folly DistributedMutex in LRUCache when available (#10179) Summary: folly DistributedMutex is faster than standard mutexes though imposes some static obligations on usage. See https://github.com/facebook/folly/blob/main/folly/synchronization/DistributedMutex.h for details. Here we use this alternative for our Cache implementations (especially LRUCache) for better locking performance, when RocksDB is compiled with folly. Also added information about which distributed mutex implementation is being used to cache_bench output and to DB LOG. Intended follow-up: * Use DMutex in more places, perhaps improving API to support non-scoped locking * Fix linking with fbcode compiler (needs ROCKSDB_NO_FBCODE=1 currently) Credit: Thanks Siying for reminding me about this line of work that was previously left unfinished. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10179 Test Plan: for correctness, existing tests. CircleCI config updated. Also Meta-internal buck build updated. For performance, ran simultaneous before & after cache_bench. Out of three comparison runs, the middle improvement to ops/sec was +21%: Baseline: USE_CLANG=1 DEBUG_LEVEL=0 make -j24 cache_bench (fbcode compiler) ``` Complete in 20.201 s; Rough parallel ops/sec = 1584062 Thread ops/sec = 107176 Operation latency (ns): Count: 32000000 Average: 9257.9421 StdDev: 122412.04 Min: 134 Median: 3623.0493 Max: 56918500 Percentiles: P50: 3623.05 P75: 10288.02 P99: 30219.35 P99.9: 683522.04 P99.99: 7302791.63 ``` New: (add USE_FOLLY=1) ``` Complete in 16.674 s; Rough parallel ops/sec = 1919135 (+21%) Thread ops/sec = 135487 Operation latency (ns): Count: 32000000 Average: 7304.9294 StdDev: 108530.28 Min: 132 Median: 3777.6012 Max: 91030902 Percentiles: P50: 3777.60 P75: 10169.89 P99: 24504.51 P99.9: 59721.59 P99.99: 1861151.83 ``` Reviewed By: anand1976 Differential Revision: D37182983 Pulled By: pdillinger fbshipit-source-id: a17eb05f25b832b6a2c1356f5c657e831a5af8d1
2 years ago
#include "util/distributed_mutex.h"
namespace ROCKSDB_NAMESPACE {
namespace fast_lru_cache {
// An experimental (under development!) alternative to LRUCache
struct LRUHandle {
void* value;
Cache::DeleterFn deleter;
LRUHandle* next_hash;
LRUHandle* next;
LRUHandle* prev;
size_t total_charge; // TODO(opt): Only allow uint32_t?
size_t key_length;
// The hash of key(). Used for fast sharding and comparisons.
uint32_t hash;
// The number of external refs to this entry. The cache itself is not counted.
uint32_t refs;
enum Flags : uint8_t {
// Whether this entry is referenced by the hash table.
IN_CACHE = (1 << 0),
};
uint8_t flags;
// Beginning of the key (MUST BE THE LAST FIELD IN THIS STRUCT!)
char key_data[1];
Slice key() const { return Slice(key_data, key_length); }
// Increase the reference count by 1.
void Ref() { refs++; }
// Just reduce the reference count by 1. Return true if it was last reference.
bool Unref() {
assert(refs > 0);
refs--;
return refs == 0;
}
// Return true if there are external refs, false otherwise.
bool HasRefs() const { return refs > 0; }
bool InCache() const { return flags & IN_CACHE; }
void SetInCache(bool in_cache) {
if (in_cache) {
flags |= IN_CACHE;
} else {
flags &= ~IN_CACHE;
}
}
void Free() {
assert(refs == 0);
if (deleter) {
(*deleter)(key(), value);
}
delete[] reinterpret_cast<char*>(this);
}
// Calculate the memory usage by metadata.
inline size_t CalcMetaCharge(
CacheMetadataChargePolicy metadata_charge_policy) const {
if (metadata_charge_policy != kFullChargeCacheMetadata) {
return 0;
} else {
#ifdef ROCKSDB_MALLOC_USABLE_SIZE
return malloc_usable_size(
const_cast<void*>(static_cast<const void*>(this)));
#else
// This is the size that is used when a new handle is created.
return sizeof(LRUHandle) - 1 + key_length;
#endif
}
}
inline void CalcTotalCharge(
size_t charge, CacheMetadataChargePolicy metadata_charge_policy) {
total_charge = charge + CalcMetaCharge(metadata_charge_policy);
}
inline size_t GetCharge(
CacheMetadataChargePolicy metadata_charge_policy) const {
size_t meta_charge = CalcMetaCharge(metadata_charge_policy);
assert(total_charge >= meta_charge);
return total_charge - meta_charge;
}
};
// We provide our own simple hash table since it removes a whole bunch
// of porting hacks and is also faster than some of the built-in hash
// table implementations in some of the compiler/runtime combinations
// we have tested. E.g., readrandom speeds up by ~5% over the g++
// 4.4.3's builtin hashtable.
class LRUHandleTable {
public:
explicit LRUHandleTable(int hash_bits);
~LRUHandleTable();
LRUHandle* Lookup(const Slice& key, uint32_t hash);
LRUHandle* Insert(LRUHandle* h);
LRUHandle* Remove(const Slice& key, uint32_t hash);
template <typename T>
void ApplyToEntriesRange(T func, uint32_t index_begin, uint32_t index_end) {
for (uint32_t i = index_begin; i < index_end; i++) {
LRUHandle* h = list_[i];
while (h != nullptr) {
auto n = h->next_hash;
assert(h->InCache());
func(h);
h = n;
}
}
}
int GetLengthBits() const { return length_bits_; }
// Return the address of the head of the chain in the bucket given
// by the hash.
inline LRUHandle** Head(uint32_t hash);
private:
// Return a pointer to slot that points to a cache entry that
// matches key/hash. If there is no such cache entry, return a
// pointer to the trailing slot in the corresponding linked list.
LRUHandle** FindPointer(const Slice& key, uint32_t hash);
// Number of hash bits (upper because lower bits used for sharding)
// used for table index. Length == 1 << length_bits_
int length_bits_;
// The table consists of an array of buckets where each bucket is
// a linked list of cache entries that hash into the bucket.
std::unique_ptr<LRUHandle*[]> list_;
};
// A single shard of sharded cache.
class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard final : public CacheShard {
public:
LRUCacheShard(size_t capacity, size_t estimated_value_size,
bool strict_capacity_limit,
CacheMetadataChargePolicy metadata_charge_policy);
~LRUCacheShard() override = default;
// Separate from constructor so caller can easily make an array of LRUCache
// if current usage is more than new capacity, the function will attempt to
// free the needed space.
void SetCapacity(size_t capacity) override;
// Set the flag to reject insertion if cache if full.
void SetStrictCapacityLimit(bool strict_capacity_limit) override;
// Like Cache methods, but with an extra "hash" parameter.
Status Insert(const Slice& key, uint32_t hash, void* value, size_t charge,
Cache::DeleterFn deleter, Cache::Handle** handle,
Cache::Priority priority) override;
Status Insert(const Slice& key, uint32_t hash, void* value,
const Cache::CacheItemHelper* helper, size_t charge,
Cache::Handle** handle, Cache::Priority priority) override {
return Insert(key, hash, value, charge, helper->del_cb, handle, priority);
}
Cache::Handle* Lookup(const Slice& key, uint32_t hash,
const Cache::CacheItemHelper* /*helper*/,
const Cache::CreateCallback& /*create_cb*/,
Cache::Priority /*priority*/, bool /*wait*/,
Statistics* /*stats*/) override {
return Lookup(key, hash);
}
Cache::Handle* Lookup(const Slice& key, uint32_t hash) override;
bool Release(Cache::Handle* handle, bool /*useful*/,
bool erase_if_last_ref) override {
return Release(handle, erase_if_last_ref);
}
bool IsReady(Cache::Handle* /*handle*/) override { return true; }
void Wait(Cache::Handle* /*handle*/) override {}
bool Ref(Cache::Handle* handle) override;
bool Release(Cache::Handle* handle, bool erase_if_last_ref = false) override;
void Erase(const Slice& key, uint32_t hash) override;
size_t GetUsage() const override;
size_t GetPinnedUsage() const override;
void ApplyToSomeEntries(
const std::function<void(const Slice& key, void* value, size_t charge,
DeleterFn deleter)>& callback,
uint32_t average_entries_per_lock, uint32_t* state) override;
void EraseUnRefEntries() override;
std::string GetPrintableOptions() const override;
private:
friend class LRUCache;
// Insert an item into the hash table and, if handle is null, insert into
// the LRU list. Older items are evicted as necessary. If the cache is full
// and free_handle_on_fail is true, the item is deleted and handle is set to
// nullptr.
Status InsertItem(LRUHandle* item, Cache::Handle** handle,
bool free_handle_on_fail);
void LRU_Remove(LRUHandle* e);
void LRU_Insert(LRUHandle* e);
// Free some space following strict LRU policy until enough space
// to hold (usage_ + charge) is freed or the lru list is empty
// This function is not thread safe - it needs to be executed while
// holding the mutex_.
void EvictFromLRU(size_t charge, autovector<LRUHandle*>* deleted);
// Returns the number of bits used to hash an element in the per-shard
// table.
static int GetHashBits(size_t capacity, size_t estimated_value_size,
CacheMetadataChargePolicy metadata_charge_policy);
// Initialized before use.
size_t capacity_;
// Whether to reject insertion if cache reaches its full capacity.
bool strict_capacity_limit_;
// Dummy head of LRU list.
// lru.prev is newest entry, lru.next is oldest entry.
// LRU contains items which can be evicted, ie reference only by cache
LRUHandle lru_;
// Pointer to head of low-pri pool in LRU list.
LRUHandle* lru_low_pri_;
// ------------^^^^^^^^^^^^^-----------
// Not frequently modified data members
// ------------------------------------
//
// We separate data members that are updated frequently from the ones that
// are not frequently updated so that they don't share the same cache line
// which will lead into false cache sharing
//
// ------------------------------------
// Frequently modified data members
// ------------vvvvvvvvvvvvv-----------
LRUHandleTable table_;
// Memory size for entries residing in the cache.
size_t usage_;
// Memory size for entries residing only in the LRU list.
size_t lru_usage_;
// mutex_ protects the following state.
// We don't count mutex_ as the cache's internal state so semantically we
// don't mind mutex_ invoking the non-const actions.
Use optimized folly DistributedMutex in LRUCache when available (#10179) Summary: folly DistributedMutex is faster than standard mutexes though imposes some static obligations on usage. See https://github.com/facebook/folly/blob/main/folly/synchronization/DistributedMutex.h for details. Here we use this alternative for our Cache implementations (especially LRUCache) for better locking performance, when RocksDB is compiled with folly. Also added information about which distributed mutex implementation is being used to cache_bench output and to DB LOG. Intended follow-up: * Use DMutex in more places, perhaps improving API to support non-scoped locking * Fix linking with fbcode compiler (needs ROCKSDB_NO_FBCODE=1 currently) Credit: Thanks Siying for reminding me about this line of work that was previously left unfinished. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10179 Test Plan: for correctness, existing tests. CircleCI config updated. Also Meta-internal buck build updated. For performance, ran simultaneous before & after cache_bench. Out of three comparison runs, the middle improvement to ops/sec was +21%: Baseline: USE_CLANG=1 DEBUG_LEVEL=0 make -j24 cache_bench (fbcode compiler) ``` Complete in 20.201 s; Rough parallel ops/sec = 1584062 Thread ops/sec = 107176 Operation latency (ns): Count: 32000000 Average: 9257.9421 StdDev: 122412.04 Min: 134 Median: 3623.0493 Max: 56918500 Percentiles: P50: 3623.05 P75: 10288.02 P99: 30219.35 P99.9: 683522.04 P99.99: 7302791.63 ``` New: (add USE_FOLLY=1) ``` Complete in 16.674 s; Rough parallel ops/sec = 1919135 (+21%) Thread ops/sec = 135487 Operation latency (ns): Count: 32000000 Average: 7304.9294 StdDev: 108530.28 Min: 132 Median: 3777.6012 Max: 91030902 Percentiles: P50: 3777.60 P75: 10169.89 P99: 24504.51 P99.9: 59721.59 P99.99: 1861151.83 ``` Reviewed By: anand1976 Differential Revision: D37182983 Pulled By: pdillinger fbshipit-source-id: a17eb05f25b832b6a2c1356f5c657e831a5af8d1
2 years ago
mutable DMutex mutex_;
};
class LRUCache
#ifdef NDEBUG
final
#endif
: public ShardedCache {
public:
LRUCache(size_t capacity, size_t estimated_value_size, int num_shard_bits,
bool strict_capacity_limit,
CacheMetadataChargePolicy metadata_charge_policy =
kDontChargeCacheMetadata);
~LRUCache() override;
const char* Name() const override { return "LRUCache"; }
CacheShard* GetShard(uint32_t shard) override;
const CacheShard* GetShard(uint32_t shard) const override;
void* Value(Handle* handle) override;
size_t GetCharge(Handle* handle) const override;
uint32_t GetHash(Handle* handle) const override;
DeleterFn GetDeleter(Handle* handle) const override;
void DisownData() override;
private:
LRUCacheShard* shards_ = nullptr;
int num_shards_ = 0;
};
} // namespace fast_lru_cache
std::shared_ptr<Cache> NewFastLRUCache(
size_t capacity, size_t estimated_value_size, int num_shard_bits,
bool strict_capacity_limit,
CacheMetadataChargePolicy metadata_charge_policy);
} // namespace ROCKSDB_NAMESPACE