8805ec2f49
Summary: The first step of the `DataBlockHashIndex` implementation. A string based hash table is implemented and unit-tested. `DataBlockHashIndexBuilder`: `Add()` takes pairs of `<key, restart_index>`, and formats it into a string when `Finish()` is called. `DataBlockHashIndex`: initialized by the formatted string, and can interpret it as a hash table. Lookup for a key is supported by iterator operation. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4139 Reviewed By: sagar0 Differential Revision: D8866764 Pulled By: fgwu fbshipit-source-id: 7f015f0098632c65979a22898a50424384730b10
139 lines
5.1 KiB
C++
139 lines
5.1 KiB
C++
// 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).
|
|
|
|
#pragma once
|
|
|
|
#include <string>
|
|
#include <vector>
|
|
|
|
namespace rocksdb {
|
|
// This is an experimental feature aiming to reduce the CPU utilization of
|
|
// point-lookup within a data-block. It is not used in per-table index-blocks.
|
|
// It supports Get(), but not Seek() or Scan(). If the key does not exist,
|
|
// the iterator is set to invalid.
|
|
//
|
|
// A serialized hash index is appended to the data-block. The new block data
|
|
// format is as follows:
|
|
//
|
|
// DATA_BLOCK: [RI RI RI ... RI RI_IDX HASH_IDX FOOTER]
|
|
//
|
|
// RI: Restart Interval (the same as the default data-block format)
|
|
// RI_IDX: Restart Interval index (the same as the default data-block format)
|
|
// HASH_IDX: The new data-block hash index feature.
|
|
// FOOTER: A 32bit block footer, which is the NUM_RESTARTS with the MSB as
|
|
// the flag indicating if this hash index is in use. Note that
|
|
// given a data block < 32KB, the MSB is never used. So we can
|
|
// borrow the MSB as the hash index flag. Besides, this format is
|
|
// compatible with the legacy data-blocks < 32KB, as the MSB is 0.
|
|
//
|
|
// If we zoom in the HASH_IDX, the format of the data-block hash index is as
|
|
// follows:
|
|
//
|
|
// HASH_IDX: [B B B ... B IDX NUM_BUCK MAP_START]
|
|
//
|
|
// B: B = bucket, an array of pairs <TAG, restart index>.
|
|
// TAG is the second hash value of the string. It is used to flag a
|
|
// matching entry among different keys that are hashed to the same
|
|
// bucket. A similar tagging idea is used in [Lim et. al, SOSP'11].
|
|
// However we have a differnet hash design that is not based on cuckoo
|
|
// hashing as Lim's paper is.
|
|
// We do not have to store the length of individual buckets, as they
|
|
// are delimited by the next bucket offset.
|
|
// IDX: Array of offsets of the index hash bucket (relative to MAP_START)
|
|
// NUM_BUCK: Number of buckets, which is the length of the IDX array.
|
|
// MAP_START: the starting offset of the data-block hash index.
|
|
//
|
|
// Each bucket B has the following structure:
|
|
// [TAG RESTART_INDEX][TAG RESTART_INDEX]...[TAG RESTART_INDEX]
|
|
// where TAG is the hash value of the second hash funtion.
|
|
//
|
|
// pairs of <key, restart index> are inserted to the hash index. Queries will
|
|
// first lookup this hash index to find the restart index, then go to the
|
|
// corresponding restart interval to search linearly for the key.
|
|
//
|
|
// For a point-lookup for a key K:
|
|
//
|
|
// Hash1()
|
|
// 1) K ===========> bucket_id
|
|
//
|
|
// 2) Look up this bucket_id in the IDX table to find the offset of the bucket
|
|
//
|
|
// Hash2()
|
|
// 3) K ============> TAG
|
|
// 3) examine the first field (which is TAG) of each entry within this bucket,
|
|
// skip those without a matching TAG.
|
|
// 4) for the entries matching the TAG, get the restart interval index from the
|
|
// second field.
|
|
//
|
|
// (following step are implemented in block.cc)
|
|
// 5) lookup the restart index table (refer to the traditional block format),
|
|
// use the restart interval index to find the offset of the restart interval.
|
|
// 6) linearly search the restart interval for the key.
|
|
//
|
|
|
|
class DataBlockHashIndexBuilder {
|
|
public:
|
|
explicit DataBlockHashIndexBuilder(uint16_t n)
|
|
: num_buckets_(n),
|
|
buckets_(n),
|
|
estimate_((n + 2) *
|
|
sizeof(uint16_t) /* n buckets, 2 num at the end */) {}
|
|
void Add(const Slice& key, const uint16_t& restart_index);
|
|
void Finish(std::string& buffer);
|
|
void Reset();
|
|
inline size_t EstimateSize() { return estimate_; }
|
|
|
|
private:
|
|
uint16_t num_buckets_;
|
|
std::vector<std::vector<uint16_t>> buckets_;
|
|
size_t estimate_;
|
|
};
|
|
|
|
class DataBlockHashIndexIterator;
|
|
|
|
class DataBlockHashIndex {
|
|
public:
|
|
explicit DataBlockHashIndex(Slice block_content);
|
|
|
|
inline uint16_t DataBlockHashMapStart() const {
|
|
return static_cast<uint16_t>(map_start_ - data_);
|
|
}
|
|
|
|
DataBlockHashIndexIterator* NewIterator(const Slice& key) const;
|
|
|
|
private:
|
|
const char *data_;
|
|
// To make the serialized hash index compact and to save the space overhead,
|
|
// here all the data fields persisted in the block are in uint16 format.
|
|
// We find that a uint16 is large enough to index every offset of a 64KiB
|
|
// block.
|
|
// So in other words, DataBlockHashIndex does not support block size equal
|
|
// or greater then 64KiB.
|
|
uint16_t size_;
|
|
uint16_t num_buckets_;
|
|
const char *map_start_; // start of the map
|
|
const char *bucket_table_; // start offset of the bucket index table
|
|
};
|
|
|
|
class DataBlockHashIndexIterator {
|
|
public:
|
|
DataBlockHashIndexIterator(const char* start, const char* end,
|
|
const uint16_t tag)
|
|
: end_(end), tag_(tag) {
|
|
current_ = start - 2 * sizeof(uint16_t);
|
|
Next();
|
|
}
|
|
bool Valid();
|
|
void Next();
|
|
uint16_t Value();
|
|
|
|
private:
|
|
const char* end_; // the end of the bucket
|
|
const uint16_t tag_; // the fingerprint (2nd hash value) of the searching key
|
|
const char* current_;
|
|
};
|
|
|
|
} // namespace rocksdb
|