ARCHIVE/rippled
1
0
Fork 0
mirror of https://github.com/XRPLF/rippled.git synced 2025-12-06 17:27:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

Some refactorings on plain table

Browse Source 
Summary:
Plain table has been working well and this is just a nit-picking patch,
which is generated during my coding reading. No real functional changes.
only some changes regarding:

* Improve some comments from the perspective a "new" code reader.
* Change some magic number to constant, which can help us to parameterize them
  in the future.
* Did some style, naming, C++ convention changes.
* Fix warnings from new "arc lint"

Test Plan: make check

Reviewers: sdong, haobo

CC: leveldb

Differential Revision: https://reviews.facebook.net/D15429
This commit is contained in:
Kai Liu
2014-01-24 21:10:19 -08:00
parent 0ab766132b
commit 4b51dffcf8
3 changed files with 200 additions and 202 deletions
Download Patch File Download Diff File Expand all files Collapse all files

125
table/plain_table_reader.h
View File

@@ -5,6 +5,7 @@
#pragma once
#include <unordered_map>
#include <memory>
#include <vector>
#include <stdint.h>
#include "rocksdb/env.h"
#include "rocksdb/iterator.h"
@@ -35,7 +36,7 @@ using std::unordered_map;
//
// The implementation of IndexedTableReader requires output file is mmaped
class PlainTableReader: public TableReader {
public:
public:
static Status Open(const Options& options, const EnvOptions& soptions,
unique_ptr<RandomAccessFile> && file, uint64_t file_size,
unique_ptr<TableReader>* table, const int bloom_num_bits,
@@ -65,12 +66,12 @@ public:
const TableProperties& table_properties);
~PlainTableReader();
private:
private:
struct IndexRecord;
class IndexRecordList;
uint32_t* hash_table_ = nullptr;
int hash_table_size_;
int hash_table_size_ = 0;
char* sub_index_ = nullptr;
Options options_;
@@ -82,24 +83,30 @@ private:
uint32_t version_;
uint32_t file_size_;
const double hash_table_ratio_;
const int bloom_bits_per_key_;
DynamicBloom* bloom_;
const double kHashTableRatio;
const int kBloomBitsPerKey;
DynamicBloom* bloom_ = nullptr;
TableProperties table_properties_;
const uint32_t data_start_offset_;
const uint32_t data_start_offset_ = 0;
const uint32_t data_end_offset_;
const size_t user_key_len_;
static const size_t kNumInternalBytes = 8;
static const uint32_t kSubIndexMask = 0x80000000;
static const size_t kOffsetLen = sizeof(uint32_t);
static const uint64_t kMaxFileSize = 1u << 31;
static const size_t kRecordsPerGroup = 256;
// To speed up the search for keys with same prefix, we'll add index key for
// every N keys, where the "N" is determined by
// kIndexIntervalForSamePrefixKeys
static const size_t kIndexIntervalForSamePrefixKeys = 16;
bool IsFixedLength() {
bool IsFixedLength() const {
return user_key_len_ != PlainTableFactory::kVariableLength;
}
size_t GetFixedInternalKeyLength() {
size_t GetFixedInternalKeyLength() const {
return user_key_len_ + kNumInternalBytes;
}
@@ -108,32 +115,67 @@ private:
// Internal helper function to generate an IndexRecordList object from all
// the rows, which contains index records as a list.
int PopulateIndexRecordList(IndexRecordList& record_list);
int PopulateIndexRecordList(IndexRecordList* record_list);
// Internal helper function to allocate memory for indexes and bloom filters
void Allocate(int num_prefixes);
void AllocateIndexAndBloom(int num_prefixes);
// Internal helper function to bucket index record list to hash buckets.
// hash2offsets is sized of of hash_table_size_, each contains a linked list
// hash_to_offsets is sized of of hash_table_size_, each contains a linked
// list
// of offsets for the hash, in reversed order.
// bucket_count is sized of hash_table_size_. The value is how many index
// records are there in hash2offsets for the same bucket.
// records are there in hash_to_offsets for the same bucket.
size_t BucketizeIndexesAndFillBloom(
IndexRecordList& record_list, int num_prefixes,
std::vector<IndexRecord*>& hash2offsets,
std::vector<uint32_t>& bucket_count);
std::vector<IndexRecord*>* hash_to_offsets,
std::vector<uint32_t>* bucket_count);
// Internal helper class to fill the indexes and bloom filters to internal
// data structures. hash2offsets and bucket_count are bucketized indexes and
// counts generated by BucketizeIndexesAndFillBloom().
// data structures. hash_to_offsets and bucket_count are bucketized indexes
// and counts generated by BucketizeIndexesAndFillBloom().
void FillIndexes(size_t sub_index_size_needed,
std::vector<IndexRecord*>& hash2offsets,
std::vector<uint32_t>& bucket_count);
const std::vector<IndexRecord*>& hash_to_offsets,
const std::vector<uint32_t>& bucket_count);
// Populate the internal indexes. It must be called before
// any query to the table.
// This query will populate the hash table hash_table_, the second
// level of indexes sub_index_ and bloom filter filter_slice_ if enabled.
// PopulateIndex() builds index of keys. It must be called before any query
// to the table.
//
// hash_table_ contains buckets size of hash_table_size_, each is a 32-bit
// integer. The lower 31 bits contain an offset value (explained below) and
// the first bit of the integer indicates type of the offset.
//
// +--------------+------------------------------------------------------+
// | Flag (1 bit) | Offset to binary search buffer or file (31 bits) +
// +--------------+------------------------------------------------------+
//
// Explanation for the "flag bit":
//
// 0 indicates that the bucket contains only one prefix (no conflict when
// hashing this prefix), whose first row starts from this offset of the
// file.
// 1 indicates that the bucket contains more than one prefixes, or there
// are too many rows for one prefix so we need a binary search for it. In
// this case, the offset indicates the offset of sub_index_ holding the
// binary search indexes of keys for those rows. Those binary search indexes
// are organized in this way:
//
// The first 4 bytes, indicate how many indexes (N) are stored after it. After
// it, there are N 32-bit integers, each points of an offset of the file,
// which
// points to starting of a row. Those offsets need to be guaranteed to be in
// ascending order so the keys they are pointing to are also in ascending
// order
// to make sure we can use them to do binary searches. Below is visual
// presentation of a bucket.
//
// <begin>
// number_of_records: varint32
// record 1 file offset: fixedint32
// record 2 file offset: fixedint32
// ....
// record N file offset: fixedint32
// <end>
Status PopulateIndex();
// Check bloom filter to see whether it might contain this prefix.
@@ -163,41 +205,4 @@ private:
explicit PlainTableReader(const TableReader&) = delete;
void operator=(const TableReader&) = delete;
};
// Iterator to iterate IndexedTable
class PlainTableIterator: public Iterator {
public:
explicit PlainTableIterator(PlainTableReader* table);
~PlainTableIterator();
bool Valid() const;
void SeekToFirst();
void SeekToLast();
void Seek(const Slice& target);
void Next();
void Prev();
Slice key() const;
Slice value() const;
Status status() const;
private:
PlainTableReader* table_;
uint32_t offset_;
uint32_t next_offset_;
Slice key_;
Slice value_;
Status status_;
// No copying allowed
PlainTableIterator(const PlainTableIterator&) = delete;
void operator=(const Iterator&) = delete;
};
} // namespace rocksdb