openethereum/src/chainfilter.rs

407 lines
11 KiB
Rust
Raw Normal View History

2015-11-28 21:42:50 +01:00
//! Multilevel blockchain bloom filter.
2015-11-28 19:11:04 +01:00
use std::collections::{HashMap};
2015-11-28 17:08:38 +01:00
use hash::*;
use sha3::*;
use num::pow;
2015-11-28 21:42:50 +01:00
/// Represents bloom index in cache
///
/// On cache level 0, every block bloom is represented by different index.
/// On higher cache levels, multiple block blooms are represented by one
/// index. Their `BloomIndex` can be created from block number and given level.
#[derive(Eq, PartialEq, Hash, Clone, Debug)]
pub struct BloomIndex {
pub level: u8,
pub index: usize,
}
impl BloomIndex {
/// Default constructor for `BloomIndex`
pub fn new(level: u8, index: usize) -> BloomIndex {
BloomIndex {
level: level,
index: index,
}
}
}
/// Types implementing this trait should provide read access for bloom filters database.
pub trait FilterDataSource {
/// returns reference to log at given position if it exists
fn bloom_at_index(&self, index: &BloomIndex) -> Option<&H2048>;
}
/// In memory cache for blooms.
///
/// Stores all blooms in HashMap, which indexes them by `BloomIndex`.
2015-11-28 17:08:38 +01:00
pub struct MemoryCache {
blooms: HashMap<BloomIndex, H2048>,
}
impl MemoryCache {
2015-11-28 21:42:50 +01:00
/// Default constructor for MemoryCache
2015-11-28 17:08:38 +01:00
pub fn new() -> MemoryCache {
MemoryCache { blooms: HashMap::new() }
}
2015-11-28 19:11:04 +01:00
/// inserts all blooms into cache
///
2015-11-28 19:48:41 +01:00
/// if bloom at given index already exists, overwrites it
2015-11-28 17:08:38 +01:00
pub fn insert_blooms(&mut self, blooms: HashMap<BloomIndex, H2048>) {
self.blooms.extend(blooms);
}
}
impl FilterDataSource for MemoryCache {
fn bloom_at_index(&self, index: &BloomIndex) -> Option<&H2048> {
self.blooms.get(index)
}
}
2015-11-28 21:42:50 +01:00
/// Should be used for search operations on blockchain.
2015-11-28 17:08:38 +01:00
pub struct ChainFilter<'a, D>
where D: FilterDataSource + 'a
{
data_source: &'a D,
index_size: usize,
level_sizes: Vec<usize>,
2015-11-28 17:08:38 +01:00
}
impl<'a, D> ChainFilter<'a, D> where D: FilterDataSource
{
2015-11-28 21:42:50 +01:00
/// Creates new filter instance.
///
/// Borrows `FilterDataSource` for reading.
2015-11-28 17:08:38 +01:00
pub fn new(data_source: &'a D, index_size: usize, levels: u8) -> Self {
if levels == 0 {
2015-11-28 21:54:41 +01:00
panic!("ChainFilter requires at least 1 level");
}
2015-11-28 17:08:38 +01:00
let mut filter = ChainFilter {
data_source: data_source,
index_size: index_size,
level_sizes: vec![]
2015-11-28 17:08:38 +01:00
};
// cache level sizes, so we do not have to calculate them all the time
for i in 0..levels {
filter.level_sizes.push(pow(index_size, i as usize));
2015-11-28 17:08:38 +01:00
}
filter
}
/// unsafely get level size
fn level_size(&self, level: u8) -> usize {
self.level_sizes[level as usize]
2015-11-28 17:08:38 +01:00
}
/// converts block number and level to `BloomIndex`
fn bloom_index(&self, block_number: usize, level: u8) -> BloomIndex {
BloomIndex {
level: level,
index: block_number / self.level_size(level),
}
}
/// return bloom which are dependencies for given index
2015-11-28 19:11:04 +01:00
///
/// bloom indexes are ordered from lowest to highest
fn lower_level_bloom_indexes(&self, index: &BloomIndex) -> Vec<BloomIndex> {
// this is the lowest level
2015-11-28 17:08:38 +01:00
if index.level == 0 {
return vec![];
2015-11-28 17:08:38 +01:00
}
let new_level = index.level - 1;
let offset = self.index_size * index.index;
(0..self.index_size).map(|i| BloomIndex::new(new_level, offset + i)).collect()
2015-11-28 17:08:38 +01:00
}
2015-11-28 19:11:04 +01:00
/// return number of levels
fn levels(&self) -> u8 {
self.level_sizes.len() as u8
}
2015-11-28 19:11:04 +01:00
/// returns max filter level
fn max_level(&self) -> u8 {
self.level_sizes.len() as u8 - 1
2015-11-28 19:11:04 +01:00
}
2015-11-28 21:42:50 +01:00
/// internal function which does bloom search recursively
2015-11-28 20:44:38 +01:00
fn blocks(&self, bloom: &H2048, from_block: usize, to_block: usize, level: u8, offset: usize) -> Option<Vec<usize>> {
2015-11-28 19:11:04 +01:00
let index = self.bloom_index(offset, level);
2015-11-28 20:44:38 +01:00
match self.data_source.bloom_at_index(&index) {
None => return None,
2015-11-28 19:11:04 +01:00
Some(level_bloom) => match level {
2015-11-28 20:00:14 +01:00
// if we are on the lowest level
// take the value, exclude to_block
2015-11-28 20:44:38 +01:00
0 if offset < to_block => return Some(vec![offset]),
// return None if it is is equal to to_block
0 => return None,
// return None if current level doesnt contain given bloom
_ if !level_bloom.contains(bloom) => return None,
// continue processing && go down
_ => ()
2015-11-28 19:11:04 +01:00
}
2015-11-28 20:00:14 +01:00
};
2015-11-28 20:44:38 +01:00
let level_size = self.level_size(level - 1);
let from_index = self.bloom_index(from_block, level - 1);
let to_index = self.bloom_index(to_block, level - 1);
let res: Vec<usize> = self.lower_level_bloom_indexes(&index).into_iter()
// chose only blooms in range
.filter(|li| li.index >= from_index.index && li.index <= to_index.index)
// map them to offsets
.map(|li| li.index * level_size)
// get all blocks that may contain our bloom
.map(|off| self.blocks(bloom, from_block, to_block, level - 1, off))
// filter existing ones
.filter_map(|x| x)
// flatten nested structures
2015-11-28 20:44:38 +01:00
.flat_map(|v| v)
.collect();
Some(res)
2015-11-28 19:11:04 +01:00
}
2015-11-28 17:08:38 +01:00
2015-11-28 21:42:50 +01:00
/// Adds new bloom to all filter levels
pub fn add_bloom(&self, bloom: &H2048, block_number: usize) -> HashMap<BloomIndex, H2048> {
2015-11-28 17:08:38 +01:00
let mut result: HashMap<BloomIndex, H2048> = HashMap::new();
2015-11-28 17:10:14 +01:00
for level in 0..self.levels() {
2015-11-28 17:10:14 +01:00
let bloom_index = self.bloom_index(block_number, level);
2015-11-28 17:08:38 +01:00
let new_bloom = match self.data_source.bloom_at_index(&bloom_index) {
Some(old_bloom) => old_bloom | bloom,
2015-11-28 17:10:14 +01:00
None => bloom.clone(),
2015-11-28 17:08:38 +01:00
};
result.insert(bloom_index, new_bloom);
}
result
}
2015-11-28 21:42:50 +01:00
/// Adds new blooms starting from block number.
pub fn add_blooms(&self, blooms: &[H2048], block_number: usize) -> HashMap<BloomIndex, H2048> {
2015-11-28 17:08:38 +01:00
let mut result: HashMap<BloomIndex, H2048> = HashMap::new();
for level in 0..self.levels() {
2015-11-28 17:08:38 +01:00
for i in 0..blooms.len() {
let bloom_index = self.bloom_index(block_number + i, level);
let is_new_bloom = match result.get_mut(&bloom_index) {
// it was already modified
2015-11-28 17:10:14 +01:00
Some(to_shift) => {
2015-11-28 17:08:38 +01:00
*to_shift = &blooms[i] | to_shift;
false
2015-11-28 17:10:14 +01:00
}
None => true,
2015-11-28 17:08:38 +01:00
};
// it hasn't been modified yet
if is_new_bloom {
let new_bloom = match self.data_source.bloom_at_index(&bloom_index) {
Some(old_bloom) => old_bloom | &blooms[i],
2015-11-28 17:10:14 +01:00
None => blooms[i].clone(),
2015-11-28 17:08:38 +01:00
};
result.insert(bloom_index, new_bloom);
}
}
}
result
}
2015-11-28 21:42:50 +01:00
/// Resets bloom at level 0 and forces rebuild on higher levels.
pub fn reset_bloom(&self, bloom: &H2048, block_number: usize) -> HashMap<BloomIndex, H2048> {
2015-11-28 17:08:38 +01:00
let mut result: HashMap<BloomIndex, H2048> = HashMap::new();
let mut reset_index = self.bloom_index(block_number, 0);
result.insert(reset_index.clone(), bloom.clone());
2015-11-28 17:10:14 +01:00
for level in 1..self.levels() {
2015-11-28 17:08:38 +01:00
let index = self.bloom_index(block_number, level);
// get all bloom indexes that were used to construct this bloom
2015-11-28 17:08:38 +01:00
let lower_indexes = self.lower_level_bloom_indexes(&index);
let new_bloom = lower_indexes.into_iter()
2015-11-28 21:42:50 +01:00
// skip reseted one
.filter(|li| li != &reset_index)
// get blooms for these indexes
.map(|li| self.data_source.bloom_at_index(&li))
// filter existing ones
.filter_map(|b| b)
// BitOr all of them
.fold(H2048::new(), |acc, bloom| &acc | bloom);
2015-11-28 17:10:14 +01:00
2015-11-28 17:08:38 +01:00
reset_index = index.clone();
result.insert(index, &new_bloom | bloom);
}
result
}
2015-11-28 21:42:50 +01:00
/// Sets lowest level bloom to 0 and forces rebuild on higher levels.
pub fn clear_bloom(&self, block_number: usize) -> HashMap<BloomIndex, H2048> {
2015-11-28 17:08:38 +01:00
self.reset_bloom(&H2048::new(), block_number)
}
2015-11-28 21:42:50 +01:00
/// Returns numbers of blocks that may contain Address.
pub fn blocks_with_address(&self, address: &Address, from_block: usize, to_block: usize) -> Vec<usize> {
2015-11-28 17:08:38 +01:00
let mut bloom = H2048::new();
bloom.shift_bloom(&address.sha3());
self.blocks_with_bloom(&bloom, from_block, to_block)
}
2015-11-28 21:42:50 +01:00
/// Returns numbers of blocks that may contain Topic.
pub fn blocks_with_topics(&self, topic: &H256, from_block: usize, to_block: usize) -> Vec<usize> {
2015-11-28 17:08:38 +01:00
let mut bloom = H2048::new();
bloom.shift_bloom(&topic.sha3());
self.blocks_with_bloom(&bloom, from_block, to_block)
}
2015-11-28 21:42:50 +01:00
/// Returns numbers of blocks that may log bloom.
pub fn blocks_with_bloom(&self, bloom: &H2048, from_block: usize, to_block: usize) -> Vec<usize> {
2015-11-28 19:11:04 +01:00
let mut result = vec![];
// lets start from highest level
let max_level = self.max_level();
let level_size = self.level_size(max_level);
let from_index = self.bloom_index(from_block, max_level);
let to_index = self.bloom_index(to_block, max_level);
for index in from_index.index..to_index.index + 1 {
// offset will be used to calculate where we are right now
let offset = level_size * index;
// go doooown!
2015-11-28 20:44:38 +01:00
match self.blocks(bloom, from_block, to_block, max_level, offset) {
Some(blocks) => result.extend(blocks),
None => ()
};
2015-11-28 19:11:04 +01:00
}
result
2015-11-28 17:08:38 +01:00
}
}
#[cfg(test)]
mod tests {
use hash::*;
use chainfilter::*;
2015-11-28 19:11:04 +01:00
use sha3::*;
use std::str::FromStr;
2015-11-28 17:08:38 +01:00
#[test]
fn test_level_size() {
let cache = MemoryCache::new();
let filter = ChainFilter::new(&cache, 16, 3);
assert_eq!(filter.level_size(0), 1);
assert_eq!(filter.level_size(1), 16);
assert_eq!(filter.level_size(2), 256);
}
#[test]
fn test_bloom_index() {
let cache = MemoryCache::new();
let filter = ChainFilter::new(&cache, 16, 3);
let bi0 = filter.bloom_index(0, 0);
assert_eq!(bi0.level, 0);
assert_eq!(bi0.index, 0);
let bi1 = filter.bloom_index(1, 0);
assert_eq!(bi1.level, 0);
assert_eq!(bi1.index, 1);
let bi2 = filter.bloom_index(2, 0);
assert_eq!(bi2.level, 0);
assert_eq!(bi2.index, 2);
let bi3 = filter.bloom_index(3, 1);
assert_eq!(bi3.level, 1);
assert_eq!(bi3.index, 0);
let bi4 = filter.bloom_index(15, 1);
assert_eq!(bi4.level, 1);
assert_eq!(bi4.index, 0);
let bi5 = filter.bloom_index(16, 1);
assert_eq!(bi5.level, 1);
assert_eq!(bi5.index, 1);
let bi6 = filter.bloom_index(255, 2);
assert_eq!(bi6.level, 2);
assert_eq!(bi6.index, 0);
let bi7 = filter.bloom_index(256, 2);
assert_eq!(bi7.level, 2);
assert_eq!(bi7.index, 1);
}
#[test]
fn test_lower_level_bloom_indexes() {
let cache = MemoryCache::new();
let filter = ChainFilter::new(&cache, 16, 3);
let bi = filter.bloom_index(256, 2);
assert_eq!(bi.level, 2);
assert_eq!(bi.index, 1);
2015-11-28 19:11:04 +01:00
let mut ebis = vec![];
2015-11-28 17:08:38 +01:00
for i in 16..32 {
2015-11-28 19:11:04 +01:00
ebis.push(BloomIndex::new(1, i));
2015-11-28 17:08:38 +01:00
}
let bis = filter.lower_level_bloom_indexes(&bi);
assert_eq!(ebis, bis);
}
2015-11-28 19:11:04 +01:00
#[test]
2015-11-28 19:48:41 +01:00
fn test_topic_basic_search() {
2015-11-28 19:11:04 +01:00
let index_size = 16;
let bloom_levels = 3;
let mut cache = MemoryCache::new();
let topic = H256::from_str("8d936b1bd3fc635710969ccfba471fb17d598d9d1971b538dd712e1e4b4f4dba").unwrap();
let modified_blooms = {
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let block_number = 23;
let mut bloom = H2048::new();
bloom.shift_bloom(&topic.sha3());
filter.add_bloom(&bloom, block_number)
};
// number of modified blooms should always be equal number of levels
assert_eq!(modified_blooms.len(), bloom_levels as usize);
cache.insert_blooms(modified_blooms);
{
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let blocks = filter.blocks_with_topics(&topic, 0, 100);
2015-11-28 19:48:41 +01:00
assert_eq!(blocks.len(), 1);
assert_eq!(blocks[0], 23);
2015-11-28 19:11:04 +01:00
}
2015-11-28 19:48:41 +01:00
{
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let blocks = filter.blocks_with_topics(&topic, 0, 23);
assert_eq!(blocks.len(), 0);
}
{
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let blocks = filter.blocks_with_topics(&topic, 23, 24);
assert_eq!(blocks.len(), 1);
assert_eq!(blocks[0], 23);
}
{
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let blocks = filter.blocks_with_topics(&topic, 24, 100);
assert_eq!(blocks.len(), 0);
}
2015-11-28 19:11:04 +01:00
}
2015-11-28 17:08:38 +01:00
}