implement ancestors as rolling bit field (#17482)

This commit is contained in:
Jeff Washington (jwash) 2021-06-02 09:32:32 -05:00 committed by GitHub
parent 107af52deb
commit eec996ba41
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 28 additions and 138 deletions

View File

@ -1,47 +1,34 @@
use crate::accounts_index::RollingBitField;
use solana_sdk::clock::Slot;
use std::collections::{HashMap, HashSet};
use std::collections::HashMap;
pub type AncestorsForSerialization = HashMap<Slot, usize>;
#[derive(Debug, Default, Clone, PartialEq, Serialize, Deserialize, AbiExample)]
#[derive(Debug, Clone, PartialEq, AbiExample)]
pub struct Ancestors {
min: Slot,
slots: Vec<Option<usize>>,
count: usize,
max: Slot,
large_range_slots: HashSet<Slot>,
ancestors: RollingBitField,
}
// some tests produce ancestors ranges that are too large such
// that we prefer to implement them in a sparse HashMap
const ANCESTORS_HASH_MAP_SIZE: u64 = 10_000;
const ANCESTORS_HASH_MAP_SIZE: u64 = 8192;
impl Default for Ancestors {
fn default() -> Self {
Self {
ancestors: RollingBitField::new(ANCESTORS_HASH_MAP_SIZE),
}
}
}
impl From<Vec<Slot>> for Ancestors {
fn from(source: Vec<Slot>) -> Ancestors {
fn from(mut source: Vec<Slot>) -> Ancestors {
// bitfield performs optimally when we insert the minimum value first so that it knows the correct start/end values
source.sort_unstable();
let mut result = Ancestors::default();
if !source.is_empty() {
result.min = Slot::MAX;
result.max = Slot::MIN;
source.iter().for_each(|slot| {
result.min = std::cmp::min(result.min, *slot);
result.max = std::cmp::max(result.max, *slot + 1);
});
let range = result.range();
if range > ANCESTORS_HASH_MAP_SIZE {
result.large_range_slots = source.into_iter().collect();
result.min = 0;
result.max = 0;
} else {
result.slots = vec![None; range as usize];
source.into_iter().for_each(|slot| {
let slot = result.slot_index(&slot);
if result.slots[slot].is_none() {
result.count += 1;
}
result.slots[slot] = Some(0);
});
}
}
source.into_iter().for_each(|slot| {
result.ancestors.insert(slot);
});
result
}
@ -49,32 +36,8 @@ impl From<Vec<Slot>> for Ancestors {
impl From<&HashMap<Slot, usize>> for Ancestors {
fn from(source: &HashMap<Slot, usize>) -> Ancestors {
let mut result = Ancestors::default();
if !source.is_empty() {
result.min = Slot::MAX;
result.max = Slot::MIN;
source.iter().for_each(|(slot, _)| {
result.min = std::cmp::min(result.min, *slot);
result.max = std::cmp::max(result.max, *slot + 1);
});
let range = result.range();
if range > ANCESTORS_HASH_MAP_SIZE {
result.large_range_slots = source.iter().map(|(slot, _size)| *slot).collect();
result.min = 0;
result.max = 0;
} else {
result.slots = vec![None; range as usize];
source.iter().for_each(|(slot, size)| {
let slot = result.slot_index(&slot);
if result.slots[slot].is_none() {
result.count += 1;
}
result.slots[slot] = Some(*size);
});
}
}
result
let vec = source.iter().map(|(slot, _)| *slot).collect::<Vec<_>>();
Ancestors::from(vec)
}
}
@ -90,75 +53,28 @@ impl From<&Ancestors> for HashMap<Slot, usize> {
impl Ancestors {
pub fn keys(&self) -> Vec<Slot> {
if self.large_range_slots.is_empty() {
self.slots
.iter()
.enumerate()
.filter_map(|(size, i)| i.map(|_| size as u64 + self.min))
.collect::<Vec<_>>()
} else {
self.large_range_slots.iter().copied().collect::<Vec<_>>()
}
self.ancestors.get_all()
}
pub fn get(&self, slot: &Slot) -> bool {
if self.large_range_slots.is_empty() {
if slot < &self.min || slot >= &self.max {
return false;
}
let slot = self.slot_index(slot);
self.slots[slot].is_some()
} else {
self.large_range_slots.get(slot).is_some()
}
self.ancestors.contains(slot)
}
pub fn remove(&mut self, slot: &Slot) {
if self.large_range_slots.is_empty() {
if slot < &self.min || slot >= &self.max {
return;
}
let slot = self.slot_index(slot);
if self.slots[slot].is_some() {
self.count -= 1;
self.slots[slot] = None;
}
} else {
self.large_range_slots.remove(slot);
}
self.ancestors.remove(slot);
}
pub fn contains_key(&self, slot: &Slot) -> bool {
if self.large_range_slots.is_empty() {
if slot < &self.min || slot >= &self.max {
return false;
}
let slot = self.slot_index(slot);
self.slots[slot].is_some()
} else {
self.large_range_slots.contains(slot)
}
self.ancestors.contains(slot)
}
pub fn len(&self) -> usize {
if self.large_range_slots.is_empty() {
self.count
} else {
self.large_range_slots.len()
}
self.ancestors.len()
}
pub fn is_empty(&self) -> bool {
self.len() == 0
}
fn slot_index(&self, slot: &Slot) -> usize {
(slot - self.min) as usize
}
fn range(&self) -> Slot {
self.max - self.min
}
}
#[cfg(test)]
pub mod tests {
@ -187,34 +103,8 @@ pub mod tests {
}
}
impl Ancestors {
pub fn insert(&mut self, mut slot: Slot, size: usize) {
if self.large_range_slots.is_empty() {
if slot < self.min || slot >= self.max {
let new_min = std::cmp::min(self.min, slot);
let new_max = std::cmp::max(self.max, slot + 1);
let new_range = new_max - new_min;
if new_min == self.min {
self.max = slot + 1;
self.slots.resize(new_range as usize, None);
} else {
// min changed
let mut new_slots = vec![None; new_range as usize];
self.slots.iter().enumerate().for_each(|(i, size)| {
new_slots[i as usize + self.min as usize - slot as usize] = *size
});
self.slots = new_slots;
self.min = slot;
// fall through and set this value in
}
}
slot -= self.min;
if self.slots[slot as usize].is_none() {
self.count += 1;
}
self.slots[slot as usize] = Some(size);
} else {
self.large_range_slots.insert(slot);
}
pub fn insert(&mut self, slot: Slot, _size: usize) {
self.ancestors.insert(slot);
}
}