solana/core/src/crds_gossip_pull.rs

418 lines
16 KiB
Rust

//! Crds Gossip Pull overlay
//! This module implements the anti-entropy protocol for the network.
//!
//! The basic strategy is as follows:
//! 1. Construct a bloom filter of the local data set
//! 2. Randomly ask a node on the network for data that is not contained in the bloom filter.
//!
//! Bloom filters have a false positive rate. Each requests uses a different bloom filter
//! with random hash functions. So each subsequent request will have a different distribution
//! of false positives.
use crate::contact_info::ContactInfo;
use crate::crds::Crds;
use crate::crds_gossip::{get_stake, get_weight, CRDS_GOSSIP_BLOOM_SIZE};
use crate::crds_gossip_error::CrdsGossipError;
use crate::crds_value::{CrdsValue, CrdsValueLabel};
use crate::packet::BLOB_DATA_SIZE;
use bincode::serialized_size;
use rand;
use rand::distributions::{Distribution, WeightedIndex};
use solana_runtime::bloom::Bloom;
use solana_sdk::hash::Hash;
use solana_sdk::pubkey::Pubkey;
use std::cmp;
use std::collections::HashMap;
use std::collections::VecDeque;
pub const CRDS_GOSSIP_PULL_CRDS_TIMEOUT_MS: u64 = 15000;
#[derive(Clone)]
pub struct CrdsGossipPull {
/// timestamp of last request
pub pull_request_time: HashMap<Pubkey, u64>,
/// hash and insert time
purged_values: VecDeque<(Hash, u64)>,
/// max bytes per message
pub max_bytes: usize,
pub crds_timeout: u64,
}
impl Default for CrdsGossipPull {
fn default() -> Self {
Self {
purged_values: VecDeque::new(),
pull_request_time: HashMap::new(),
max_bytes: BLOB_DATA_SIZE,
crds_timeout: CRDS_GOSSIP_PULL_CRDS_TIMEOUT_MS,
}
}
}
impl CrdsGossipPull {
/// generate a random request
pub fn new_pull_request(
&self,
crds: &Crds,
self_id: &Pubkey,
now: u64,
stakes: &HashMap<Pubkey, u64>,
) -> Result<(Pubkey, Bloom<Hash>, CrdsValue), CrdsGossipError> {
let options = self.pull_options(crds, &self_id, now, stakes);
if options.is_empty() {
return Err(CrdsGossipError::NoPeers);
}
let filter = self.build_crds_filter(crds);
let index = WeightedIndex::new(options.iter().map(|weighted| weighted.0)).unwrap();
let random = index.sample(&mut rand::thread_rng());
let self_info = crds
.lookup(&CrdsValueLabel::ContactInfo(*self_id))
.unwrap_or_else(|| panic!("self_id invalid {}", self_id));
Ok((options[random].1.id, filter, self_info.clone()))
}
fn pull_options<'a>(
&self,
crds: &'a Crds,
self_id: &Pubkey,
now: u64,
stakes: &HashMap<Pubkey, u64>,
) -> Vec<(f32, &'a ContactInfo)> {
crds.table
.values()
.filter_map(|v| v.value.contact_info())
.filter(|v| v.id != *self_id && ContactInfo::is_valid_address(&v.gossip))
.map(|item| {
let max_weight = f32::from(u16::max_value()) - 1.0;
let req_time: u64 = *self.pull_request_time.get(&item.id).unwrap_or(&0);
let since = ((now - req_time) / 1024) as u32;
let stake = get_stake(&item.id, stakes);
let weight = get_weight(max_weight, since, stake);
(weight, item)
})
.collect()
}
/// time when a request to `from` was initiated
/// This is used for weighted random selection during `new_pull_request`
/// It's important to use the local nodes request creation time as the weight
/// instead of the response received time otherwise failed nodes will increase their weight.
pub fn mark_pull_request_creation_time(&mut self, from: &Pubkey, now: u64) {
self.pull_request_time.insert(*from, now);
}
/// Store an old hash in the purged values set
pub fn record_old_hash(&mut self, hash: Hash, timestamp: u64) {
self.purged_values.push_back((hash, timestamp))
}
/// process a pull request and create a response
pub fn process_pull_request(
&mut self,
crds: &mut Crds,
caller: CrdsValue,
mut filter: Bloom<Hash>,
now: u64,
) -> Vec<CrdsValue> {
let rv = self.filter_crds_values(crds, &mut filter);
let key = caller.label().pubkey();
let old = crds.insert(caller, now);
if let Some(val) = old.ok().and_then(|opt| opt) {
self.purged_values
.push_back((val.value_hash, val.local_timestamp))
}
crds.update_record_timestamp(&key, now);
rv
}
/// process a pull response
pub fn process_pull_response(
&mut self,
crds: &mut Crds,
from: &Pubkey,
response: Vec<CrdsValue>,
now: u64,
) -> usize {
let mut failed = 0;
for r in response {
let owner = r.label().pubkey();
let old = crds.insert(r, now);
failed += old.is_err() as usize;
old.ok().map(|opt| {
crds.update_record_timestamp(&owner, now);
opt.map(|val| {
self.purged_values
.push_back((val.value_hash, val.local_timestamp))
})
});
}
crds.update_record_timestamp(from, now);
failed
}
/// build a filter of the current crds table
pub fn build_crds_filter(&self, crds: &Crds) -> Bloom<Hash> {
let num = cmp::max(
CRDS_GOSSIP_BLOOM_SIZE,
crds.table.values().count() + self.purged_values.len(),
);
let mut bloom = Bloom::random(num, 0.1, 4 * 1024 * 8 - 1);
for v in crds.table.values() {
bloom.add(&v.value_hash);
}
for (value_hash, _insert_timestamp) in &self.purged_values {
bloom.add(value_hash);
}
bloom
}
/// filter values that fail the bloom filter up to max_bytes
fn filter_crds_values(&self, crds: &Crds, filter: &mut Bloom<Hash>) -> Vec<CrdsValue> {
let mut max_bytes = self.max_bytes as isize;
let mut ret = vec![];
for v in crds.table.values() {
if filter.contains(&v.value_hash) {
continue;
}
max_bytes -= serialized_size(&v.value).unwrap() as isize;
if max_bytes < 0 {
break;
}
ret.push(v.value.clone());
}
ret
}
/// Purge values from the crds that are older then `active_timeout`
/// The value_hash of an active item is put into self.purged_values queue
pub fn purge_active(&mut self, crds: &mut Crds, self_id: &Pubkey, min_ts: u64) {
let old = crds.find_old_labels(min_ts);
let mut purged: VecDeque<_> = old
.iter()
.filter(|label| label.pubkey() != *self_id)
.filter_map(|label| {
let rv = crds
.lookup_versioned(label)
.map(|val| (val.value_hash, val.local_timestamp));
crds.remove(label);
rv
})
.collect();
self.purged_values.append(&mut purged);
}
/// Purge values from the `self.purged_values` queue that are older then purge_timeout
pub fn purge_purged(&mut self, min_ts: u64) {
let cnt = self
.purged_values
.iter()
.take_while(|v| v.1 < min_ts)
.count();
self.purged_values.drain(..cnt);
}
}
#[cfg(test)]
mod test {
use super::*;
use crate::contact_info::ContactInfo;
#[test]
fn test_new_pull_with_stakes() {
let mut crds = Crds::default();
let mut stakes = HashMap::new();
let node = CrdsGossipPull::default();
let me = CrdsValue::ContactInfo(ContactInfo::new_localhost(&Pubkey::new_rand(), 0));
crds.insert(me.clone(), 0).unwrap();
for i in 1..=30 {
let entry = CrdsValue::ContactInfo(ContactInfo::new_localhost(&Pubkey::new_rand(), 0));
let id = entry.label().pubkey();
crds.insert(entry.clone(), 0).unwrap();
stakes.insert(id, i * 100);
}
let now = 1024;
let mut options = node.pull_options(&crds, &me.label().pubkey(), now, &stakes);
assert!(!options.is_empty());
options.sort_by(|(weight_l, _), (weight_r, _)| weight_r.partial_cmp(weight_l).unwrap());
// check that the highest stake holder is also the heaviest weighted.
assert_eq!(
*stakes.get(&options.get(0).unwrap().1.id).unwrap(),
3000_u64
);
}
#[test]
fn test_new_pull_request() {
let mut crds = Crds::default();
let entry = CrdsValue::ContactInfo(ContactInfo::new_localhost(&Pubkey::new_rand(), 0));
let id = entry.label().pubkey();
let node = CrdsGossipPull::default();
assert_eq!(
node.new_pull_request(&crds, &id, 0, &HashMap::new()),
Err(CrdsGossipError::NoPeers)
);
crds.insert(entry.clone(), 0).unwrap();
assert_eq!(
node.new_pull_request(&crds, &id, 0, &HashMap::new()),
Err(CrdsGossipError::NoPeers)
);
let new = CrdsValue::ContactInfo(ContactInfo::new_localhost(&Pubkey::new_rand(), 0));
crds.insert(new.clone(), 0).unwrap();
let req = node.new_pull_request(&crds, &id, 0, &HashMap::new());
let (to, _, self_info) = req.unwrap();
assert_eq!(to, new.label().pubkey());
assert_eq!(self_info, entry);
}
#[test]
fn test_new_mark_creation_time() {
let mut crds = Crds::default();
let entry = CrdsValue::ContactInfo(ContactInfo::new_localhost(&Pubkey::new_rand(), 0));
let node_pubkey = entry.label().pubkey();
let mut node = CrdsGossipPull::default();
crds.insert(entry.clone(), 0).unwrap();
let old = CrdsValue::ContactInfo(ContactInfo::new_localhost(&Pubkey::new_rand(), 0));
crds.insert(old.clone(), 0).unwrap();
let new = CrdsValue::ContactInfo(ContactInfo::new_localhost(&Pubkey::new_rand(), 0));
crds.insert(new.clone(), 0).unwrap();
// set request creation time to max_value
node.mark_pull_request_creation_time(&new.label().pubkey(), u64::max_value());
// odds of getting the other request should be 1 in u64::max_value()
for _ in 0..10 {
let req = node.new_pull_request(&crds, &node_pubkey, u64::max_value(), &HashMap::new());
let (to, _, self_info) = req.unwrap();
assert_eq!(to, old.label().pubkey());
assert_eq!(self_info, entry);
}
}
#[test]
fn test_process_pull_request() {
let mut node_crds = Crds::default();
let entry = CrdsValue::ContactInfo(ContactInfo::new_localhost(&Pubkey::new_rand(), 0));
let node_pubkey = entry.label().pubkey();
let node = CrdsGossipPull::default();
node_crds.insert(entry.clone(), 0).unwrap();
let new = CrdsValue::ContactInfo(ContactInfo::new_localhost(&Pubkey::new_rand(), 0));
node_crds.insert(new.clone(), 0).unwrap();
let req = node.new_pull_request(&node_crds, &node_pubkey, 0, &HashMap::new());
let mut dest_crds = Crds::default();
let mut dest = CrdsGossipPull::default();
let (_, filter, caller) = req.unwrap();
let rsp = dest.process_pull_request(&mut dest_crds, caller.clone(), filter, 1);
assert!(rsp.is_empty());
assert!(dest_crds.lookup(&caller.label()).is_some());
assert_eq!(
dest_crds
.lookup_versioned(&caller.label())
.unwrap()
.insert_timestamp,
1
);
assert_eq!(
dest_crds
.lookup_versioned(&caller.label())
.unwrap()
.local_timestamp,
1
);
}
#[test]
fn test_process_pull_request_response() {
let mut node_crds = Crds::default();
let entry = CrdsValue::ContactInfo(ContactInfo::new_localhost(&Pubkey::new_rand(), 0));
let node_pubkey = entry.label().pubkey();
let mut node = CrdsGossipPull::default();
node_crds.insert(entry.clone(), 0).unwrap();
let new = CrdsValue::ContactInfo(ContactInfo::new_localhost(&Pubkey::new_rand(), 0));
node_crds.insert(new.clone(), 0).unwrap();
let mut dest = CrdsGossipPull::default();
let mut dest_crds = Crds::default();
let new_id = Pubkey::new_rand();
let new = CrdsValue::ContactInfo(ContactInfo::new_localhost(&new_id, 1));
dest_crds.insert(new.clone(), 0).unwrap();
// node contains a key from the dest node, but at an older local timestamp
let same_key = CrdsValue::ContactInfo(ContactInfo::new_localhost(&new_id, 0));
assert_eq!(same_key.label(), new.label());
assert!(same_key.wallclock() < new.wallclock());
node_crds.insert(same_key.clone(), 0).unwrap();
assert_eq!(
node_crds
.lookup_versioned(&same_key.label())
.unwrap()
.local_timestamp,
0
);
let mut done = false;
for _ in 0..30 {
// there is a chance of a false positive with bloom filters
let req = node.new_pull_request(&node_crds, &node_pubkey, 0, &HashMap::new());
let (_, filter, caller) = req.unwrap();
let rsp = dest.process_pull_request(&mut dest_crds, caller, filter, 0);
// if there is a false positive this is empty
// prob should be around 0.1 per iteration
if rsp.is_empty() {
continue;
}
assert_eq!(rsp.len(), 1);
let failed = node.process_pull_response(&mut node_crds, &node_pubkey, rsp, 1);
assert_eq!(failed, 0);
assert_eq!(
node_crds
.lookup_versioned(&new.label())
.unwrap()
.local_timestamp,
1
);
// verify that the whole record was updated for dest since this is a response from dest
assert_eq!(
node_crds
.lookup_versioned(&same_key.label())
.unwrap()
.local_timestamp,
1
);
done = true;
break;
}
assert!(done);
}
#[test]
fn test_gossip_purge() {
let mut node_crds = Crds::default();
let entry = CrdsValue::ContactInfo(ContactInfo::new_localhost(&Pubkey::new_rand(), 0));
let node_label = entry.label();
let node_pubkey = node_label.pubkey();
let mut node = CrdsGossipPull::default();
node_crds.insert(entry.clone(), 0).unwrap();
let old = CrdsValue::ContactInfo(ContactInfo::new_localhost(&Pubkey::new_rand(), 0));
node_crds.insert(old.clone(), 0).unwrap();
let value_hash = node_crds.lookup_versioned(&old.label()).unwrap().value_hash;
//verify self is valid
assert_eq!(node_crds.lookup(&node_label).unwrap().label(), node_label);
// purge
node.purge_active(&mut node_crds, &node_pubkey, 1);
//verify self is still valid after purge
assert_eq!(node_crds.lookup(&node_label).unwrap().label(), node_label);
assert_eq!(node_crds.lookup_versioned(&old.label()), None);
assert_eq!(node.purged_values.len(), 1);
for _ in 0..30 {
// there is a chance of a false positive with bloom filters
// assert that purged value is still in the set
// chance of 30 consecutive false positives is 0.1^30
let filter = node.build_crds_filter(&node_crds);
assert!(filter.contains(&value_hash));
}
// purge the value
node.purge_purged(1);
assert_eq!(node.purged_values.len(), 0);
}
}