Remove sync await points (#3036)

## Issue Addressed

Removes the await points in sync waiting for a processor response for rpc block processing. Built on top of #3029 
This also handles a couple of bugs in the previous code and adds a relatively comprehensive test suite.
This commit is contained in:
Divma
2022-03-23 01:09:39 +00:00
parent af50130e21
commit 788b6af3c4
13 changed files with 1722 additions and 727 deletions

View File

@@ -0,0 +1,637 @@
use std::collections::hash_map::Entry;
use std::time::Duration;
use beacon_chain::{BeaconChainTypes, BlockError};
use fnv::FnvHashMap;
use lighthouse_network::{PeerAction, PeerId};
use lru_cache::LRUCache;
use slog::{crit, debug, error, trace, warn, Logger};
use smallvec::SmallVec;
use store::{Hash256, SignedBeaconBlock};
use strum::AsStaticRef;
use tokio::sync::mpsc;
use crate::beacon_processor::{ChainSegmentProcessId, WorkEvent};
use crate::metrics;
use self::{
parent_lookup::{ParentLookup, VerifyError},
single_block_lookup::SingleBlockRequest,
};
use super::BatchProcessResult;
use super::{
manager::{BlockProcessType, Id},
network_context::SyncNetworkContext,
};
mod parent_lookup;
mod single_block_lookup;
#[cfg(test)]
mod tests;
const FAILED_CHAINS_CACHE_SIZE: usize = 500;
const SINGLE_BLOCK_LOOKUP_MAX_ATTEMPTS: u8 = 3;
pub(crate) struct BlockLookups<T: BeaconChainTypes> {
/// A collection of parent block lookups.
parent_queue: SmallVec<[ParentLookup<T::EthSpec>; 3]>,
/// A cache of failed chain lookups to prevent duplicate searches.
failed_chains: LRUCache<Hash256>,
/// A collection of block hashes being searched for and a flag indicating if a result has been
/// received or not.
///
/// The flag allows us to determine if the peer returned data or sent us nothing.
single_block_lookups: FnvHashMap<Id, SingleBlockRequest<SINGLE_BLOCK_LOOKUP_MAX_ATTEMPTS>>,
/// A multi-threaded, non-blocking processor for applying messages to the beacon chain.
beacon_processor_send: mpsc::Sender<WorkEvent<T>>,
/// The logger for the import manager.
log: Logger,
}
impl<T: BeaconChainTypes> BlockLookups<T> {
pub fn new(beacon_processor_send: mpsc::Sender<WorkEvent<T>>, log: Logger) -> Self {
Self {
parent_queue: Default::default(),
failed_chains: LRUCache::new(FAILED_CHAINS_CACHE_SIZE),
single_block_lookups: Default::default(),
beacon_processor_send,
log,
}
}
/* Lookup requests */
pub fn search_block(
&mut self,
hash: Hash256,
peer_id: PeerId,
cx: &mut SyncNetworkContext<T::EthSpec>,
) {
// Do not re-request a block that is already being requested
if self
.single_block_lookups
.values_mut()
.any(|single_block_request| single_block_request.add_peer(&hash, &peer_id))
{
return;
}
debug!(
self.log,
"Searching for block";
"peer_id" => %peer_id,
"block" => %hash
);
let mut single_block_request = SingleBlockRequest::new(hash, peer_id);
let (peer_id, request) = single_block_request.request_block().unwrap();
if let Ok(request_id) = cx.single_block_lookup_request(peer_id, request) {
self.single_block_lookups
.insert(request_id, single_block_request);
metrics::set_gauge(
&metrics::SYNC_SINGLE_BLOCK_LOOKUPS,
self.single_block_lookups.len() as i64,
);
}
}
pub fn search_parent(
&mut self,
block: Box<SignedBeaconBlock<T::EthSpec>>,
peer_id: PeerId,
cx: &mut SyncNetworkContext<T::EthSpec>,
) {
let block_root = block.canonical_root();
let parent_root = block.parent_root();
// If this block or it's parent is part of a known failed chain, ignore it.
if self.failed_chains.contains(&parent_root) || self.failed_chains.contains(&block_root) {
debug!(self.log, "Block is from a past failed chain. Dropping";
"block_root" => ?block_root, "block_slot" => block.slot());
return;
}
// Make sure this block is not already downloaded, and that neither it or its parent is
// being searched for.
if self.parent_queue.iter_mut().any(|parent_req| {
parent_req.contains_block(&block)
|| parent_req.add_peer(&block_root, &peer_id)
|| parent_req.add_peer(&parent_root, &peer_id)
}) {
// we are already searching for this block, ignore it
return;
}
let parent_lookup = ParentLookup::new(*block, peer_id);
self.request_parent(parent_lookup, cx);
}
/* Lookup responses */
pub fn single_block_lookup_response(
&mut self,
id: Id,
peer_id: PeerId,
block: Option<Box<SignedBeaconBlock<T::EthSpec>>>,
seen_timestamp: Duration,
cx: &mut SyncNetworkContext<T::EthSpec>,
) {
let mut request = match self.single_block_lookups.entry(id) {
Entry::Occupied(req) => req,
Entry::Vacant(_) => {
if block.is_some() {
crit!(
self.log,
"Block returned for single block lookup not present"
);
#[cfg(debug_assertions)]
panic!("block returned for single block lookup not present");
}
return;
}
};
match request.get_mut().verify_block(block) {
Ok(Some(block)) => {
// This is the correct block, send it for processing
if self
.send_block_for_processing(
block,
seen_timestamp,
BlockProcessType::SingleBlock { id },
)
.is_err()
{
// Remove to avoid inconsistencies
self.single_block_lookups.remove(&id);
}
}
Ok(None) => {
// request finished correctly, it will be removed after the block is processed.
}
Err(error) => {
let msg: &str = error.as_static();
cx.report_peer(peer_id, PeerAction::LowToleranceError, msg);
// Remove the request, if it can be retried it will be added with a new id.
let mut req = request.remove();
debug!(self.log, "Single block lookup failed";
"peer_id" => %peer_id, "error" => msg, "block_root" => %req.hash);
// try the request again if possible
if let Ok((peer_id, request)) = req.request_block() {
if let Ok(id) = cx.single_block_lookup_request(peer_id, request) {
self.single_block_lookups.insert(id, req);
}
}
}
}
metrics::set_gauge(
&metrics::SYNC_SINGLE_BLOCK_LOOKUPS,
self.single_block_lookups.len() as i64,
);
}
pub fn parent_lookup_response(
&mut self,
id: Id,
peer_id: PeerId,
block: Option<Box<SignedBeaconBlock<T::EthSpec>>>,
seen_timestamp: Duration,
cx: &mut SyncNetworkContext<T::EthSpec>,
) {
let mut parent_lookup = if let Some(pos) = self
.parent_queue
.iter()
.position(|request| request.pending_response(id))
{
self.parent_queue.remove(pos)
} else {
if block.is_some() {
debug!(self.log, "Response for a parent lookup request that was not found"; "peer_id" => %peer_id);
}
return;
};
match parent_lookup.verify_block(block, &self.failed_chains) {
Ok(Some(block)) => {
// Block is correct, send to the beacon processor.
let chain_hash = parent_lookup.chain_hash();
if self
.send_block_for_processing(
block,
seen_timestamp,
BlockProcessType::ParentLookup { chain_hash },
)
.is_ok()
{
self.parent_queue.push(parent_lookup)
}
}
Ok(None) => {
// Request finished successfully, nothing else to do. It will be removed after the
// processing result arrives.
self.parent_queue.push(parent_lookup);
}
Err(e) => match e {
VerifyError::RootMismatch
| VerifyError::NoBlockReturned
| VerifyError::ExtraBlocksReturned => {
let e = e.as_static();
warn!(self.log, "Peer sent invalid response to parent request.";
"peer_id" => %peer_id, "reason" => e);
// We do not tolerate these kinds of errors. We will accept a few but these are signs
// of a faulty peer.
cx.report_peer(peer_id, PeerAction::LowToleranceError, e);
// We try again if possible.
self.request_parent(parent_lookup, cx);
}
VerifyError::PreviousFailure { parent_root } => {
self.failed_chains.insert(parent_lookup.chain_hash());
debug!(
self.log,
"Parent chain ignored due to past failure";
"block" => %parent_root,
);
// Add the root block to failed chains
self.failed_chains.insert(parent_lookup.chain_hash());
cx.report_peer(
peer_id,
PeerAction::MidToleranceError,
"bbroot_failed_chains",
);
}
},
};
metrics::set_gauge(
&metrics::SYNC_PARENT_BLOCK_LOOKUPS,
self.parent_queue.len() as i64,
);
}
/* Error responses */
#[allow(clippy::needless_collect)] // false positive
pub fn peer_disconnected(&mut self, peer_id: &PeerId, cx: &mut SyncNetworkContext<T::EthSpec>) {
/* Check disconnection for single block lookups */
// better written after https://github.com/rust-lang/rust/issues/59618
let remove_retry_ids: Vec<Id> = self
.single_block_lookups
.iter_mut()
.filter_map(|(id, req)| {
if req.check_peer_disconnected(peer_id).is_err() {
Some(*id)
} else {
None
}
})
.collect();
for mut req in remove_retry_ids
.into_iter()
.map(|id| self.single_block_lookups.remove(&id).unwrap())
.collect::<Vec<_>>()
{
// retry the request
match req.request_block() {
Ok((peer_id, block_request)) => {
if let Ok(request_id) = cx.single_block_lookup_request(peer_id, block_request) {
self.single_block_lookups.insert(request_id, req);
}
}
Err(e) => {
trace!(self.log, "Single block request failed on peer disconnection";
"block_root" => %req.hash, "peer_id" => %peer_id, "reason" => e.as_static());
}
}
}
/* Check disconnection for parent lookups */
while let Some(pos) = self
.parent_queue
.iter_mut()
.position(|req| req.check_peer_disconnected(peer_id).is_err())
{
let parent_lookup = self.parent_queue.remove(pos);
trace!(self.log, "Parent lookup's peer disconnected"; &parent_lookup);
self.request_parent(parent_lookup, cx);
}
}
pub fn parent_lookup_failed(
&mut self,
id: Id,
peer_id: PeerId,
cx: &mut SyncNetworkContext<T::EthSpec>,
) {
if let Some(pos) = self
.parent_queue
.iter()
.position(|request| request.pending_response(id))
{
let mut parent_lookup = self.parent_queue.remove(pos);
parent_lookup.download_failed();
trace!(self.log, "Parent lookup request failed"; &parent_lookup);
self.request_parent(parent_lookup, cx);
} else {
return debug!(self.log, "RPC failure for a parent lookup request that was not found"; "peer_id" => %peer_id);
};
metrics::set_gauge(
&metrics::SYNC_PARENT_BLOCK_LOOKUPS,
self.parent_queue.len() as i64,
);
}
pub fn single_block_lookup_failed(&mut self, id: Id, cx: &mut SyncNetworkContext<T::EthSpec>) {
if let Some(mut request) = self.single_block_lookups.remove(&id) {
request.register_failure();
trace!(self.log, "Single block lookup failed"; "block" => %request.hash);
if let Ok((peer_id, block_request)) = request.request_block() {
if let Ok(request_id) = cx.single_block_lookup_request(peer_id, block_request) {
self.single_block_lookups.insert(request_id, request);
}
}
}
metrics::set_gauge(
&metrics::SYNC_SINGLE_BLOCK_LOOKUPS,
self.single_block_lookups.len() as i64,
);
}
/* Processing responses */
pub fn single_block_processed(
&mut self,
id: Id,
result: Result<(), BlockError<T::EthSpec>>,
cx: &mut SyncNetworkContext<T::EthSpec>,
) {
let mut req = match self.single_block_lookups.remove(&id) {
Some(req) => req,
None => {
#[cfg(debug_assertions)]
panic!("block processed for single block lookup not present");
#[cfg(not(debug_assertions))]
return crit!(
self.log,
"Block processed for single block lookup not present"
);
}
};
let root = req.hash;
let peer_id = match req.processing_peer() {
Ok(peer) => peer,
Err(_) => return,
};
if let Err(e) = &result {
trace!(self.log, "Single block processing failed"; "block" => %root, "error" => %e);
} else {
trace!(self.log, "Single block processing succeeded"; "block" => %root);
}
match result {
Err(e) => match e {
BlockError::BlockIsAlreadyKnown => {
// No error here
}
BlockError::BeaconChainError(e) => {
// Internal error
error!(self.log, "Beacon chain error processing single block"; "block_root" => %root, "error" => ?e);
}
BlockError::ParentUnknown(block) => {
self.search_parent(block, peer_id, cx);
}
other => {
warn!(self.log, "Peer sent invalid block in single block lookup"; "root" => %root, "error" => ?other, "peer_id" => %peer_id);
cx.report_peer(
peer_id,
PeerAction::MidToleranceError,
"single_block_failure",
);
// Try it again if possible.
req.register_failure();
if let Ok((peer_id, request)) = req.request_block() {
if let Ok(request_id) = cx.single_block_lookup_request(peer_id, request) {
// insert with the new id
self.single_block_lookups.insert(request_id, req);
}
}
}
},
Ok(()) => {
// No error here
}
}
metrics::set_gauge(
&metrics::SYNC_SINGLE_BLOCK_LOOKUPS,
self.single_block_lookups.len() as i64,
);
}
pub fn parent_block_processed(
&mut self,
chain_hash: Hash256,
result: Result<(), BlockError<T::EthSpec>>,
cx: &mut SyncNetworkContext<T::EthSpec>,
) {
let (mut parent_lookup, peer_id) = if let Some((pos, peer)) = self
.parent_queue
.iter()
.enumerate()
.find_map(|(pos, request)| {
request
.get_processing_peer(chain_hash)
.map(|peer| (pos, peer))
}) {
(self.parent_queue.remove(pos), peer)
} else {
#[cfg(debug_assertions)]
panic!(
"Process response for a parent lookup request that was not found. Chain_hash: {}",
chain_hash
);
#[cfg(not(debug_assertions))]
return crit!(self.log, "Process response for a parent lookup request that was not found"; "chain_hash" => %chain_hash);
};
if let Err(e) = &result {
trace!(self.log, "Parent block processing failed"; &parent_lookup, "error" => %e);
} else {
trace!(self.log, "Parent block processing succeeded"; &parent_lookup);
}
match result {
Err(BlockError::ParentUnknown(block)) => {
// need to keep looking for parents
// add the block back to the queue and continue the search
parent_lookup.add_block(*block);
self.request_parent(parent_lookup, cx);
}
Ok(_) | Err(BlockError::BlockIsAlreadyKnown { .. }) => {
let chain_hash = parent_lookup.chain_hash();
let blocks = parent_lookup.chain_blocks();
let process_id = ChainSegmentProcessId::ParentLookup(chain_hash);
match self
.beacon_processor_send
.try_send(WorkEvent::chain_segment(process_id, blocks))
{
Ok(_) => {
self.parent_queue.push(parent_lookup);
}
Err(e) => {
error!(
self.log,
"Failed to send chain segment to processor";
"error" => ?e
);
}
}
}
Err(outcome) => {
// all else we consider the chain a failure and downvote the peer that sent
// us the last block
warn!(
self.log, "Invalid parent chain";
"score_adjustment" => %PeerAction::MidToleranceError,
"outcome" => ?outcome,
"last_peer" => %peer_id,
);
// Add this chain to cache of failed chains
self.failed_chains.insert(chain_hash);
// This currently can be a host of errors. We permit this due to the partial
// ambiguity.
cx.report_peer(peer_id, PeerAction::MidToleranceError, "parent_request_err");
}
}
metrics::set_gauge(
&metrics::SYNC_PARENT_BLOCK_LOOKUPS,
self.parent_queue.len() as i64,
);
}
pub fn parent_chain_processed(
&mut self,
chain_hash: Hash256,
result: BatchProcessResult,
cx: &mut SyncNetworkContext<T::EthSpec>,
) {
let parent_lookup = if let Some(pos) = self
.parent_queue
.iter()
.position(|request| request.chain_hash() == chain_hash)
{
self.parent_queue.remove(pos)
} else {
#[cfg(debug_assertions)]
panic!(
"Chain process response for a parent lookup request that was not found. Chain_hash: {}",
chain_hash
);
#[cfg(not(debug_assertions))]
return crit!(self.log, "Chain process response for a parent lookup request that was not found"; "chain_hash" => %chain_hash);
};
debug!(self.log, "Parent chain processed"; "chain_hash" => %chain_hash, "result" => ?result);
match result {
BatchProcessResult::Success(_) => {
// nothing to do.
}
BatchProcessResult::Failed {
imported_blocks: _,
peer_action,
} => {
self.failed_chains.insert(parent_lookup.chain_hash());
if let Some(peer_action) = peer_action {
for &peer_id in parent_lookup.used_peers() {
cx.report_peer(peer_id, peer_action, "parent_chain_failure")
}
}
}
}
metrics::set_gauge(
&metrics::SYNC_PARENT_BLOCK_LOOKUPS,
self.parent_queue.len() as i64,
);
}
/* Helper functions */
fn send_block_for_processing(
&mut self,
block: Box<SignedBeaconBlock<T::EthSpec>>,
duration: Duration,
process_type: BlockProcessType,
) -> Result<(), ()> {
trace!(self.log, "Sending block for processing"; "block" => %block.canonical_root(), "process" => ?process_type);
let event = WorkEvent::rpc_beacon_block(block, duration, process_type);
if let Err(e) = self.beacon_processor_send.try_send(event) {
error!(
self.log,
"Failed to send sync block to processor";
"error" => ?e
);
return Err(());
}
Ok(())
}
fn request_parent(
&mut self,
mut parent_lookup: ParentLookup<T::EthSpec>,
cx: &mut SyncNetworkContext<T::EthSpec>,
) {
match parent_lookup.request_parent(cx) {
Err(e) => {
debug!(self.log, "Failed to request parent"; &parent_lookup, "error" => e.as_static());
match e {
parent_lookup::RequestError::SendFailed(_) => {
// Probably shutting down, nothing to do here. Drop the request
}
parent_lookup::RequestError::ChainTooLong
| parent_lookup::RequestError::TooManyAttempts => {
self.failed_chains.insert(parent_lookup.chain_hash());
// This indicates faulty peers.
for &peer_id in parent_lookup.used_peers() {
cx.report_peer(peer_id, PeerAction::LowToleranceError, e.as_static())
}
}
parent_lookup::RequestError::NoPeers => {
// This happens if the peer disconnects while the block is being
// processed. Drop the request without extra penalty
}
}
}
Ok(_) => {
debug!(self.log, "Requesting parent"; &parent_lookup);
self.parent_queue.push(parent_lookup)
}
}
// We remove and add back again requests so we want this updated regardless of outcome.
metrics::set_gauge(
&metrics::SYNC_PARENT_BLOCK_LOOKUPS,
self.parent_queue.len() as i64,
);
}
}

View File

@@ -0,0 +1,201 @@
use lighthouse_network::PeerId;
use store::{EthSpec, Hash256, SignedBeaconBlock};
use strum::AsStaticStr;
use crate::sync::{
manager::{Id, SLOT_IMPORT_TOLERANCE},
network_context::SyncNetworkContext,
};
use super::single_block_lookup::{self, SingleBlockRequest};
/// How many attempts we try to find a parent of a block before we give up trying .
pub(crate) const PARENT_FAIL_TOLERANCE: u8 = 5;
/// The maximum depth we will search for a parent block. In principle we should have sync'd any
/// canonical chain to its head once the peer connects. A chain should not appear where it's depth
/// is further back than the most recent head slot.
pub(crate) const PARENT_DEPTH_TOLERANCE: usize = SLOT_IMPORT_TOLERANCE * 2;
/// Maintains a sequential list of parents to lookup and the lookup's current state.
pub(crate) struct ParentLookup<T: EthSpec> {
/// The root of the block triggering this parent request.
chain_hash: Hash256,
/// The blocks that have currently been downloaded.
downloaded_blocks: Vec<SignedBeaconBlock<T>>,
/// Request of the last parent.
current_parent_request: SingleBlockRequest<PARENT_FAIL_TOLERANCE>,
/// Id of the last parent request.
current_parent_request_id: Option<Id>,
}
#[derive(Debug, PartialEq, Eq, AsStaticStr)]
pub enum VerifyError {
RootMismatch,
NoBlockReturned,
ExtraBlocksReturned,
PreviousFailure { parent_root: Hash256 },
}
#[derive(Debug, PartialEq, Eq)]
pub enum RequestError {
SendFailed(&'static str),
ChainTooLong,
TooManyAttempts,
NoPeers,
}
impl<T: EthSpec> ParentLookup<T> {
pub fn contains_block(&self, block: &SignedBeaconBlock<T>) -> bool {
self.downloaded_blocks
.iter()
.any(|d_block| d_block == block)
}
pub fn new(block: SignedBeaconBlock<T>, peer_id: PeerId) -> Self {
let current_parent_request = SingleBlockRequest::new(block.parent_root(), peer_id);
Self {
chain_hash: block.canonical_root(),
downloaded_blocks: vec![block],
current_parent_request,
current_parent_request_id: None,
}
}
/// Attempts to request the next unknown parent. If the request fails, it should be removed.
pub fn request_parent(&mut self, cx: &mut SyncNetworkContext<T>) -> Result<(), RequestError> {
// check to make sure this request hasn't failed
if self.downloaded_blocks.len() >= PARENT_DEPTH_TOLERANCE {
return Err(RequestError::ChainTooLong);
}
let (peer_id, request) = self.current_parent_request.request_block()?;
match cx.parent_lookup_request(peer_id, request) {
Ok(request_id) => {
self.current_parent_request_id = Some(request_id);
Ok(())
}
Err(reason) => {
self.current_parent_request_id = None;
Err(RequestError::SendFailed(reason))
}
}
}
pub fn check_peer_disconnected(&mut self, peer_id: &PeerId) -> Result<(), ()> {
self.current_parent_request.check_peer_disconnected(peer_id)
}
pub fn add_block(&mut self, block: SignedBeaconBlock<T>) {
let next_parent = block.parent_root();
self.downloaded_blocks.push(block);
self.current_parent_request.hash = next_parent;
self.current_parent_request.state = single_block_lookup::State::AwaitingDownload;
self.current_parent_request_id = None;
}
pub fn pending_response(&self, req_id: Id) -> bool {
self.current_parent_request_id == Some(req_id)
}
/// Get the parent lookup's chain hash.
pub fn chain_hash(&self) -> Hash256 {
self.chain_hash
}
pub fn download_failed(&mut self) {
self.current_parent_request.register_failure();
self.current_parent_request_id = None;
}
pub fn chain_blocks(&mut self) -> Vec<SignedBeaconBlock<T>> {
std::mem::take(&mut self.downloaded_blocks)
}
/// Verifies that the received block is what we requested. If so, parent lookup now waits for
/// the processing result of the block.
pub fn verify_block(
&mut self,
block: Option<Box<SignedBeaconBlock<T>>>,
failed_chains: &lru_cache::LRUCache<Hash256>,
) -> Result<Option<Box<SignedBeaconBlock<T>>>, VerifyError> {
let block = self.current_parent_request.verify_block(block)?;
// check if the parent of this block isn't in the failed cache. If it is, this chain should
// be dropped and the peer downscored.
if let Some(parent_root) = block.as_ref().map(|block| block.parent_root()) {
if failed_chains.contains(&parent_root) {
self.current_parent_request.register_failure();
self.current_parent_request_id = None;
return Err(VerifyError::PreviousFailure { parent_root });
}
}
Ok(block)
}
pub fn get_processing_peer(&self, chain_hash: Hash256) -> Option<PeerId> {
if self.chain_hash == chain_hash {
return self.current_parent_request.processing_peer().ok();
}
None
}
#[cfg(test)]
pub fn failed_attempts(&self) -> u8 {
self.current_parent_request.failed_attempts
}
pub fn add_peer(&mut self, block_root: &Hash256, peer_id: &PeerId) -> bool {
self.current_parent_request.add_peer(block_root, peer_id)
}
pub fn used_peers(&self) -> impl Iterator<Item = &PeerId> + '_ {
self.current_parent_request.used_peers.iter()
}
}
impl From<super::single_block_lookup::VerifyError> for VerifyError {
fn from(e: super::single_block_lookup::VerifyError) -> Self {
use super::single_block_lookup::VerifyError as E;
match e {
E::RootMismatch => VerifyError::RootMismatch,
E::NoBlockReturned => VerifyError::NoBlockReturned,
E::ExtraBlocksReturned => VerifyError::ExtraBlocksReturned,
}
}
}
impl From<super::single_block_lookup::LookupRequestError> for RequestError {
fn from(e: super::single_block_lookup::LookupRequestError) -> Self {
use super::single_block_lookup::LookupRequestError as E;
match e {
E::TooManyAttempts => RequestError::TooManyAttempts,
E::NoPeers => RequestError::NoPeers,
}
}
}
impl<T: EthSpec> slog::KV for ParentLookup<T> {
fn serialize(
&self,
record: &slog::Record,
serializer: &mut dyn slog::Serializer,
) -> slog::Result {
serializer.emit_arguments("chain_hash", &format_args!("{}", self.chain_hash))?;
slog::Value::serialize(&self.current_parent_request, record, "parent", serializer)?;
serializer.emit_usize("downloaded_blocks", self.downloaded_blocks.len())?;
slog::Result::Ok(())
}
}
impl RequestError {
pub fn as_static(&self) -> &'static str {
match self {
RequestError::SendFailed(e) => e,
RequestError::ChainTooLong => "chain_too_long",
RequestError::TooManyAttempts => "too_many_attempts",
RequestError::NoPeers => "no_peers",
}
}
}

View File

@@ -0,0 +1,209 @@
use std::collections::HashSet;
use lighthouse_network::{rpc::BlocksByRootRequest, PeerId};
use rand::seq::IteratorRandom;
use ssz_types::VariableList;
use store::{EthSpec, Hash256, SignedBeaconBlock};
use strum::AsStaticStr;
/// Object representing a single block lookup request.
#[derive(PartialEq, Eq)]
pub struct SingleBlockRequest<const MAX_ATTEMPTS: u8> {
/// The hash of the requested block.
pub hash: Hash256,
/// State of this request.
pub state: State,
/// Peers that should have this block.
pub available_peers: HashSet<PeerId>,
/// Peers from which we have requested this block.
pub used_peers: HashSet<PeerId>,
/// How many times have we attempted this block.
pub failed_attempts: u8,
}
#[derive(Debug, PartialEq, Eq)]
pub enum State {
AwaitingDownload,
Downloading { peer_id: PeerId },
Processing { peer_id: PeerId },
}
#[derive(Debug, PartialEq, Eq, AsStaticStr)]
pub enum VerifyError {
RootMismatch,
NoBlockReturned,
ExtraBlocksReturned,
}
#[derive(Debug, PartialEq, Eq, AsStaticStr)]
pub enum LookupRequestError {
TooManyAttempts,
NoPeers,
}
impl<const MAX_ATTEMPTS: u8> SingleBlockRequest<MAX_ATTEMPTS> {
pub fn new(hash: Hash256, peer_id: PeerId) -> Self {
Self {
hash,
state: State::AwaitingDownload,
available_peers: HashSet::from([peer_id]),
used_peers: HashSet::default(),
failed_attempts: 0,
}
}
pub fn register_failure(&mut self) {
self.failed_attempts += 1;
self.state = State::AwaitingDownload;
}
pub fn add_peer(&mut self, hash: &Hash256, peer_id: &PeerId) -> bool {
let is_useful = &self.hash == hash;
if is_useful {
self.available_peers.insert(*peer_id);
}
is_useful
}
/// If a peer disconnects, this request could be failed. If so, an error is returned
pub fn check_peer_disconnected(&mut self, dc_peer_id: &PeerId) -> Result<(), ()> {
self.available_peers.remove(dc_peer_id);
if let State::Downloading { peer_id } = &self.state {
if peer_id == dc_peer_id {
// Peer disconnected before providing a block
self.register_failure();
return Err(());
}
}
Ok(())
}
/// Verifies if the received block matches the requested one.
/// Returns the block for processing if the response is what we expected.
pub fn verify_block<T: EthSpec>(
&mut self,
block: Option<Box<SignedBeaconBlock<T>>>,
) -> Result<Option<Box<SignedBeaconBlock<T>>>, VerifyError> {
match self.state {
State::AwaitingDownload => {
self.register_failure();
Err(VerifyError::ExtraBlocksReturned)
}
State::Downloading { peer_id } => match block {
Some(block) => {
if block.canonical_root() != self.hash {
// return an error and drop the block
self.register_failure();
Err(VerifyError::RootMismatch)
} else {
// Return the block for processing.
self.state = State::Processing { peer_id };
Ok(Some(block))
}
}
None => {
self.register_failure();
Err(VerifyError::NoBlockReturned)
}
},
State::Processing { peer_id: _ } => match block {
Some(_) => {
// We sent the block for processing and received an extra block.
self.register_failure();
Err(VerifyError::ExtraBlocksReturned)
}
None => {
// This is simply the stream termination and we are already processing the
// block
Ok(None)
}
},
}
}
pub fn request_block(&mut self) -> Result<(PeerId, BlocksByRootRequest), LookupRequestError> {
debug_assert!(matches!(self.state, State::AwaitingDownload));
if self.failed_attempts <= MAX_ATTEMPTS {
if let Some(&peer_id) = self.available_peers.iter().choose(&mut rand::thread_rng()) {
let request = BlocksByRootRequest {
block_roots: VariableList::from(vec![self.hash]),
};
self.state = State::Downloading { peer_id };
self.used_peers.insert(peer_id);
Ok((peer_id, request))
} else {
Err(LookupRequestError::NoPeers)
}
} else {
Err(LookupRequestError::TooManyAttempts)
}
}
pub fn processing_peer(&self) -> Result<PeerId, ()> {
if let State::Processing { peer_id } = &self.state {
Ok(*peer_id)
} else {
Err(())
}
}
}
impl<const MAX_ATTEMPTS: u8> slog::Value for SingleBlockRequest<MAX_ATTEMPTS> {
fn serialize(
&self,
record: &slog::Record,
key: slog::Key,
serializer: &mut dyn slog::Serializer,
) -> slog::Result {
serializer.emit_str("request", key)?;
serializer.emit_arguments("hash", &format_args!("{}", self.hash))?;
match &self.state {
State::AwaitingDownload => {
"awaiting_download".serialize(record, "state", serializer)?
}
State::Downloading { peer_id } => {
serializer.emit_arguments("downloading_peer", &format_args!("{}", peer_id))?
}
State::Processing { peer_id } => {
serializer.emit_arguments("processing_peer", &format_args!("{}", peer_id))?
}
}
slog::Result::Ok(())
}
}
#[cfg(test)]
mod tests {
use super::*;
use types::test_utils::{SeedableRng, TestRandom, XorShiftRng};
use types::MinimalEthSpec as E;
fn rand_block() -> SignedBeaconBlock<E> {
let mut rng = XorShiftRng::from_seed([42; 16]);
SignedBeaconBlock::from_block(
types::BeaconBlock::Base(types::BeaconBlockBase {
..<_>::random_for_test(&mut rng)
}),
types::Signature::random_for_test(&mut rng),
)
}
#[test]
fn test_happy_path() {
let peer_id = PeerId::random();
let block = rand_block();
let mut sl = SingleBlockRequest::<4>::new(block.canonical_root(), peer_id);
sl.request_block().unwrap();
sl.verify_block(Some(Box::new(block))).unwrap().unwrap();
}
#[test]
fn test_max_attempts() {
let peer_id = PeerId::random();
let block = rand_block();
let mut sl = SingleBlockRequest::<4>::new(block.canonical_root(), peer_id);
sl.register_failure();
}
}

View File

@@ -0,0 +1,460 @@
use std::sync::Arc;
use crate::service::RequestId;
use crate::sync::manager::RequestId as SyncId;
use crate::NetworkMessage;
use super::*;
use beacon_chain::builder::Witness;
use beacon_chain::eth1_chain::CachingEth1Backend;
use lighthouse_network::{NetworkGlobals, Request};
use slog::{Drain, Level};
use slot_clock::SystemTimeSlotClock;
use store::MemoryStore;
use types::test_utils::{SeedableRng, TestRandom, XorShiftRng};
use types::MinimalEthSpec as E;
type T = Witness<SystemTimeSlotClock, CachingEth1Backend<E>, E, MemoryStore<E>, MemoryStore<E>>;
struct TestRig {
beacon_processor_rx: mpsc::Receiver<WorkEvent<T>>,
network_rx: mpsc::UnboundedReceiver<NetworkMessage<E>>,
rng: XorShiftRng,
}
const D: Duration = Duration::new(0, 0);
impl TestRig {
fn test_setup(log_level: Option<Level>) -> (BlockLookups<T>, SyncNetworkContext<E>, Self) {
let log = {
let decorator = slog_term::TermDecorator::new().build();
let drain = slog_term::FullFormat::new(decorator).build().fuse();
let drain = slog_async::Async::new(drain).build().fuse();
if let Some(log_level) = log_level {
slog::Logger::root(drain.filter_level(log_level).fuse(), slog::o!())
} else {
slog::Logger::root(drain.filter(|_| false).fuse(), slog::o!())
}
};
let (beacon_processor_tx, beacon_processor_rx) = mpsc::channel(100);
let (network_tx, network_rx) = mpsc::unbounded_channel();
let rng = XorShiftRng::from_seed([42; 16]);
let rig = TestRig {
beacon_processor_rx,
network_rx,
rng,
};
let bl = BlockLookups::new(
beacon_processor_tx,
log.new(slog::o!("component" => "block_lookups")),
);
let cx = {
let globals = Arc::new(NetworkGlobals::new_test_globals(&log));
SyncNetworkContext::new(
network_tx,
globals,
log.new(slog::o!("component" => "network_context")),
)
};
(bl, cx, rig)
}
fn rand_block(&mut self) -> SignedBeaconBlock<E> {
SignedBeaconBlock::from_block(
types::BeaconBlock::Base(types::BeaconBlockBase {
..<_>::random_for_test(&mut self.rng)
}),
types::Signature::random_for_test(&mut self.rng),
)
}
#[track_caller]
fn expect_block_request(&mut self) -> Id {
match self.network_rx.try_recv() {
Ok(NetworkMessage::SendRequest {
peer_id: _,
request: Request::BlocksByRoot(_request),
request_id: RequestId::Sync(SyncId::SingleBlock { id }),
}) => id,
other => {
panic!("Expected block request, found {:?}", other);
}
}
}
#[track_caller]
fn expect_parent_request(&mut self) -> Id {
match self.network_rx.try_recv() {
Ok(NetworkMessage::SendRequest {
peer_id: _,
request: Request::BlocksByRoot(_request),
request_id: RequestId::Sync(SyncId::ParentLookup { id }),
}) => id,
other => panic!("Expected parent request, found {:?}", other),
}
}
#[track_caller]
fn expect_block_process(&mut self) {
match self.beacon_processor_rx.try_recv() {
Ok(work) => {
assert_eq!(work.work_type(), crate::beacon_processor::RPC_BLOCK);
}
other => panic!("Expected block process, found {:?}", other),
}
}
#[track_caller]
fn expect_parent_chain_process(&mut self) {
match self.beacon_processor_rx.try_recv() {
Ok(work) => {
assert_eq!(work.work_type(), crate::beacon_processor::CHAIN_SEGMENT);
}
other => panic!("Expected chain segment process, found {:?}", other),
}
}
#[track_caller]
fn expect_empty_network(&mut self) {
assert_eq!(
self.network_rx.try_recv().expect_err("must err"),
mpsc::error::TryRecvError::Empty
);
}
#[track_caller]
pub fn expect_penalty(&mut self) {
match self.network_rx.try_recv() {
Ok(NetworkMessage::ReportPeer { .. }) => {}
other => panic!("Expected peer penalty, found {:?}", other),
}
}
pub fn block_with_parent(&mut self, parent_root: Hash256) -> SignedBeaconBlock<E> {
SignedBeaconBlock::from_block(
types::BeaconBlock::Base(types::BeaconBlockBase {
parent_root,
..<_>::random_for_test(&mut self.rng)
}),
types::Signature::random_for_test(&mut self.rng),
)
}
}
#[test]
fn test_single_block_lookup_happy_path() {
let (mut bl, mut cx, mut rig) = TestRig::test_setup(None);
let block = rig.rand_block();
let peer_id = PeerId::random();
// Trigger the request
bl.search_block(block.canonical_root(), peer_id, &mut cx);
let id = rig.expect_block_request();
// The peer provides the correct block, should not be penalized. Now the block should be sent
// for processing.
bl.single_block_lookup_response(id, peer_id, Some(Box::new(block)), D, &mut cx);
rig.expect_empty_network();
rig.expect_block_process();
// The request should still be active.
assert_eq!(bl.single_block_lookups.len(), 1);
// Send the stream termination. Peer should have not been penalized, and the request removed
// after processing.
bl.single_block_lookup_response(id, peer_id, None, D, &mut cx);
bl.single_block_processed(id, Ok(()), &mut cx);
rig.expect_empty_network();
assert_eq!(bl.single_block_lookups.len(), 0);
}
#[test]
fn test_single_block_lookup_empty_response() {
let (mut bl, mut cx, mut rig) = TestRig::test_setup(None);
let block_hash = Hash256::random();
let peer_id = PeerId::random();
// Trigger the request
bl.search_block(block_hash, peer_id, &mut cx);
let id = rig.expect_block_request();
// The peer does not have the block. It should be penalized.
bl.single_block_lookup_response(id, peer_id, None, D, &mut cx);
rig.expect_penalty();
rig.expect_block_request(); // it should be retried
}
#[test]
fn test_single_block_lookup_wrong_response() {
let (mut bl, mut cx, mut rig) = TestRig::test_setup(None);
let block_hash = Hash256::random();
let peer_id = PeerId::random();
// Trigger the request
bl.search_block(block_hash, peer_id, &mut cx);
let id = rig.expect_block_request();
// Peer sends something else. It should be penalized.
let bad_block = rig.rand_block();
bl.single_block_lookup_response(id, peer_id, Some(Box::new(bad_block)), D, &mut cx);
rig.expect_penalty();
rig.expect_block_request(); // should be retried
// Send the stream termination. This should not produce an additional penalty.
bl.single_block_lookup_response(id, peer_id, None, D, &mut cx);
rig.expect_empty_network();
}
#[test]
fn test_single_block_lookup_failure() {
let (mut bl, mut cx, mut rig) = TestRig::test_setup(None);
let block_hash = Hash256::random();
let peer_id = PeerId::random();
// Trigger the request
bl.search_block(block_hash, peer_id, &mut cx);
let id = rig.expect_block_request();
// The request fails. RPC failures are handled elsewhere so we should not penalize the peer.
bl.single_block_lookup_failed(id, &mut cx);
rig.expect_block_request();
rig.expect_empty_network();
}
#[test]
fn test_single_block_lookup_becomes_parent_request() {
let (mut bl, mut cx, mut rig) = TestRig::test_setup(None);
let block = rig.rand_block();
let peer_id = PeerId::random();
// Trigger the request
bl.search_block(block.canonical_root(), peer_id, &mut cx);
let id = rig.expect_block_request();
// The peer provides the correct block, should not be penalized. Now the block should be sent
// for processing.
bl.single_block_lookup_response(id, peer_id, Some(Box::new(block.clone())), D, &mut cx);
rig.expect_empty_network();
rig.expect_block_process();
// The request should still be active.
assert_eq!(bl.single_block_lookups.len(), 1);
// Send the stream termination. Peer should have not been penalized, and the request moved to a
// parent request after processing.
bl.single_block_processed(id, Err(BlockError::ParentUnknown(Box::new(block))), &mut cx);
assert_eq!(bl.single_block_lookups.len(), 0);
rig.expect_parent_request();
rig.expect_empty_network();
assert_eq!(bl.parent_queue.len(), 1);
}
#[test]
fn test_parent_lookup_happy_path() {
let (mut bl, mut cx, mut rig) = TestRig::test_setup(None);
let parent = rig.rand_block();
let block = rig.block_with_parent(parent.canonical_root());
let chain_hash = block.canonical_root();
let peer_id = PeerId::random();
// Trigger the request
bl.search_parent(Box::new(block), peer_id, &mut cx);
let id = rig.expect_parent_request();
// Peer sends the right block, it should be sent for processing. Peer should not be penalized.
bl.parent_lookup_response(id, peer_id, Some(Box::new(parent)), D, &mut cx);
rig.expect_block_process();
rig.expect_empty_network();
// Processing succeeds, now the rest of the chain should be sent for processing.
bl.parent_block_processed(chain_hash, Err(BlockError::BlockIsAlreadyKnown), &mut cx);
rig.expect_parent_chain_process();
bl.parent_chain_processed(chain_hash, BatchProcessResult::Success(true), &mut cx);
assert_eq!(bl.parent_queue.len(), 0);
}
#[test]
fn test_parent_lookup_wrong_response() {
let (mut bl, mut cx, mut rig) = TestRig::test_setup(None);
let parent = rig.rand_block();
let block = rig.block_with_parent(parent.canonical_root());
let chain_hash = block.canonical_root();
let peer_id = PeerId::random();
// Trigger the request
bl.search_parent(Box::new(block), peer_id, &mut cx);
let id1 = rig.expect_parent_request();
// Peer sends the wrong block, peer should be penalized and the block re-requested.
let bad_block = rig.rand_block();
bl.parent_lookup_response(id1, peer_id, Some(Box::new(bad_block)), D, &mut cx);
rig.expect_penalty();
let id2 = rig.expect_parent_request();
// Send the stream termination for the first request. This should not produce extra penalties.
bl.parent_lookup_response(id1, peer_id, None, D, &mut cx);
rig.expect_empty_network();
// Send the right block this time.
bl.parent_lookup_response(id2, peer_id, Some(Box::new(parent)), D, &mut cx);
rig.expect_block_process();
// Processing succeeds, now the rest of the chain should be sent for processing.
bl.parent_block_processed(chain_hash, Ok(()), &mut cx);
rig.expect_parent_chain_process();
bl.parent_chain_processed(chain_hash, BatchProcessResult::Success(true), &mut cx);
assert_eq!(bl.parent_queue.len(), 0);
}
#[test]
fn test_parent_lookup_empty_response() {
let (mut bl, mut cx, mut rig) = TestRig::test_setup(None);
let parent = rig.rand_block();
let block = rig.block_with_parent(parent.canonical_root());
let chain_hash = block.canonical_root();
let peer_id = PeerId::random();
// Trigger the request
bl.search_parent(Box::new(block), peer_id, &mut cx);
let id1 = rig.expect_parent_request();
// Peer sends an empty response, peer should be penalized and the block re-requested.
bl.parent_lookup_response(id1, peer_id, None, D, &mut cx);
rig.expect_penalty();
let id2 = rig.expect_parent_request();
// Send the right block this time.
bl.parent_lookup_response(id2, peer_id, Some(Box::new(parent)), D, &mut cx);
rig.expect_block_process();
// Processing succeeds, now the rest of the chain should be sent for processing.
bl.parent_block_processed(chain_hash, Ok(()), &mut cx);
rig.expect_parent_chain_process();
bl.parent_chain_processed(chain_hash, BatchProcessResult::Success(true), &mut cx);
assert_eq!(bl.parent_queue.len(), 0);
}
#[test]
fn test_parent_lookup_rpc_failure() {
let (mut bl, mut cx, mut rig) = TestRig::test_setup(None);
let parent = rig.rand_block();
let block = rig.block_with_parent(parent.canonical_root());
let chain_hash = block.canonical_root();
let peer_id = PeerId::random();
// Trigger the request
bl.search_parent(Box::new(block), peer_id, &mut cx);
let id1 = rig.expect_parent_request();
// The request fails. It should be tried again.
bl.parent_lookup_failed(id1, peer_id, &mut cx);
let id2 = rig.expect_parent_request();
// Send the right block this time.
bl.parent_lookup_response(id2, peer_id, Some(Box::new(parent)), D, &mut cx);
rig.expect_block_process();
// Processing succeeds, now the rest of the chain should be sent for processing.
bl.parent_block_processed(chain_hash, Ok(()), &mut cx);
rig.expect_parent_chain_process();
bl.parent_chain_processed(chain_hash, BatchProcessResult::Success(true), &mut cx);
assert_eq!(bl.parent_queue.len(), 0);
}
#[test]
fn test_parent_lookup_too_many_attempts() {
let (mut bl, mut cx, mut rig) = TestRig::test_setup(None);
let parent = rig.rand_block();
let block = rig.block_with_parent(parent.canonical_root());
let chain_hash = block.canonical_root();
let peer_id = PeerId::random();
// Trigger the request
bl.search_parent(Box::new(block), peer_id, &mut cx);
for i in 1..=parent_lookup::PARENT_FAIL_TOLERANCE + 1 {
let id = rig.expect_parent_request();
match i % 2 {
// make sure every error is accounted for
0 => {
// The request fails. It should be tried again.
bl.parent_lookup_failed(id, peer_id, &mut cx);
}
_ => {
// Send a bad block this time. It should be tried again.
let bad_block = rig.rand_block();
bl.parent_lookup_response(id, peer_id, Some(Box::new(bad_block)), D, &mut cx);
rig.expect_penalty();
}
}
if i < parent_lookup::PARENT_FAIL_TOLERANCE {
assert_eq!(bl.parent_queue[0].failed_attempts(), dbg!(i));
}
}
assert_eq!(bl.parent_queue.len(), 0);
assert!(bl.failed_chains.contains(&chain_hash));
}
#[test]
fn test_parent_lookup_too_deep() {
let (mut bl, mut cx, mut rig) = TestRig::test_setup(None);
let mut blocks =
Vec::<SignedBeaconBlock<E>>::with_capacity(parent_lookup::PARENT_DEPTH_TOLERANCE);
while blocks.len() < parent_lookup::PARENT_DEPTH_TOLERANCE {
let parent = blocks
.last()
.map(|b| b.canonical_root())
.unwrap_or_else(Hash256::random);
let block = rig.block_with_parent(parent);
blocks.push(block);
}
let peer_id = PeerId::random();
let trigger_block = blocks.pop().unwrap();
let chain_hash = trigger_block.canonical_root();
bl.search_parent(Box::new(trigger_block), peer_id, &mut cx);
for block in blocks.into_iter().rev() {
let id = rig.expect_parent_request();
// the block
bl.parent_lookup_response(id, peer_id, Some(Box::new(block.clone())), D, &mut cx);
// the stream termination
bl.parent_lookup_response(id, peer_id, None, D, &mut cx);
// the processing request
rig.expect_block_process();
// the processing result
bl.parent_block_processed(
chain_hash,
Err(BlockError::ParentUnknown(Box::new(block))),
&mut cx,
)
}
rig.expect_penalty();
assert!(bl.failed_chains.contains(&chain_hash));
}
#[test]
fn test_parent_lookup_disconnection() {
let (mut bl, mut cx, mut rig) = TestRig::test_setup(None);
let peer_id = PeerId::random();
let trigger_block = rig.rand_block();
bl.search_parent(Box::new(trigger_block), peer_id, &mut cx);
bl.peer_disconnected(&peer_id, &mut cx);
assert!(bl.parent_queue.is_empty());
}