Files
lighthouse/beacon_node/network/src/sync/block_lookups/mod.rs

912 lines
39 KiB
Rust

//! Implements block lookup sync.
//!
//! Block lookup sync is triggered when a peer claims to have imported a block we don't know about.
//! For example, a peer attesting to a head block root that is not in our fork-choice. Lookup sync
//! is recursive in nature, as we may discover that this attested head block root has a parent that
//! is also unknown to us.
//!
//! Block lookup is implemented as an event-driven state machine. It sends events to the network and
//! beacon processor, and expects some set of events back. A discrepancy in the expected event API
//! will result in lookups getting "stuck". A lookup becomes stuck when there is no future event
//! that will trigger the lookup to make progress. There's a fallback mechanism that drops lookups
//! that live for too long, logging the line "Notify the devs a sync lookup is stuck".
//!
//! The expected event API is documented in the code paths that are making assumptions with the
//! comment prefix "Lookup sync event safety:"
//!
//! Block lookup sync attempts to not re-download or re-process data that we already have. Block
//! components are cached temporarily in multiple places before they are imported into fork-choice.
//! Therefore, block lookup sync must peek these caches correctly to decide when to skip a download
//! or consider a lookup complete. These caches are read from the `SyncNetworkContext` and its state
//! returned to this module as `LookupRequestResult` variants.
use self::parent_chain::{NodeChain, compute_parent_chains};
pub use self::single_block_lookup::DownloadResult;
use self::single_block_lookup::{LookupRequestError, PeerType, SingleBlockLookup};
use super::manager::{BlockProcessType, SLOT_IMPORT_TOLERANCE};
use super::network_context::{RpcResponseError, SyncNetworkContext};
use crate::metrics;
use crate::network_beacon_processor::BlockProcessingResult;
use crate::sync::SyncMessage;
use crate::sync::block_lookups::parent_chain::find_oldest_fork_ancestor;
use crate::sync::block_lookups::single_block_lookup::{AwaitingParent, ImportedAction};
use beacon_chain::BeaconChainTypes;
use fnv::FnvHashMap;
use lighthouse_network::PeerId;
use lighthouse_network::service::api_types::SingleLookupReqId;
use lru_cache::LRUTimeCache;
use std::collections::hash_map::Entry;
use std::sync::Arc;
use std::time::Duration;
use store::Hash256;
use tracing::{debug, error, warn};
use types::{
DataColumnSidecarList, EthSpec, ExecutionBlockHash, SignedBeaconBlock,
SignedExecutionPayloadEnvelope,
};
pub mod parent_chain;
mod single_block_lookup;
/// 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.
///
/// Have the same value as range's sync tolerance to consider a peer synced. Once sync lookup
/// reaches the maximum depth it will force trigger range sync.
pub(crate) const PARENT_DEPTH_TOLERANCE: usize = SLOT_IMPORT_TOLERANCE;
const IGNORED_CHAINS_CACHE_EXPIRY_SECONDS: u64 = 60;
pub const SINGLE_BLOCK_LOOKUP_MAX_ATTEMPTS: u8 = 4;
/// Maximum time we allow a lookup to exist before assuming it is stuck and will never make
/// progress. Assume the worse case processing time per block component set * times max depth.
/// 15 * 2 * 32 = 16 minutes.
const LOOKUP_MAX_DURATION_STUCK_SECS: u64 = 15 * PARENT_DEPTH_TOLERANCE as u64;
/// The most common case of child-lookup without peers is receiving block components before the
/// attestation deadline when the node is lagging behind. Once peers start attesting for the child
/// lookup at most after 4 seconds, the lookup should gain peers.
const LOOKUP_MAX_DURATION_NO_PEERS_SECS: u64 = 10;
/// Lookups contain untrusted data, including blocks that have not yet been validated. In case of
/// bugs or malicious activity we want to bound how much memory these lookups can consume. Aprox the
/// max size of a lookup is ~ 10 MB (current max size of gossip and RPC blocks). 200 lookups can
/// take at most 2 GB. 200 lookups allow 3 parallel chains of depth 64 (current maximum).
const MAX_LOOKUPS: usize = 200;
type BlockDownloadResponse<E> = Result<DownloadResult<Arc<SignedBeaconBlock<E>>>, RpcResponseError>;
type CustodyDownloadResponse<E> =
Result<DownloadResult<DataColumnSidecarList<E>>, RpcResponseError>;
type PayloadDownloadResponse<E> =
Result<DownloadResult<Arc<SignedExecutionPayloadEnvelope<E>>>, RpcResponseError>;
pub enum BlockComponent<E: EthSpec> {
Block(DownloadResult<Arc<SignedBeaconBlock<E>>>),
Sidecar,
}
pub type SingleLookupId = u32;
pub struct BlockLookups<T: BeaconChainTypes> {
/// A cache of block roots that must be ignored for some time to prevent useless searches. For
/// example if a chain is too long, its lookup chain is dropped, and range sync is expected to
/// eventually sync those blocks
ignored_chains: LRUTimeCache<Hash256>,
// TODO: Why not index lookups by block_root?
single_block_lookups: FnvHashMap<SingleLookupId, SingleBlockLookup<T>>,
/// Used for testing assertions
metrics: BlockLookupsMetrics,
}
#[cfg(test)]
use lighthouse_network::service::api_types::Id;
#[cfg(test)]
#[derive(Debug)]
pub(crate) struct BlockLookupSummary {
/// Lookup ID
pub id: Id,
/// Requested block root
pub block_root: Hash256,
/// List of peers that claim to have imported this set of block components.
pub peers: Vec<PeerId>,
}
impl<T: BeaconChainTypes> BlockLookups<T> {
pub fn new() -> Self {
Self {
ignored_chains: LRUTimeCache::new(Duration::from_secs(
IGNORED_CHAINS_CACHE_EXPIRY_SECONDS,
)),
single_block_lookups: Default::default(),
metrics: <_>::default(),
}
}
#[cfg(test)]
pub(crate) fn metrics(&self) -> &BlockLookupsMetrics {
&self.metrics
}
#[cfg(test)]
pub(crate) fn insert_ignored_chain(&mut self, block_root: Hash256) {
self.ignored_chains.insert(block_root);
}
#[cfg(test)]
pub(crate) fn get_ignored_chains(&mut self) -> Vec<Hash256> {
self.ignored_chains.keys().cloned().collect()
}
#[cfg(test)]
pub(crate) fn active_single_lookups(&self) -> Vec<BlockLookupSummary> {
self.single_block_lookups
.iter()
.map(|(id, l)| BlockLookupSummary {
id: *id,
block_root: l.block_root(),
peers: l.all_peers(),
})
.collect()
}
/// Returns a vec of all parent lookup chains by tip, in descending slot order (tip first)
pub(crate) fn active_parent_lookups(&self) -> Vec<NodeChain> {
compute_parent_chains(
&self
.single_block_lookups
.values()
.map(|lookup| lookup.into())
.collect::<Vec<_>>(),
)
}
/* Lookup requests */
/// Creates a parent lookup for the block with the given `block_root` and immediately triggers it.
/// If a parent lookup exists or is triggered, a current lookup will be created.
///
/// Returns true if the lookup is created or already exists
#[must_use = "only reference the new lookup if returns true"]
pub fn search_child_and_parent(
&mut self,
block_root: Hash256,
block_component: BlockComponent<T::EthSpec>,
parent_root: Hash256,
parent_block_hash: Option<ExecutionBlockHash>,
peer_id: PeerId,
cx: &mut SyncNetworkContext<T>,
) -> bool {
let parent_lookup_exists = self.search_parent_of_child(
parent_root,
&PeerType::new(parent_block_hash),
block_root,
&[peer_id],
cx,
);
// Only create the child lookup if the parent exists
if parent_lookup_exists {
// `search_parent_of_child` ensures that the parent lookup exists so we can safely wait for it
self.new_current_lookup(
block_root,
Some(block_component),
Some(AwaitingParent::new(parent_root, parent_block_hash)),
// On a `UnknownParentBlock` or `UnknownParentSidecarHeader` event the peer is not
// required to have the rest of the block components. Create the lookup with zero
// peers to house the block components. We don't know the child's fork yet, so use
// `Block` conservatively; the correct peer set is established when the child's
// block downloads and its FULL children begin attesting.
&[],
&PeerType::Block,
cx,
)
} else {
false
}
}
/// Search a block whose parent root is unknown.
///
/// Returns true if the lookup is created or already exists
#[must_use = "only reference the new lookup if returns true"]
pub fn search_unknown_block(
&mut self,
block_root: Hash256,
peer_source: &[PeerId],
cx: &mut SyncNetworkContext<T>,
) -> bool {
self.new_current_lookup(block_root, None, None, peer_source, &PeerType::Block, cx)
}
/// A block or blob triggers the search of a parent.
/// Check if this new lookup extends a bad chain:
/// - Extending `child_block_root_trigger` would exceed the max depth
/// - `block_root_to_search` is a failed chain
///
/// Returns true if the lookup is created or already exists
#[must_use = "only reference the new lookup if returns true"]
pub fn search_parent_of_child(
&mut self,
block_root_to_search: Hash256,
// Classifies `peers` relative to the parent being searched: `GloasChild` when they imported
// the FULL child (and so can serve the parent's payload envelope and data columns), else
// `Block`.
peer_type: &PeerType,
child_block_root_trigger: Hash256,
peers: &[PeerId],
cx: &mut SyncNetworkContext<T>,
) -> bool {
let parent_chains = self.active_parent_lookups();
for (chain_idx, parent_chain) in parent_chains.iter().enumerate() {
// `block_root_to_search` will trigger a new lookup, and it will extend a parent_chain
// beyond its max length
let block_would_extend_chain = parent_chain.ancestor() == child_block_root_trigger;
// `block_root_to_search` already has a lookup, and with the block trigger it extends
// the parent_chain beyond its length. This can happen because when creating a lookup
// for a new root we don't do any parent chain length checks
let trigger_is_chain_tip = parent_chain.tip == child_block_root_trigger;
if (block_would_extend_chain || trigger_is_chain_tip)
&& parent_chain.len() >= PARENT_DEPTH_TOLERANCE
{
debug!(block_root = ?block_root_to_search, "Parent lookup chain too long");
// Searching for this parent would extend a parent chain over the max
// Insert the tip only to chains to ignore
self.ignored_chains.insert(parent_chain.tip);
// Note: Drop only the chain that's too long until it merges with another chain
// that's not too long. Consider this attack: there's a chain of valid unknown
// blocks A -> B. A malicious peer builds `PARENT_DEPTH_TOLERANCE` garbage
// blocks on top of A forming A -> C. The malicious peer forces us to fetch C
// from it, which will result in parent A hitting the chain_too_long error. Then
// the valid chain A -> B is dropped too.
//
// `find_oldest_fork_ancestor` should never return Err, unwrapping to tip for
// complete-ness
let parent_chain_tip = parent_chain.tip;
let block_to_drop =
find_oldest_fork_ancestor(parent_chains, chain_idx).unwrap_or(parent_chain_tip);
// Drop all lookups descending from the child of the too long parent chain
if let Some((lookup_id, lookup)) = self
.single_block_lookups
.iter()
.find(|(_, l)| l.block_root() == block_to_drop)
{
// If a lookup chain is too long, we can't distinguish a valid chain from a
// malicious one. We must attempt to sync this chain to not lose liveness. If
// the chain grows too long, we stop lookup sync and transition this head to
// forward range sync. We need to tell range sync which head to sync to, and
// from which peers. The lookup of the very tip of this chain may contain zero
// peers if it's the parent-child lookup. So we do a bit of a trick here:
// - Tell range sync to sync to the tip's root (if available, else its ancestor)
// - But use all peers in the ancestor lookup, which should have at least one
// peer, and its peer set is a strict superset of the tip's lookup.
if let Some((_, tip_lookup)) = self
.single_block_lookups
.iter()
.find(|(_, l)| l.block_root() == parent_chain_tip)
{
cx.send_sync_message(SyncMessage::AddPeersForceRangeSync {
peers: lookup.all_peers(),
head_slot: tip_lookup.peek_downloaded_block_slot(),
head_root: parent_chain_tip,
});
} else {
// Should never happen, log error and continue the lookup drop
error!(
error = "Parent chain tip lookup not found",
block_root = ?parent_chain_tip,
"Unable to transition lookup to range sync"
);
}
// Do not downscore peers here. Because we can't distinguish a valid chain from
// a malicious one we may penalize honest peers for attempting to discover us a
// valid chain. Until blocks_by_range allows to specify a tip, for example with
// https://github.com/ethereum/consensus-specs/pull/3845 we will have poor
// attributability. A peer can send us garbage blocks over blocks_by_root, and
// then correct blocks via blocks_by_range.
self.drop_lookup_and_children(*lookup_id, "chain_too_long");
} else {
// Should never happen
error!(
error = "Block to drop lookup not found",
block_root = ?block_to_drop,
"Unable to transition lookup to range sync"
);
}
return false;
}
}
// `block_root_to_search` is a failed chain check happens inside new_current_lookup
self.new_current_lookup(block_root_to_search, None, None, peers, peer_type, cx)
}
/// Searches for a single block hash. If the blocks parent is unknown, a chain of blocks is
/// constructed.
/// Returns true if the lookup is created or already exists
#[must_use = "only reference the new lookup if returns true"]
fn new_current_lookup(
&mut self,
block_root: Hash256,
block_component: Option<BlockComponent<T::EthSpec>>,
awaiting_parent: Option<AwaitingParent>,
peers: &[PeerId],
peer_type: &PeerType,
cx: &mut SyncNetworkContext<T>,
) -> bool {
// If this block or it's parent is part of a known ignored chain, ignore it.
if self.ignored_chains.contains(&block_root) {
debug!(?block_root, "Dropping lookup for block marked ignored");
return false;
}
// Do not re-request a block that is already being requested
if let Some((&lookup_id, lookup)) = self
.single_block_lookups
.iter_mut()
.find(|(_id, lookup)| lookup.is_for_block(block_root))
{
if let Some(block_component) = block_component {
let imported = lookup.add_child_components(block_component);
if !imported {
debug!(?block_root, "Lookup child component ignored");
}
}
if let Err(e) = self.add_peers_to_lookup_and_ancestors(lookup_id, peers, peer_type, cx)
{
warn!(error = ?e, "Error adding peers to ancestor lookup");
}
return true;
}
// Ensure that awaiting parent exists, otherwise this lookup won't be able to make progress
if let Some(awaiting_parent) = awaiting_parent
&& !self
.single_block_lookups
.iter()
.any(|(_, lookup)| lookup.is_for_block(awaiting_parent.parent_root()))
{
warn!(block_root = ?awaiting_parent, "Ignoring child lookup parent lookup not found");
return false;
}
// Lookups contain untrusted data, bound the total count of lookups hold in memory to reduce
// the risk of OOM in case of bugs of malicious activity.
if self.single_block_lookups.len() >= MAX_LOOKUPS {
warn!(?block_root, "Dropping lookup reached max");
return false;
}
// If we know that this lookup has unknown parent (is awaiting a parent lookup to resolve),
// signal here to hold processing downloaded data.
let mut lookup =
SingleBlockLookup::new(block_root, peers, peer_type, cx.next_id(), awaiting_parent);
let _guard = lookup.span.clone().entered();
// Add block components to the new request
if let Some(block_component) = block_component {
lookup.add_child_components(block_component);
}
let id = lookup.id;
let lookup = match self.single_block_lookups.entry(id) {
Entry::Vacant(entry) => entry.insert(lookup),
Entry::Occupied(_) => {
// Should never happen
warn!(id, "Lookup exists with same id");
return false;
}
};
debug!(
?peers,
?block_root,
?awaiting_parent,
id = lookup.id,
"Created block lookup"
);
metrics::inc_counter(&metrics::SYNC_LOOKUP_CREATED);
self.metrics.created_lookups += 1;
let result = lookup.continue_requests(cx);
if self.on_lookup_result(id, result, "new_current_lookup", cx) {
self.update_metrics();
true
} else {
false
}
}
/* Lookup responses */
/// Process a block response received from a single lookup request.
pub fn on_block_download_response(
&mut self,
id: SingleLookupReqId,
response: BlockDownloadResponse<T::EthSpec>,
cx: &mut SyncNetworkContext<T>,
) {
let Some(lookup) = self.single_block_lookups.get_mut(&id.lookup_id) else {
debug!(?id, "Block returned for single block lookup not present");
return;
};
let result = lookup.on_block_download_response(id.req_id, response, cx);
self.on_lookup_result(id.lookup_id, result, "block_download_response", cx);
}
pub fn on_custody_download_response(
&mut self,
id: SingleLookupReqId,
response: CustodyDownloadResponse<T::EthSpec>,
cx: &mut SyncNetworkContext<T>,
) {
let Some(lookup) = self.single_block_lookups.get_mut(&id.lookup_id) else {
debug!(?id, "Custody returned for single block lookup not present");
return;
};
let result = lookup.on_custody_download_response(id.req_id, response, cx);
self.on_lookup_result(id.lookup_id, result, "custody_download_response", cx);
}
pub fn on_payload_download_response(
&mut self,
id: SingleLookupReqId,
response: PayloadDownloadResponse<T::EthSpec>,
cx: &mut SyncNetworkContext<T>,
) {
let Some(lookup) = self.single_block_lookups.get_mut(&id.lookup_id) else {
debug!(
?id,
"Payload envelope returned for single block lookup not present"
);
return;
};
let result = lookup.on_payload_download_response(id.req_id, response, cx);
self.on_lookup_result(id.lookup_id, result, "payload_download_response", cx);
}
/* Error responses */
pub fn peer_disconnected(&mut self, peer_id: &PeerId) {
for (id, lookup) in self.single_block_lookups.iter_mut() {
lookup.remove_peer(peer_id);
if lookup.has_no_peers() {
debug!(%id, "Lookup has no peers");
}
}
}
/* Processing responses */
pub fn on_processing_result(
&mut self,
process_type: BlockProcessType,
result: BlockProcessingResult,
cx: &mut SyncNetworkContext<T>,
) {
let id = process_type.id();
let Some(lookup) = self.single_block_lookups.get_mut(&id) else {
debug!(id, "Unknown single block lookup");
return;
};
let block_root = lookup.block_root();
debug!(
?block_root,
id,
?process_type,
?result,
"Received lookup processing result"
);
let lookup_result = match process_type {
BlockProcessType::SingleBlock { .. } => lookup.on_block_processing_result(&result, cx),
BlockProcessType::SingleCustodyColumn(_) => {
lookup.on_data_processing_result(&result, cx)
}
BlockProcessType::SinglePayloadEnvelope(_) => {
lookup.on_payload_processing_result(&result, cx)
}
};
match &result {
BlockProcessingResult::Imported(fully_imported, _) => {
// Some component got imported potentially continue
if lookup.is_complete() {
if self.single_block_lookups.remove(&id).is_some() {
debug!(?block_root, id, "Dropping completed lookup");
metrics::inc_counter(&metrics::SYNC_LOOKUP_COMPLETED);
self.metrics.completed_lookups += 1;
// Block imported, continue the requests of pending child blocks
self.continue_child_lookups(
ImportedAction::LookupComplete { block_root },
cx,
);
self.update_metrics();
} else {
debug!(id, "Attempting to drop non-existent lookup");
}
} else if *fully_imported
&& matches!(process_type, BlockProcessType::SingleBlock { .. })
{
// The block imported into fork choice but the lookup is not `is_complete`: its
// data may have become available via the da_checker (so the lookup's own
// request never completed), or it is a Gloas block whose payload arrives
// separately. Unblock the appropriate children, and complete the lookup unless
// a FULL Gloas child still awaits the payload.
let import_action = match lookup.peek_downloaded_bid_block_hash() {
Some(bid_block_hash) => ImportedAction::GloasBlockComplete {
block_root,
bid_block_hash,
},
None => ImportedAction::LookupComplete { block_root },
};
self.continue_child_lookups(import_action, cx);
if !self.has_any_awaiting_children(block_root) {
self.single_block_lookups.remove(&id);
metrics::inc_counter(&metrics::SYNC_LOOKUP_COMPLETED);
self.metrics.completed_lookups += 1;
debug!(
?block_root,
id, "Dropping completed lookup after block import"
);
}
self.update_metrics();
}
}
BlockProcessingResult::ParentUnknown {
parent_root,
parent_block_hash,
} => {
// Parent unknown error, create parent lookup
let peers = lookup.all_peers();
if !self.search_parent_of_child(
*parent_root,
&PeerType::new(*parent_block_hash),
block_root,
&peers,
cx,
) {
self.drop_lookup_and_children(id, "Failed");
self.update_metrics();
}
}
BlockProcessingResult::Error { .. } => {}
}
self.on_lookup_result(id, lookup_result, "processing_result", cx);
}
pub fn has_any_awaiting_children(&self, block_root: Hash256) -> bool {
self.single_block_lookups
.iter()
.any(|(_, lookup)| lookup.is_awaiting_block(block_root))
}
/// Makes progress on the immediate children of `block_root`
pub fn continue_child_lookups(
&mut self,
import_action: ImportedAction,
cx: &mut SyncNetworkContext<T>,
) {
let mut lookup_results = vec![]; // < need to buffer lookup results to not re-borrow &mut self
for (id, lookup) in self.single_block_lookups.iter_mut() {
// If lookup is awaiting parent?
// - If Some
// - If parent_root lookup got block
// - Check if the child is FULL, if so keep waiting, otherwise continue and resolve
if lookup.maybe_resolve_awaiting_parent(import_action) {
debug!(
?import_action,
id,
block_root = ?lookup.block_root(),
"Continuing child lookup"
);
let result = lookup.continue_requests(cx);
lookup_results.push((*id, result));
}
}
for (id, result) in lookup_results {
self.on_lookup_result(id, result, "continue_child_lookups", cx);
}
}
/// Drops `dropped_id` lookup and all its children recursively. Lookups awaiting a parent need
/// the parent to make progress to resolve, therefore we must drop them if the parent is
/// dropped.
pub fn drop_lookup_and_children(&mut self, dropped_id: SingleLookupId, reason: &'static str) {
if let Some(dropped_lookup) = self.single_block_lookups.remove(&dropped_id) {
debug!(
id = ?dropped_id,
block_root = ?dropped_lookup.block_root(),
awaiting_parent = ?dropped_lookup.awaiting_parent(),
reason,
"Dropping lookup"
);
metrics::inc_counter_vec(&metrics::SYNC_LOOKUP_DROPPED, &[reason]);
self.metrics.dropped_lookups += 1;
let child_lookups = self
.single_block_lookups
.iter()
.filter(|(_, lookup)| lookup.is_awaiting_block(dropped_lookup.block_root()))
.map(|(id, _)| *id)
.collect::<Vec<_>>();
for id in child_lookups {
self.drop_lookup_and_children(id, reason);
}
}
}
/// Common handler a lookup request error, drop it and update metrics
/// Returns true if the lookup is created or already exists
fn on_lookup_result(
&mut self,
id: SingleLookupId,
result: Result<(), LookupRequestError>,
source: &str,
cx: &mut SyncNetworkContext<T>,
) -> bool {
match result {
Ok(_) => {
// The lookup may have become complete from already-cached components during
// `continue_requests` (e.g. the block became available via the da_checker), in
// which case no `Imported` processing result is emitted. Detect that here.
if self
.single_block_lookups
.get(&id)
.is_some_and(|lookup| lookup.is_complete())
&& let Some(lookup) = self.single_block_lookups.remove(&id)
{
let block_root = lookup.block_root();
debug!(?block_root, id, "Dropping completed lookup (cached)");
metrics::inc_counter(&metrics::SYNC_LOOKUP_COMPLETED);
self.metrics.completed_lookups += 1;
self.continue_child_lookups(ImportedAction::LookupComplete { block_root }, cx);
self.update_metrics();
}
true
}
// If UnknownLookup do not log the request error. No need to drop child lookups nor
// update metrics because the lookup does not exist.
Err(error) => {
// A FULL Gloas child re-awaits its parent's payload once the parent's block
// imports. A failed payload download must not cascade-drop the parent (and the
// child) — the payload may still arrive (e.g. via gossip). Retain the parent;
// genuinely stuck lookups are pruned by `drop_stuck_lookups`.
if source == "payload_download_response"
&& let Some(block_root) =
self.single_block_lookups.get(&id).map(|l| l.block_root())
&& self.has_any_awaiting_children(block_root)
{
debug!(
id,
source,
?error,
"Retaining parent with a child awaiting its payload"
);
return false;
}
debug!(id, source, ?error, "Dropping lookup on request error");
self.drop_lookup_and_children(id, error.into());
self.update_metrics();
false
}
}
}
/* Helper functions */
/// Drops all the single block requests and returns how many requests were dropped.
pub fn drop_single_block_requests(&mut self) -> usize {
let requests_to_drop = self.single_block_lookups.len();
self.single_block_lookups.clear();
requests_to_drop
}
pub fn update_metrics(&self) {
metrics::set_gauge(
&metrics::SYNC_SINGLE_BLOCK_LOOKUPS,
self.single_block_lookups.len() as i64,
);
}
/// Perform some prune operations on lookups on some interval
pub fn prune_lookups(&mut self) {
self.drop_lookups_without_peers();
self.drop_stuck_lookups();
}
/// Lookups without peers are allowed to exist for some time. See this common race condition:
///
/// 1. Receive unknown block parent event
/// 2. Create child lookup with zero peers
/// 3. Parent is processed, before receiving any attestation for the child block
/// 4. Child lookup is attempted to make progress but has no peers
/// 5. We receive an attestion for child block and add a peer to the child block lookup
///
/// On step 4 we could drop the lookup because we attempt to issue a request with no peers
/// available. This has two issues:
/// - We may drop the lookup while some other block component is processing, triggering an
/// unknown lookup error. This can potentially cause un-related child lookups to also be
/// dropped when calling `drop_lookup_and_children`.
/// - We lose all progress of the lookup, and have to re-download its components that we may
/// already have there cached.
///
/// Instead there's no negative for keeping lookups with no peers around for some time. If we
/// regularly prune them, it should not be a memory concern (TODO: maybe yes!).
fn drop_lookups_without_peers(&mut self) {
for (lookup_id, block_root) in self
.single_block_lookups
.values()
.filter(|lookup| {
// Do not drop lookup that are awaiting events to prevent inconsinstencies. If a
// lookup gets stuck, it will be eventually pruned by `drop_stuck_lookups`
lookup.has_no_peers()
&& lookup.elapsed_since_created()
> Duration::from_secs(LOOKUP_MAX_DURATION_NO_PEERS_SECS)
&& !lookup.is_awaiting_event()
})
.map(|lookup| (lookup.id, lookup.block_root()))
.collect::<Vec<_>>()
{
debug!(
id = lookup_id,
%block_root,
"Dropping lookup with no peers"
);
self.drop_lookup_and_children(lookup_id, "no_peers");
}
}
/// Safety mechanism to unstuck lookup sync. Lookup sync if purely event driven and depends on
/// external components to feed it events to make progress. If there is a bug in network, in
/// beacon processor, or here internally: lookups can get stuck forever. A stuck lookup can
/// stall a node indefinitely as other lookup will be awaiting on a parent lookup to make
/// progress.
///
/// If a lookup lasts more than LOOKUP_MAX_DURATION_SECS this function will find its oldest
/// ancestor and then drop it and all its children. This action will allow the node to unstuck
/// itself. Bugs that cause lookups to get stuck may be triggered consistently. So this strategy
/// is useful for two reasons:
///
/// - One single clear warn level log per stuck incident
/// - If the original bug is sporadic, it reduces the time a node is stuck from forever to 15 min
fn drop_stuck_lookups(&mut self) {
// While loop to find and drop all disjoint trees of potentially stuck lookups.
while let Some(stuck_lookup) = self.single_block_lookups.values().find(|lookup| {
lookup.elapsed_since_created() > Duration::from_secs(LOOKUP_MAX_DURATION_STUCK_SECS)
}) {
let ancestor_stuck_lookup = match self.find_oldest_ancestor_lookup(stuck_lookup) {
Ok(lookup) => lookup,
Err(e) => {
warn!(error = ?e,"Error finding oldest ancestor lookup");
// Default to dropping the lookup that exceeds the max duration so at least
// eventually sync should be unstuck
stuck_lookup
}
};
if stuck_lookup.id == ancestor_stuck_lookup.id {
warn!(
block_root = ?stuck_lookup.block_root(),
lookup = ?stuck_lookup,
"Notify the devs a sync lookup is stuck"
);
} else {
warn!(
block_root = ?stuck_lookup.block_root(),
lookup = ?stuck_lookup,
ancestor_block_root = ?ancestor_stuck_lookup.block_root(),
ancestor_lookup = ?ancestor_stuck_lookup,
"Notify the devs a sync lookup is stuck"
);
}
metrics::inc_counter(&metrics::SYNC_LOOKUPS_STUCK);
self.drop_lookup_and_children(ancestor_stuck_lookup.id, "lookup_stuck");
}
}
/// Recursively find the oldest ancestor lookup of another lookup
fn find_oldest_ancestor_lookup<'a>(
&'a self,
lookup: &'a SingleBlockLookup<T>,
) -> Result<&'a SingleBlockLookup<T>, String> {
if let Some(awaiting_parent) = lookup.awaiting_parent() {
if let Some(parent_lookup) = self
.single_block_lookups
.values()
.find(|l| l.is_parent_of(awaiting_parent))
{
self.find_oldest_ancestor_lookup(parent_lookup)
} else {
Err(format!(
"Lookup references unknown parent {awaiting_parent:?}"
))
}
} else {
Ok(lookup)
}
}
/// Adds peers to a lookup and its ancestors recursively.
///
/// Note: Takes a `lookup_id` as argument to allow recursion on mutable lookups, without having
/// to duplicate the code to add peers to a lookup
fn add_peers_to_lookup_and_ancestors(
&mut self,
lookup_id: SingleLookupId,
peers: &[PeerId],
peer_type: &PeerType,
cx: &mut SyncNetworkContext<T>,
) -> Result<(), String> {
let lookup = self
.single_block_lookups
.get_mut(&lookup_id)
.ok_or(format!("Unknown lookup for id {lookup_id}"))?;
let mut added_some_peer = false;
for peer in peers {
if lookup.add_peer(*peer, peer_type) {
added_some_peer = true;
debug!(
block_root = ?lookup.block_root(),
?peer,
"Adding peer to existing single block lookup"
);
}
}
if let Some(&awaiting_parent) = lookup.awaiting_parent() {
// Regardless of gloas full/empty the lookup to add peers to is keyed by block_root
if let Some(parent_id) = self
.single_block_lookups
.iter()
.find(|(_, l)| l.is_parent_of(&awaiting_parent))
.map(|(parent_id, _)| *parent_id)
{
self.add_peers_to_lookup_and_ancestors(
parent_id,
peers,
&(&awaiting_parent).into(),
cx,
)
} else {
Err(format!("Lookup references unknown {awaiting_parent:?}"))
}
} else if added_some_peer {
// If this lookup is not awaiting a parent and we added at least one peer, attempt to
// make progress. It is possible that a lookup is created with zero peers, attempted to
// make progress, and then receives peers. After that time the lookup will never be
// pruned with `drop_lookups_without_peers` because it has peers. This is rare corner
// case, but it can result in stuck lookups.
let result = lookup.continue_requests(cx);
self.on_lookup_result(lookup_id, result, "add_peers", cx);
Ok(())
} else {
Ok(())
}
}
}
#[derive(Default, Clone, Debug)]
pub(crate) struct BlockLookupsMetrics {
pub created_lookups: usize,
pub dropped_lookups: usize,
pub completed_lookups: usize,
}