litep2p/kad: Configure periodic network bootstrap for better throughput

substrate/client/network/src/litep2p/discovery.rs

@@ -51,7 +51,7 @@ use std::{
 	cmp,
 	collections::{HashMap, HashSet, VecDeque},
 	pin::Pin,
-	sync::Arc,
+	sync::{atomic::AtomicUsize, Arc},
 	task::{Context, Poll},
 	time::{Duration, Instant},
 };
@@ -62,12 +62,21 @@ const LOG_TARGET: &str = "sub-libp2p::discovery";
 /// Kademlia query interval.
 const KADEMLIA_QUERY_INTERVAL: Duration = Duration::from_secs(5);
 
+/// The convergence time between 2 `FIND_NODE` queries.
+///
+/// The time is exponentially increased after each query until it reaches 120 seconds.
+/// The time is reset to `KADEMLIA_QUERY_INTERVAL` after a failed query.
+const CONVERGENCE_QUERY_INTERVAL: Duration = Duration::from_secs(120);
+
 /// mDNS query interval.
 const MDNS_QUERY_INTERVAL: Duration = Duration::from_secs(30);
 
 /// Minimum number of confirmations received before an address is verified.
 const MIN_ADDRESS_CONFIRMATIONS: usize = 5;
 
+/// Maximum number of in-flight `FIND_NODE` queries.
+const MAX_INFLIGHT_FIND_NODE_QUERIES: usize = 16;
+
 /// Discovery events.
 #[derive(Debug)]
 pub enum DiscoveryEvent {
@@ -172,8 +181,8 @@ pub struct Discovery {
 	/// If `None`, there is currently a query pending.
 	next_kad_query: Option<Delay>,
 
-	/// Active `FIND_NODE` query if it exists.
-	find_node_query_id: Option<QueryId>,
+	/// Active `FIND_NODE` queries.
+	find_node_queries: HashMap<QueryId, std::time::Instant>,
 
 	/// Pending events.
 	pending_events: VecDeque<DiscoveryEvent>,
@@ -195,6 +204,12 @@ pub struct Discovery {
 
 	/// Delay to next `FIND_NODE` query.
 	duration_to_next_find_query: Duration,
+
+	/// Number of connected peers as reported by the blocks announcement protocol.
+	num_connected_peers: Arc<AtomicUsize>,
+
+	/// Number of active connections over which we interrupt the discovery process.
+	discovery_only_if_under_num: usize,
 }
 
 /// Legacy (fallback) Kademlia protocol name based on `protocol_id`.
@@ -229,6 +244,8 @@ impl Discovery {
 		protocol_id: &ProtocolId,
 		known_peers: HashMap<PeerId, Vec<Multiaddr>>,
 		listen_addresses: Arc<RwLock<HashSet<Multiaddr>>>,
+		num_connected_peers: Arc<AtomicUsize>,
+		discovery_only_if_under_num: usize,
 		_peerstore_handle: Arc<dyn PeerStoreProvider>,
 	) -> (Self, PingConfig, IdentifyConfig, KademliaConfig, Option<MdnsConfig>) {
 		let (ping_config, ping_event_stream) = PingConfig::default();
@@ -271,7 +288,7 @@ impl Discovery {
 				kademlia_handle,
 				_peerstore_handle,
 				listen_addresses,
-				find_node_query_id: None,
+				find_node_queries: HashMap::new(),
 				pending_events: VecDeque::new(),
 				duration_to_next_find_query: Duration::from_secs(1),
 				address_confirmations: LruMap::new(ByLength::new(8)),
@@ -282,6 +299,8 @@ impl Discovery {
 					genesis_hash,
 					fork_id,
 				)]),
+				num_connected_peers,
+				discovery_only_if_under_num,
 			},
 			ping_config,
 			identify_config,
@@ -290,6 +309,11 @@ impl Discovery {
 		)
 	}
 
+	/// Get number of connected peers.
+	fn num_connected_peers(&self) -> usize {
+		self.num_connected_peers.load(std::sync::atomic::Ordering::Relaxed)
+	}
+
 	/// Add known peer to `Kademlia`.
 	#[allow(unused)]
 	pub async fn add_known_peer(&mut self, peer: PeerId, addresses: Vec<Multiaddr>) {
@@ -448,42 +472,58 @@ impl Stream for Discovery {
 
 		if let Some(mut delay) = this.next_kad_query.take() {
 			match delay.poll_unpin(cx) {
-				Poll::Pending => {
-					this.next_kad_query = Some(delay);
-				},
 				Poll::Ready(()) => {
-					let peer = PeerId::random();
+					let num_peers = this.num_connected_peers();
+					if num_peers < this.discovery_only_if_under_num &&
+						this.find_node_queries.len() < MAX_INFLIGHT_FIND_NODE_QUERIES
+					{
+						let peer = PeerId::random();
+						log::debug!(target: LOG_TARGET, "start next kademlia query for {peer:?}");
 
-					log::trace!(target: LOG_TARGET, "start next kademlia query for {peer:?}");
+						if let Ok(query_id) = this.kademlia_handle.try_find_node(peer) {
+							this.find_node_queries.insert(query_id, std::time::Instant::now());
 
-					match this.kademlia_handle.try_find_node(peer) {
-						Ok(query_id) => {
-							this.find_node_query_id = Some(query_id);
-							return Poll::Ready(Some(DiscoveryEvent::RandomKademliaStarted))
-						},
-						Err(()) => {
 							this.duration_to_next_find_query = cmp::min(
 								this.duration_to_next_find_query * 2,
-								Duration::from_secs(60),
+								CONVERGENCE_QUERY_INTERVAL,
 							);
 							this.next_kad_query =
 								Some(Delay::new(this.duration_to_next_find_query));
-						},
+
+							return Poll::Ready(Some(DiscoveryEvent::RandomKademliaStarted))
+						}
+					} else {
+						log::debug!(
+							target: LOG_TARGET,
+							"discovery is paused: {num_peers}/{} connected peers and in flight queries: {}/{MAX_INFLIGHT_FIND_NODE_QUERIES}",
+							this.discovery_only_if_under_num,
+							this.find_node_queries.len(),
+						);
 					}
+
+					this.duration_to_next_find_query =
+						cmp::min(this.duration_to_next_find_query * 2, CONVERGENCE_QUERY_INTERVAL);
+					this.next_kad_query = Some(Delay::new(this.duration_to_next_find_query));
+				},
+				Poll::Pending => {
+					this.next_kad_query = Some(delay);
 				},
 			}
 		}
 
 		match Pin::new(&mut this.kademlia_handle).poll_next(cx) {
 			Poll::Pending => {},
 			Poll::Ready(None) => return Poll::Ready(None),
-			Poll::Ready(Some(KademliaEvent::FindNodeSuccess { peers, .. })) => {
+			Poll::Ready(Some(KademliaEvent::FindNodeSuccess { peers, query_id, .. })) => {
 				// the addresses are already inserted into the DHT and in `TransportManager` so
 				// there is no need to add them again. The found peers must be registered to
 				// `Peerstore` so other protocols are aware of them through `Peerset`.
-				log::trace!(target: LOG_TARGET, "dht random walk yielded {} peers", peers.len());
 
-				this.next_kad_query = Some(Delay::new(KADEMLIA_QUERY_INTERVAL));
+				if let Some(instant) = this.find_node_queries.remove(&query_id) {
+					log::trace!(target: LOG_TARGET, "dht random walk yielded {} peers for {query_id:?} in {:?}", peers.len(), instant.elapsed());
+				} else {
+					log::trace!(target: LOG_TARGET, "dht random walk yielded {} peers for {query_id:?}", peers.len());
+				}
 
 				return Poll::Ready(Some(DiscoveryEvent::RoutingTableUpdate {
 					peers: peers.into_iter().map(|(peer, _)| peer).collect(),
@@ -507,15 +547,14 @@ impl Stream for Discovery {
 			Poll::Ready(Some(KademliaEvent::PutRecordSucess { query_id, key: _ })) =>
 				return Poll::Ready(Some(DiscoveryEvent::PutRecordSuccess { query_id })),
 			Poll::Ready(Some(KademliaEvent::QueryFailed { query_id })) => {
-				match this.find_node_query_id == Some(query_id) {
-					true => {
-						this.find_node_query_id = None;
-						this.duration_to_next_find_query =
-							cmp::min(this.duration_to_next_find_query * 2, Duration::from_secs(60));
-						this.next_kad_query = Some(Delay::new(this.duration_to_next_find_query));
-					},
-					false => return Poll::Ready(Some(DiscoveryEvent::QueryFailed { query_id })),
+				if let Some(instant) = this.find_node_queries.remove(&query_id) {
+					this.duration_to_next_find_query = KADEMLIA_QUERY_INTERVAL;
+					this.next_kad_query = Some(Delay::new(this.duration_to_next_find_query));
+
+					log::debug!(target: LOG_TARGET, "dht random walk failed for {query_id:?} in {:?}", instant.elapsed());
 				}
+
+				return Poll::Ready(Some(DiscoveryEvent::QueryFailed { query_id }));
 			},
 			Poll::Ready(Some(KademliaEvent::IncomingRecord { record })) => {
 				log::trace!(

substrate/client/network/src/litep2p/mod.rs

@@ -259,6 +259,17 @@ impl Litep2pNetworkBackend {
 		Ok((local_identity, local_peer_id))
 	}
 
+	/// Fetch the number of connected peers from the peerset handle and update
+	/// the atomic `num_connected` shared between the network backend.
+	fn fetch_connected_peers(&self) -> usize {
+		let num_connected_peers = self
+			.peerset_handles
+			.get(&self.block_announce_protocol)
+			.map_or(0usize, |handle| handle.connected_peers.load(Ordering::Relaxed));
+		self.num_connected.store(num_connected_peers, Ordering::Relaxed);
+		num_connected_peers
+	}
+
 	/// Configure transport protocols for `Litep2pNetworkBackend`.
 	fn configure_transport<B: BlockT + 'static, H: ExHashT>(
 		config: &FullNetworkConfiguration<B, H, Self>,
@@ -439,6 +450,9 @@ impl<B: BlockT + 'static, H: ExHashT> NetworkBackend<B, H> for Litep2pNetworkBac
 		let peer_store_handle = params.network_config.peer_store_handle();
 		let executor = Arc::new(Litep2pExecutor { executor: params.executor });
 
+		let limit_discovery_under =
+			params.network_config.network_config.default_peers_set.out_peers as usize + 15;
+
 		let FullNetworkConfiguration {
 			notification_protocols,
 			request_response_protocols,
@@ -541,6 +555,8 @@ impl<B: BlockT + 'static, H: ExHashT> NetworkBackend<B, H> for Litep2pNetworkBac
 
 		// enable ipfs ping, identify and kademlia, and potentially mdns if user enabled it
 		let listen_addresses = Arc::new(Default::default());
+		let num_connected = Arc::new(Default::default());
+
 		let (discovery, ping_config, identify_config, kademlia_config, maybe_mdns_config) =
 			Discovery::new(
 				&network_config,
@@ -549,6 +565,8 @@ impl<B: BlockT + 'static, H: ExHashT> NetworkBackend<B, H> for Litep2pNetworkBac
 				&params.protocol_id,
 				known_addresses.clone(),
 				Arc::clone(&listen_addresses),
+				Arc::clone(&num_connected),
+				limit_discovery_under,
 				Arc::clone(&peer_store_handle),
 			);
 
@@ -592,7 +610,6 @@ impl<B: BlockT + 'static, H: ExHashT> NetworkBackend<B, H> for Litep2pNetworkBac
 		));
 
 		// register rest of the metrics now that `Litep2p` has been created
-		let num_connected = Arc::new(Default::default());
 		let bandwidth: Arc<dyn BandwidthSink> =
 			Arc::new(Litep2pBandwidthSink { sink: litep2p.bandwidth_sink() });
 
@@ -682,11 +699,7 @@ impl<B: BlockT + 'static, H: ExHashT> NetworkBackend<B, H> for Litep2pNetworkBac
 		log::debug!(target: LOG_TARGET, "starting litep2p network backend");
 
 		loop {
-			let num_connected_peers = self
-				.peerset_handles
-				.get(&self.block_announce_protocol)
-				.map_or(0usize, |handle| handle.connected_peers.load(Ordering::Relaxed));
-			self.num_connected.store(num_connected_peers, Ordering::Relaxed);
+			self.fetch_connected_peers();
 
 			tokio::select! {
 				command = self.cmd_rx.next() => match command {
@@ -707,11 +720,10 @@ impl<B: BlockT + 'static, H: ExHashT> NetworkBackend<B, H> for Litep2pNetworkBac
 							self.event_streams.push(tx);
 						}
 						NetworkServiceCommand::Status { tx } => {
+							let num_connected_peers = self.fetch_connected_peers();
+
 							let _ = tx.send(NetworkStatus {
-								num_connected_peers: self
-									.peerset_handles
-									.get(&self.block_announce_protocol)
-									.map_or(0usize, |handle| handle.connected_peers.load(Ordering::Relaxed)),
+								num_connected_peers,
 								total_bytes_inbound: self.litep2p.bandwidth_sink().inbound() as u64,
 								total_bytes_outbound: self.litep2p.bandwidth_sink().outbound() as u64,
 							});