Extract common part of relay loops (#660)

* extract common parts of relay loops: begin

* merge client impls

* backoff in exchange loop

* reconnect without clone

bridges/relays/ethereum-client/src/client.rs

@@ -28,22 +28,32 @@ use jsonrpsee::Client as RpcClient;
 /// The client used to interact with an Ethereum node through RPC.
 #[derive(Clone)]
 pub struct Client {
+	params: ConnectionParams,
 	client: RpcClient,
 }
 
 impl Client {
 	/// Create a new Ethereum RPC Client.
 	pub fn new(params: ConnectionParams) -> Self {
+		Self {
+			client: Self::build_client(&params),
+			params,
+		}
+	}
+
+	/// Build client to use in connection.
+	fn build_client(params: &ConnectionParams) -> RpcClient {
 		let uri = format!("http://{}:{}", params.host, params.port);
 		let transport = HttpTransportClient::new(&uri);
 		let raw_client = RawClient::new(transport);
-		let client: RpcClient = raw_client.into();
-
-		Self { client }
+		raw_client.into()
+	}
+
+	/// Reopen client connection.
+	pub fn reconnect(&mut self) {
+		self.client = Self::build_client(&self.params);
 	}
-}
 
-impl Client {
 	/// Estimate gas usage for the given call.
 	pub async fn estimate_gas(&self, call_request: CallRequest) -> Result<U256> {
 		Ok(Ethereum::estimate_gas(&self.client, call_request).await?)

bridges/relays/ethereum/src/ethereum_exchange.rs

@@ -39,7 +39,7 @@ use relay_rialto_client::{Rialto, SigningParams as RialtoSigningParams};
 use relay_substrate_client::{
 	Chain as SubstrateChain, Client as SubstrateClient, ConnectionParams as SubstrateConnectionParams,
 };
-use relay_utils::{metrics::MetricsParams, HeaderId};
+use relay_utils::{metrics::MetricsParams, relay_loop::Client as RelayClient, HeaderId};
 use rialto_runtime::exchange::EthereumTransactionInclusionProof;
 use std::{sync::Arc, time::Duration};
 
@@ -120,19 +120,28 @@ impl SourceTransaction for EthereumSourceTransaction {
 }
 
 /// Ethereum node as transactions proof source.
+#[derive(Clone)]
 struct EthereumTransactionsSource {
 	client: EthereumClient,
 }
 
 #[async_trait]
-impl SourceClient<EthereumToSubstrateExchange> for EthereumTransactionsSource {
+impl RelayClient for EthereumTransactionsSource {
 	type Error = RpcError;
 
+	async fn reconnect(&mut self) -> Result<(), RpcError> {
+		self.client.reconnect();
+		Ok(())
+	}
+}
+
+#[async_trait]
+impl SourceClient<EthereumToSubstrateExchange> for EthereumTransactionsSource {
 	async fn tick(&self) {
 		async_std::task::sleep(ETHEREUM_TICK_INTERVAL).await;
 	}
 
-	async fn block_by_hash(&self, hash: H256) -> Result<EthereumSourceBlock, Self::Error> {
+	async fn block_by_hash(&self, hash: H256) -> Result<EthereumSourceBlock, RpcError> {
 		self.client
 			.header_by_hash_with_transactions(hash)
 			.await
@@ -140,7 +149,7 @@ impl SourceClient<EthereumToSubstrateExchange> for EthereumTransactionsSource {
 			.map_err(Into::into)
 	}
 
-	async fn block_by_number(&self, number: u64) -> Result<EthereumSourceBlock, Self::Error> {
+	async fn block_by_number(&self, number: u64) -> Result<EthereumSourceBlock, RpcError> {
 		self.client
 			.header_by_number_with_transactions(number)
 			.await
@@ -151,7 +160,7 @@ impl SourceClient<EthereumToSubstrateExchange> for EthereumTransactionsSource {
 	async fn transaction_block(
 		&self,
 		hash: &EthereumTransactionHash,
-	) -> Result<Option<(EthereumHeaderId, usize)>, Self::Error> {
+	) -> Result<Option<(EthereumHeaderId, usize)>, RpcError> {
 		let eth_tx = match self.client.transaction_by_hash(*hash).await? {
 			Some(eth_tx) => eth_tx,
 			None => return Ok(None),
@@ -173,7 +182,7 @@ impl SourceClient<EthereumToSubstrateExchange> for EthereumTransactionsSource {
 		&self,
 		block: &EthereumSourceBlock,
 		tx_index: usize,
-	) -> Result<EthereumTransactionInclusionProof, Self::Error> {
+	) -> Result<EthereumTransactionInclusionProof, RpcError> {
 		const TRANSACTION_HAS_RAW_FIELD_PROOF: &str = "RPC level checks that transactions from Ethereum\
 			node are having `raw` field; qed";
 		const BLOCK_HAS_HASH_FIELD_PROOF: &str = "RPC level checks that block has `hash` field; qed";
@@ -199,6 +208,7 @@ impl SourceClient<EthereumToSubstrateExchange> for EthereumTransactionsSource {
 }
 
 /// Substrate node as transactions proof target.
+#[derive(Clone)]
 struct SubstrateTransactionsTarget {
 	client: SubstrateClient<Rialto>,
 	sign_params: RialtoSigningParams,
@@ -206,18 +216,25 @@ struct SubstrateTransactionsTarget {
 }
 
 #[async_trait]
-impl TargetClient<EthereumToSubstrateExchange> for SubstrateTransactionsTarget {
+impl RelayClient for SubstrateTransactionsTarget {
 	type Error = RpcError;
 
+	async fn reconnect(&mut self) -> Result<(), RpcError> {
+		Ok(self.client.reconnect().await?)
+	}
+}
+
+#[async_trait]
+impl TargetClient<EthereumToSubstrateExchange> for SubstrateTransactionsTarget {
 	async fn tick(&self) {
 		async_std::task::sleep(Rialto::AVERAGE_BLOCK_INTERVAL).await;
 	}
 
-	async fn is_header_known(&self, id: &EthereumHeaderId) -> Result<bool, Self::Error> {
+	async fn is_header_known(&self, id: &EthereumHeaderId) -> Result<bool, RpcError> {
 		self.client.ethereum_header_known(*id).await
 	}
 
-	async fn is_header_finalized(&self, id: &EthereumHeaderId) -> Result<bool, Self::Error> {
+	async fn is_header_finalized(&self, id: &EthereumHeaderId) -> Result<bool, RpcError> {
 		// we check if header is finalized by simple comparison of the header number and
 		// number of best finalized PoA header known to Substrate node.
 		//
@@ -230,11 +247,11 @@ impl TargetClient<EthereumToSubstrateExchange> for SubstrateTransactionsTarget {
 		Ok(id.0 <= best_finalized_ethereum_block.0)
 	}
 
-	async fn best_finalized_header_id(&self) -> Result<EthereumHeaderId, Self::Error> {
+	async fn best_finalized_header_id(&self) -> Result<EthereumHeaderId, RpcError> {
 		self.client.best_ethereum_finalized_block().await
 	}
 
-	async fn filter_transaction_proof(&self, proof: &EthereumTransactionInclusionProof) -> Result<bool, Self::Error> {
+	async fn filter_transaction_proof(&self, proof: &EthereumTransactionInclusionProof) -> Result<bool, RpcError> {
 		// let's try to parse transaction locally
 		let (raw_tx, raw_tx_receipt) = &proof.proof[proof.index as usize];
 		let parse_result = rialto_runtime::exchange::EthTransaction::parse(raw_tx);
@@ -253,7 +270,7 @@ impl TargetClient<EthereumToSubstrateExchange> for SubstrateTransactionsTarget {
 		self.client.verify_exchange_transaction_proof(proof.clone()).await
 	}
 
-	async fn submit_transaction_proof(&self, proof: EthereumTransactionInclusionProof) -> Result<(), Self::Error> {
+	async fn submit_transaction_proof(&self, proof: EthereumTransactionInclusionProof) -> Result<(), RpcError> {
 		let (sign_params, bridge_instance) = (self.sign_params.clone(), self.bridge_instance.clone());
 		self.client
 			.submit_exchange_transaction_proof(sign_params, bridge_instance, proof)

bridges/relays/ethereum/src/ethereum_sync_loop.rs

@@ -37,7 +37,7 @@ use relay_rialto_client::{Rialto, SigningParams as RialtoSigningParams};
 use relay_substrate_client::{
 	Chain as SubstrateChain, Client as SubstrateClient, ConnectionParams as SubstrateConnectionParams,
 };
-use relay_utils::metrics::MetricsParams;
+use relay_utils::{metrics::MetricsParams, relay_loop::Client as RelayClient};
 
 use std::fmt::Debug;
 use std::{collections::HashSet, sync::Arc, time::Duration};
@@ -105,6 +105,7 @@ impl HeadersSyncPipeline for EthereumHeadersSyncPipeline {
 pub type QueuedEthereumHeader = QueuedHeader<EthereumHeadersSyncPipeline>;
 
 /// Ethereum client as headers source.
+#[derive(Clone)]
 struct EthereumHeadersSource {
 	/// Ethereum node client.
 	client: EthereumClient,
@@ -117,14 +118,22 @@ impl EthereumHeadersSource {
 }
 
 #[async_trait]
-impl SourceClient<EthereumHeadersSyncPipeline> for EthereumHeadersSource {
+impl RelayClient for EthereumHeadersSource {
 	type Error = RpcError;
 
-	async fn best_block_number(&self) -> Result<u64, Self::Error> {
+	async fn reconnect(&mut self) -> Result<(), RpcError> {
+		self.client.reconnect();
+		Ok(())
+	}
+}
+
+#[async_trait]
+impl SourceClient<EthereumHeadersSyncPipeline> for EthereumHeadersSource {
+	async fn best_block_number(&self) -> Result<u64, RpcError> {
 		self.client.best_block_number().await.map_err(Into::into)
 	}
 
-	async fn header_by_hash(&self, hash: HeaderHash) -> Result<Header, Self::Error> {
+	async fn header_by_hash(&self, hash: HeaderHash) -> Result<Header, RpcError> {
 		self.client
 			.header_by_hash(hash)
 			.await
@@ -132,7 +141,7 @@ impl SourceClient<EthereumHeadersSyncPipeline> for EthereumHeadersSource {
 			.map_err(Into::into)
 	}
 
-	async fn header_by_number(&self, number: u64) -> Result<Header, Self::Error> {
+	async fn header_by_number(&self, number: u64) -> Result<Header, RpcError> {
 		self.client
 			.header_by_number(number)
 			.await
@@ -140,7 +149,7 @@ impl SourceClient<EthereumHeadersSyncPipeline> for EthereumHeadersSource {
 			.map_err(Into::into)
 	}
 
-	async fn header_completion(&self, id: EthereumHeaderId) -> Result<(EthereumHeaderId, Option<()>), Self::Error> {
+	async fn header_completion(&self, id: EthereumHeaderId) -> Result<(EthereumHeaderId, Option<()>), RpcError> {
 		Ok((id, None))
 	}
 
@@ -148,13 +157,14 @@ impl SourceClient<EthereumHeadersSyncPipeline> for EthereumHeadersSource {
 		&self,
 		id: EthereumHeaderId,
 		header: QueuedEthereumHeader,
-	) -> Result<(EthereumHeaderId, Vec<Receipt>), Self::Error> {
+	) -> Result<(EthereumHeaderId, Vec<Receipt>), RpcError> {
 		self.client
 			.transaction_receipts(id, header.header().transactions.clone())
 			.await
 	}
 }
 
+#[derive(Clone)]
 struct SubstrateHeadersTarget {
 	/// Substrate node client.
 	client: SubstrateClient<Rialto>,
@@ -183,21 +193,25 @@ impl SubstrateHeadersTarget {
 }
 
 #[async_trait]
-impl TargetClient<EthereumHeadersSyncPipeline> for SubstrateHeadersTarget {
+impl RelayClient for SubstrateHeadersTarget {
 	type Error = RpcError;
 
-	async fn best_header_id(&self) -> Result<EthereumHeaderId, Self::Error> {
+	async fn reconnect(&mut self) -> Result<(), RpcError> {
+		Ok(self.client.reconnect().await?)
+	}
+}
+
+#[async_trait]
+impl TargetClient<EthereumHeadersSyncPipeline> for SubstrateHeadersTarget {
+	async fn best_header_id(&self) -> Result<EthereumHeaderId, RpcError> {
 		self.client.best_ethereum_block().await
 	}
 
-	async fn is_known_header(&self, id: EthereumHeaderId) -> Result<(EthereumHeaderId, bool), Self::Error> {
+	async fn is_known_header(&self, id: EthereumHeaderId) -> Result<(EthereumHeaderId, bool), RpcError> {
 		Ok((id, self.client.ethereum_header_known(id).await?))
 	}
 
-	async fn submit_headers(
-		&self,
-		headers: Vec<QueuedEthereumHeader>,
-	) -> SubmittedHeaders<EthereumHeaderId, Self::Error> {
+	async fn submit_headers(&self, headers: Vec<QueuedEthereumHeader>) -> SubmittedHeaders<EthereumHeaderId, RpcError> {
 		let (sign_params, bridge_instance, sign_transactions) = (
 			self.sign_params.clone(),
 			self.bridge_instance.clone(),
@@ -208,16 +222,16 @@ impl TargetClient<EthereumHeadersSyncPipeline> for SubstrateHeadersTarget {
 			.await
 	}
 
-	async fn incomplete_headers_ids(&self) -> Result<HashSet<EthereumHeaderId>, Self::Error> {
+	async fn incomplete_headers_ids(&self) -> Result<HashSet<EthereumHeaderId>, RpcError> {
 		Ok(HashSet::new())
 	}
 
 	#[allow(clippy::unit_arg)]
-	async fn complete_header(&self, id: EthereumHeaderId, _completion: ()) -> Result<EthereumHeaderId, Self::Error> {
+	async fn complete_header(&self, id: EthereumHeaderId, _completion: ()) -> Result<EthereumHeaderId, RpcError> {
 		Ok(id)
 	}
 
-	async fn requires_extra(&self, header: QueuedEthereumHeader) -> Result<(EthereumHeaderId, bool), Self::Error> {
+	async fn requires_extra(&self, header: QueuedEthereumHeader) -> Result<(EthereumHeaderId, bool), RpcError> {
 		// we can minimize number of receipts_check calls by checking header
 		// logs bloom here, but it may give us false positives (when authorities
 		// source is contract, we never need any logs)

bridges/relays/ethereum/src/substrate_sync_loop.rs

@@ -35,7 +35,7 @@ use relay_substrate_client::{
 	headers_source::HeadersSource, Chain as SubstrateChain, Client as SubstrateClient,
 	ConnectionParams as SubstrateConnectionParams,
 };
-use relay_utils::metrics::MetricsParams;
+use relay_utils::{metrics::MetricsParams, relay_loop::Client as RelayClient};
 use sp_runtime::Justification;
 
 use std::fmt::Debug;
@@ -98,6 +98,7 @@ pub type QueuedRialtoHeader = QueuedHeader<SubstrateHeadersSyncPipeline>;
 type SubstrateHeadersSource = HeadersSource<Rialto, SubstrateHeadersSyncPipeline>;
 
 /// Ethereum client as Substrate headers target.
+#[derive(Clone)]
 struct EthereumHeadersTarget {
 	/// Ethereum node client.
 	client: EthereumClient,
@@ -118,38 +119,42 @@ impl EthereumHeadersTarget {
 }
 
 #[async_trait]
-impl TargetClient<SubstrateHeadersSyncPipeline> for EthereumHeadersTarget {
+impl RelayClient for EthereumHeadersTarget {
 	type Error = RpcError;
 
-	async fn best_header_id(&self) -> Result<RialtoHeaderId, Self::Error> {
+	async fn reconnect(&mut self) -> Result<(), RpcError> {
+		self.client.reconnect();
+		Ok(())
+	}
+}
+
+#[async_trait]
+impl TargetClient<SubstrateHeadersSyncPipeline> for EthereumHeadersTarget {
+	async fn best_header_id(&self) -> Result<RialtoHeaderId, RpcError> {
 		self.client.best_substrate_block(self.contract).await
 	}
 
-	async fn is_known_header(&self, id: RialtoHeaderId) -> Result<(RialtoHeaderId, bool), Self::Error> {
+	async fn is_known_header(&self, id: RialtoHeaderId) -> Result<(RialtoHeaderId, bool), RpcError> {
 		self.client.substrate_header_known(self.contract, id).await
 	}
 
-	async fn submit_headers(&self, headers: Vec<QueuedRialtoHeader>) -> SubmittedHeaders<RialtoHeaderId, Self::Error> {
+	async fn submit_headers(&self, headers: Vec<QueuedRialtoHeader>) -> SubmittedHeaders<RialtoHeaderId, RpcError> {
 		self.client
 			.submit_substrate_headers(self.sign_params.clone(), self.contract, headers)
 			.await
 	}
 
-	async fn incomplete_headers_ids(&self) -> Result<HashSet<RialtoHeaderId>, Self::Error> {
+	async fn incomplete_headers_ids(&self) -> Result<HashSet<RialtoHeaderId>, RpcError> {
 		self.client.incomplete_substrate_headers(self.contract).await
 	}
 
-	async fn complete_header(
-		&self,
-		id: RialtoHeaderId,
-		completion: Justification,
-	) -> Result<RialtoHeaderId, Self::Error> {
+	async fn complete_header(&self, id: RialtoHeaderId, completion: Justification) -> Result<RialtoHeaderId, RpcError> {
 		self.client
 			.complete_substrate_header(self.sign_params.clone(), self.contract, id, completion)
 			.await
 	}
 
-	async fn requires_extra(&self, header: QueuedRialtoHeader) -> Result<(RialtoHeaderId, bool), Self::Error> {
+	async fn requires_extra(&self, header: QueuedRialtoHeader) -> Result<(RialtoHeaderId, bool), RpcError> {
 		Ok((header.header().id(), false))
 	}
 }

bridges/relays/exchange-relay/src/exchange.rs

@@ -17,7 +17,9 @@
 //! Relaying proofs of exchange transaction.
 
 use async_trait::async_trait;
-use relay_utils::{MaybeConnectionError, StringifiedMaybeConnectionError};
+use relay_utils::{
+	relay_loop::Client as RelayClient, FailedClient, MaybeConnectionError, StringifiedMaybeConnectionError,
+};
 use std::{
 	fmt::{Debug, Display},
 	string::ToString,
@@ -84,10 +86,7 @@ pub type HeaderId<P> = relay_utils::HeaderId<BlockHashOf<P>, BlockNumberOf<P>>;
 
 /// Source client API.
 #[async_trait]
-pub trait SourceClient<P: TransactionProofPipeline> {
-	/// Error type.
-	type Error: Debug + MaybeConnectionError;
-
+pub trait SourceClient<P: TransactionProofPipeline>: RelayClient {
 	/// Sleep until exchange-related data is (probably) updated.
 	async fn tick(&self);
 	/// Get block by hash.
@@ -104,10 +103,7 @@ pub trait SourceClient<P: TransactionProofPipeline> {
 
 /// Target client API.
 #[async_trait]
-pub trait TargetClient<P: TransactionProofPipeline> {
-	/// Error type.
-	type Error: Debug + MaybeConnectionError;
-
+pub trait TargetClient<P: TransactionProofPipeline>: RelayClient {
 	/// Sleep until exchange-related data is (probably) updated.
 	async fn tick(&self);
 	/// Returns `Ok(true)` if header is known to the target node.
@@ -146,7 +142,7 @@ pub async fn relay_block_transactions<P: TransactionProofPipeline>(
 	target_client: &impl TargetClient<P>,
 	source_block: &P::Block,
 	mut relayed_transactions: RelayedBlockTransactions,
-) -> Result<RelayedBlockTransactions, RelayedBlockTransactions> {
+) -> Result<RelayedBlockTransactions, (FailedClient, RelayedBlockTransactions)> {
 	let transactions_to_process = source_block
 		.transactions()
 		.into_iter()
@@ -156,16 +152,21 @@ pub async fn relay_block_transactions<P: TransactionProofPipeline>(
 		let result = async {
 			let source_tx_id = format!("{}/{}", source_block.id().1, source_tx_index);
 			let source_tx_proof =
-				prepare_transaction_proof(source_client, &source_tx_id, source_block, source_tx_index).await?;
+				prepare_transaction_proof(source_client, &source_tx_id, source_block, source_tx_index)
+					.await
+					.map_err(|e| (FailedClient::Source, e))?;
 
 			let needs_to_be_relayed =
 				target_client
 					.filter_transaction_proof(&source_tx_proof)
 					.await
 					.map_err(|err| {
+						(
+							FailedClient::Target,
 							StringifiedMaybeConnectionError::new(
 								err.is_connection_error(),
 								format!("Transaction filtering has failed with {:?}", err),
+							),
 						)
 					})?;
 
@@ -176,6 +177,7 @@ pub async fn relay_block_transactions<P: TransactionProofPipeline>(
 			relay_ready_transaction_proof(target_client, &source_tx_id, source_tx_proof)
 				.await
 				.map(|_| true)
+				.map_err(|e| (FailedClient::Target, e))
 		}
 		.await;
 
@@ -205,7 +207,7 @@ pub async fn relay_block_transactions<P: TransactionProofPipeline>(
 				relayed_transactions.processed += 1;
 				relayed_transactions.relayed += 1;
 			}
-			Err(err) => {
+			Err((failed_client, err)) => {
 				log::error!(
 					target: "bridge",
 					"Error relaying {} transaction {} proof to {} node: {}. {}",
@@ -221,7 +223,7 @@ pub async fn relay_block_transactions<P: TransactionProofPipeline>(
 				);
 
 				if err.is_connection_error() {
-					return Err(relayed_transactions);
+					return Err((failed_client, relayed_transactions));
 				}
 
 				relayed_transactions.processed += 1;
@@ -529,8 +531,9 @@ pub(crate) mod tests {
 	#[derive(Debug, Clone, PartialEq)]
 	pub struct TestTransactionProof(pub TestTransactionHash);
 
+	#[derive(Clone)]
 	pub struct TestTransactionsSource {
-		pub on_tick: Box<dyn Fn(&mut TestTransactionsSourceData) + Send + Sync>,
+		pub on_tick: Arc<dyn Fn(&mut TestTransactionsSourceData) + Send + Sync>,
 		pub data: Arc<Mutex<TestTransactionsSourceData>>,
 	}
 
@@ -543,7 +546,7 @@ pub(crate) mod tests {
 	impl TestTransactionsSource {
 		pub fn new(on_tick: Box<dyn Fn(&mut TestTransactionsSourceData) + Send + Sync>) -> Self {
 			Self {
-				on_tick,
+				on_tick: Arc::new(on_tick),
 				data: Arc::new(Mutex::new(TestTransactionsSourceData {
 					block: Ok(test_block()),
 					transaction_block: Ok(Some((test_block_id(), 0))),
@@ -554,9 +557,16 @@ pub(crate) mod tests {
 	}
 
 	#[async_trait]
-	impl SourceClient<TestTransactionProofPipeline> for TestTransactionsSource {
+	impl RelayClient for TestTransactionsSource {
 		type Error = TestError;
 
+		async fn reconnect(&mut self) -> Result<(), TestError> {
+			Ok(())
+		}
+	}
+
+	#[async_trait]
+	impl SourceClient<TestTransactionProofPipeline> for TestTransactionsSource {
 		async fn tick(&self) {
 			(self.on_tick)(&mut *self.data.lock())
 		}
@@ -584,8 +594,9 @@ pub(crate) mod tests {
 		}
 	}
 
+	#[derive(Clone)]
 	pub struct TestTransactionsTarget {
-		pub on_tick: Box<dyn Fn(&mut TestTransactionsTargetData) + Send + Sync>,
+		pub on_tick: Arc<dyn Fn(&mut TestTransactionsTargetData) + Send + Sync>,
 		pub data: Arc<Mutex<TestTransactionsTargetData>>,
 	}
 
@@ -600,7 +611,7 @@ pub(crate) mod tests {
 	impl TestTransactionsTarget {
 		pub fn new(on_tick: Box<dyn Fn(&mut TestTransactionsTargetData) + Send + Sync>) -> Self {
 			Self {
-				on_tick,
+				on_tick: Arc::new(on_tick),
 				data: Arc::new(Mutex::new(TestTransactionsTargetData {
 					is_header_known: Ok(true),
 					is_header_finalized: Ok(true),
@@ -613,9 +624,16 @@ pub(crate) mod tests {
 	}
 
 	#[async_trait]
-	impl TargetClient<TestTransactionProofPipeline> for TestTransactionsTarget {
+	impl RelayClient for TestTransactionsTarget {
 		type Error = TestError;
 
+		async fn reconnect(&mut self) -> Result<(), TestError> {
+			Ok(())
+		}
+	}
+
+	#[async_trait]
+	impl TargetClient<TestTransactionProofPipeline> for TestTransactionsTarget {
 		async fn tick(&self) {
 			(self.on_tick)(&mut *self.data.lock())
 		}
@@ -784,6 +802,7 @@ pub(crate) mod tests {
 			),
 			pre_relayed,
 		))
+		.map_err(|(_, transactions)| transactions)
 	}
 
 	#[test]

bridges/relays/exchange-relay/src/exchange_loop.rs

@@ -27,13 +27,9 @@ use futures::{future::FutureExt, select};
 use num_traits::One;
 use relay_utils::{
 	metrics::{start as metrics_start, GlobalMetrics, MetricsParams},
-	retry_backoff,
+	retry_backoff, FailedClient, MaybeConnectionError,
 };
-use std::{future::Future, time::Duration};
-
-/// Delay after connection-related error happened before we'll try
-/// reconnection again.
-const CONNECTION_ERROR_DELAY: Duration = Duration::from_secs(10);
+use std::future::Future;
 
 /// Transactions proofs relay state.
 #[derive(Debug)]
@@ -43,7 +39,7 @@ pub struct TransactionProofsRelayState<BlockNumber> {
 }
 
 /// Transactions proofs relay storage.
-pub trait TransactionProofsRelayStorage {
+pub trait TransactionProofsRelayStorage: Clone {
 	/// Associated block number.
 	type BlockNumber;
 
@@ -54,7 +50,7 @@ pub trait TransactionProofsRelayStorage {
 }
 
 /// In-memory storage for auto-relay loop.
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 pub struct InMemoryStorage<BlockNumber> {
 	best_processed_header_number: BlockNumber,
 }
@@ -84,21 +80,15 @@ impl<BlockNumber: Clone + Copy> TransactionProofsRelayStorage for InMemoryStorag
 
 /// Run proofs synchronization.
 pub fn run<P: TransactionProofPipeline>(
-	mut storage: impl TransactionProofsRelayStorage<BlockNumber = BlockNumberOf<P>>,
+	storage: impl TransactionProofsRelayStorage<BlockNumber = BlockNumberOf<P>>,
 	source_client: impl SourceClient<P>,
 	target_client: impl TargetClient<P>,
 	metrics_params: Option<MetricsParams>,
 	exit_signal: impl Future<Output = ()>,
 ) {
-	let mut local_pool = futures::executor::LocalPool::new();
-
-	local_pool.run_until(async move {
-		let mut retry_backoff = retry_backoff();
-		let mut state = storage.state();
-		let mut current_finalized_block = None;
-
-		let mut metrics_global = GlobalMetrics::default();
-		let mut metrics_exch = ExchangeLoopMetrics::default();
+	let exit_signal = exit_signal.shared();
+	let metrics_global = GlobalMetrics::default();
+	let metrics_exch = ExchangeLoopMetrics::default();
 	let metrics_enabled = metrics_params.is_some();
 	metrics_start(
 		format!("{}_to_{}_Exchange", P::SOURCE_NAME, P::TARGET_NAME),
@@ -107,6 +97,44 @@ pub fn run<P: TransactionProofPipeline>(
 		&metrics_exch,
 	);
 
+	relay_utils::relay_loop::run(
+		relay_utils::relay_loop::RECONNECT_DELAY,
+		source_client,
+		target_client,
+		|source_client, target_client| {
+			run_until_connection_lost(
+				storage.clone(),
+				source_client,
+				target_client,
+				if metrics_enabled {
+					Some(metrics_global.clone())
+				} else {
+					None
+				},
+				if metrics_enabled {
+					Some(metrics_exch.clone())
+				} else {
+					None
+				},
+				exit_signal.clone(),
+			)
+		},
+	);
+}
+
+/// Run proofs synchronization.
+async fn run_until_connection_lost<P: TransactionProofPipeline>(
+	mut storage: impl TransactionProofsRelayStorage<BlockNumber = BlockNumberOf<P>>,
+	source_client: impl SourceClient<P>,
+	target_client: impl TargetClient<P>,
+	metrics_global: Option<GlobalMetrics>,
+	metrics_exch: Option<ExchangeLoopMetrics>,
+	exit_signal: impl Future<Output = ()>,
+) -> Result<(), FailedClient> {
+	let mut retry_backoff = retry_backoff();
+	let mut state = storage.state();
+	let mut current_finalized_block = None;
+
 	let exit_signal = exit_signal.fuse();
 
 	futures::pin_mut!(exit_signal);
@@ -118,34 +146,35 @@ pub fn run<P: TransactionProofPipeline>(
 			&target_client,
 			&mut state,
 			&mut current_finalized_block,
-				if metrics_enabled { Some(&mut metrics_exch) } else { None },
+			metrics_exch.as_ref(),
 		)
 		.await;
 
-			if metrics_enabled {
-				metrics_global.update();
+		if let Some(ref metrics_global) = metrics_global {
+			metrics_global.update().await;
 		}
 
-			match iteration_result {
-				Ok(_) => {
+		if let Err((is_connection_error, failed_client)) = iteration_result {
+			if is_connection_error {
+				return Err(failed_client);
+			}
+
+			let retry_timeout = retry_backoff
+				.next_backoff()
+				.unwrap_or(relay_utils::relay_loop::RECONNECT_DELAY);
+			select! {
+				_ = async_std::task::sleep(retry_timeout).fuse() => {},
+				_ = exit_signal => return Ok(()),
+			}
+		} else {
 			retry_backoff.reset();
 
 			select! {
 				_ = source_client.tick().fuse() => {},
-						_ = exit_signal => return,
-					}
-				}
-				Err(_) => {
-					let retry_timeout = retry_backoff.next_backoff().unwrap_or(CONNECTION_ERROR_DELAY);
-
-					select! {
-						_ = async_std::task::sleep(retry_timeout).fuse() => {},
-						_ = exit_signal => return,
+				_ = exit_signal => return Ok(()),
 			}
 		}
 	}
-		}
-	});
 }
 
 /// Run exchange loop until we need to break.
@@ -155,8 +184,8 @@ async fn run_loop_iteration<P: TransactionProofPipeline>(
 	target_client: &impl TargetClient<P>,
 	state: &mut TransactionProofsRelayState<BlockNumberOf<P>>,
 	current_finalized_block: &mut Option<(P::Block, RelayedBlockTransactions)>,
-	mut exchange_loop_metrics: Option<&mut ExchangeLoopMetrics>,
-) -> Result<(), ()> {
+	exchange_loop_metrics: Option<&ExchangeLoopMetrics>,
+) -> Result<(), (bool, FailedClient)> {
 	let best_finalized_header_id = match target_client.best_finalized_header_id().await {
 		Ok(best_finalized_header_id) => {
 			log::debug!(
@@ -178,7 +207,7 @@ async fn run_loop_iteration<P: TransactionProofPipeline>(
 				err,
 			);
 
-			return Err(());
+			return Err((err.is_connection_error(), FailedClient::Target));
 		}
 	};
 
@@ -202,7 +231,7 @@ async fn run_loop_iteration<P: TransactionProofPipeline>(
 					state.best_processed_header_number = state.best_processed_header_number + One::one();
 					storage.set_state(state);
 
-					if let Some(exchange_loop_metrics) = exchange_loop_metrics.as_mut() {
+					if let Some(ref exchange_loop_metrics) = exchange_loop_metrics {
 						exchange_loop_metrics.update::<P>(
 							state.best_processed_header_number,
 							best_finalized_header_id.0,
@@ -212,9 +241,9 @@ async fn run_loop_iteration<P: TransactionProofPipeline>(
 
 					// we have just updated state => proceed to next block retrieval
 				}
-				Err(relayed_transactions) => {
+				Err((failed_client, relayed_transactions)) => {
 					*current_finalized_block = Some((block, relayed_transactions));
-					return Err(());
+					return Err((true, failed_client));
 				}
 			}
 		}
@@ -240,7 +269,7 @@ async fn run_loop_iteration<P: TransactionProofPipeline>(
 						err,
 					);
 
-					return Err(());
+					return Err((err.is_connection_error(), FailedClient::Source));
 				}
 			}
 		}

bridges/relays/exchange-relay/src/exchange_loop_metrics.rs

@@ -20,6 +20,7 @@ use crate::exchange::{BlockNumberOf, RelayedBlockTransactions, TransactionProofP
 use relay_utils::metrics::{register, Counter, CounterVec, GaugeVec, Metrics, Opts, Registry, U64};
 
 /// Exchange transactions relay metrics.
+#[derive(Clone)]
 pub struct ExchangeLoopMetrics {
 	/// Best finalized block numbers - "processed" and "known".
 	best_block_numbers: GaugeVec<U64>,
@@ -60,7 +61,7 @@ impl Default for ExchangeLoopMetrics {
 impl ExchangeLoopMetrics {
 	/// Update metrics when single block is relayed.
 	pub fn update<P: TransactionProofPipeline>(
-		&mut self,
+		&self,
 		best_processed_block_number: BlockNumberOf<P>,
 		best_known_block_number: BlockNumberOf<P>,
 		relayed_transactions: RelayedBlockTransactions,

bridges/relays/headers-relay/src/sync_loop.rs

@@ -26,7 +26,9 @@ use num_traits::{Saturating, Zero};
 use relay_utils::{
 	format_ids, interval,
 	metrics::{start as metrics_start, GlobalMetrics, MetricsParams},
-	process_future_result, retry_backoff, MaybeConnectionError, StringifiedMaybeConnectionError,
+	process_future_result,
+	relay_loop::Client as RelayClient,
+	retry_backoff, FailedClient, MaybeConnectionError, StringifiedMaybeConnectionError,
 };
 use std::{
 	collections::HashSet,
@@ -53,10 +55,7 @@ const MAINTAIN_INTERVAL: Duration = Duration::from_secs(30);
 
 /// Source client trait.
 #[async_trait]
-pub trait SourceClient<P: HeadersSyncPipeline> {
-	/// Type of error this clients returns.
-	type Error: std::fmt::Debug + MaybeConnectionError;
-
+pub trait SourceClient<P: HeadersSyncPipeline>: RelayClient {
 	/// Get best block number.
 	async fn best_block_number(&self) -> Result<P::Number, Self::Error>;
 
@@ -80,10 +79,7 @@ pub trait SourceClient<P: HeadersSyncPipeline> {
 
 /// Target client trait.
 #[async_trait]
-pub trait TargetClient<P: HeadersSyncPipeline> {
-	/// Type of error this clients returns.
-	type Error: std::fmt::Debug + MaybeConnectionError;
-
+pub trait TargetClient<P: HeadersSyncPipeline>: RelayClient {
 	/// Returns ID of best header known to the target node.
 	async fn best_header_id(&self) -> Result<HeaderIdOf<P>, Self::Error>;
 
@@ -106,7 +102,7 @@ pub trait TargetClient<P: HeadersSyncPipeline> {
 
 /// Synchronization maintain procedure.
 #[async_trait]
-pub trait SyncMaintain<P: HeadersSyncPipeline>: Send + Sync {
+pub trait SyncMaintain<P: HeadersSyncPipeline>: Clone + Send + Sync {
 	/// Run custom maintain procedures. This is guaranteed to be called when both source and target
 	/// clients are unoccupied.
 	async fn maintain(&self, _sync: &mut HeadersSync<P>) {}
@@ -126,17 +122,10 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 	metrics_params: Option<MetricsParams>,
 	exit_signal: impl Future<Output = ()>,
 ) {
-	#![allow(unused_variables)] // this is to suppress weird errors from clippy
-	let mut local_pool = futures::executor::LocalPool::new();
-	let mut progress_context = (Instant::now(), None, None);
+	let exit_signal = exit_signal.shared();
 
-	local_pool.run_until(async move {
-		let mut sync = HeadersSync::<P>::new(sync_params);
-		let mut stall_countdown = None;
-		let mut last_update_time = Instant::now();
-
-		let mut metrics_global = GlobalMetrics::default();
-		let mut metrics_sync = SyncLoopMetrics::default();
+	let metrics_global = GlobalMetrics::default();
+	let metrics_sync = SyncLoopMetrics::default();
 	let metrics_enabled = metrics_params.is_some();
 	metrics_start(
 		format!("{}_to_{}_Sync", P::SOURCE_NAME, P::TARGET_NAME),
@@ -145,6 +134,53 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 		&metrics_sync,
 	);
 
+	relay_utils::relay_loop::run(
+		relay_utils::relay_loop::RECONNECT_DELAY,
+		source_client,
+		target_client,
+		|source_client, target_client| {
+			run_until_connection_lost(
+				source_client,
+				source_tick,
+				target_client,
+				target_tick,
+				sync_maintain.clone(),
+				sync_params.clone(),
+				if metrics_enabled {
+					Some(metrics_global.clone())
+				} else {
+					None
+				},
+				if metrics_enabled {
+					Some(metrics_sync.clone())
+				} else {
+					None
+				},
+				exit_signal.clone(),
+			)
+		},
+	);
+}
+
+/// Run headers synchronization.
+#[allow(clippy::too_many_arguments)]
+async fn run_until_connection_lost<P: HeadersSyncPipeline, TC: TargetClient<P>>(
+	source_client: impl SourceClient<P>,
+	source_tick: Duration,
+	target_client: TC,
+	target_tick: Duration,
+	sync_maintain: impl SyncMaintain<P>,
+	sync_params: HeadersSyncParams,
+	metrics_global: Option<GlobalMetrics>,
+	metrics_sync: Option<SyncLoopMetrics>,
+	exit_signal: impl Future<Output = ()>,
+) -> Result<(), FailedClient> {
+	let mut progress_context = (Instant::now(), None, None);
+
+	let mut sync = HeadersSync::<P>::new(sync_params);
+	let mut stall_countdown = None;
+	let mut last_update_time = Instant::now();
+
 	let mut source_retry_backoff = retry_backoff();
 	let mut source_client_is_online = false;
 	let mut source_best_block_number_required = false;
@@ -206,7 +242,7 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 					&mut source_go_offline_future,
 					async_std::task::sleep,
 					|| format!("Error retrieving best header number from {}", P::SOURCE_NAME),
-					).is_ok();
+				).fail_if_connection_error(FailedClient::Source)?;
 			},
 			source_new_header = source_new_header_future => {
 				source_client_is_online = process_future_result(
@@ -216,7 +252,7 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 					&mut source_go_offline_future,
 					async_std::task::sleep,
 					|| format!("Error retrieving header from {} node", P::SOURCE_NAME),
-					).is_ok();
+				).fail_if_connection_error(FailedClient::Source)?;
 			},
 			source_orphan_header = source_orphan_header_future => {
 				source_client_is_online = process_future_result(
@@ -226,7 +262,7 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 					&mut source_go_offline_future,
 					async_std::task::sleep,
 					|| format!("Error retrieving orphan header from {} node", P::SOURCE_NAME),
-					).is_ok();
+				).fail_if_connection_error(FailedClient::Source)?;
 			},
 			source_extra = source_extra_future => {
 				source_client_is_online = process_future_result(
@@ -236,7 +272,7 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 					&mut source_go_offline_future,
 					async_std::task::sleep,
 					|| format!("Error retrieving extra data from {} node", P::SOURCE_NAME),
-					).is_ok();
+				).fail_if_connection_error(FailedClient::Source)?;
 			},
 			source_completion = source_completion_future => {
 				source_client_is_online = process_future_result(
@@ -246,9 +282,9 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 					&mut source_go_offline_future,
 					async_std::task::sleep,
 					|| format!("Error retrieving completion data from {} node", P::SOURCE_NAME),
-					).is_ok();
+				).fail_if_connection_error(FailedClient::Source)?;
 			},
-				source_client = source_go_offline_future => {
+			_ = source_go_offline_future => {
 				source_client_is_online = true;
 			},
 			_ = source_tick_stream.next() => {
@@ -297,7 +333,7 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 					&mut target_go_offline_future,
 					async_std::task::sleep,
 					|| format!("Error retrieving best known {} header from {} node", P::SOURCE_NAME, P::TARGET_NAME),
-					).is_ok();
+				).fail_if_connection_error(FailedClient::Target)?;
 			},
 			incomplete_headers_ids = target_incomplete_headers_future => {
 				target_incomplete_headers_required = false;
@@ -309,7 +345,7 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 					&mut target_go_offline_future,
 					async_std::task::sleep,
 					|| format!("Error retrieving incomplete headers from {} node", P::TARGET_NAME),
-					).is_ok();
+				).fail_if_connection_error(FailedClient::Target)?;
 			},
 			target_existence_status = target_existence_status_future => {
 				target_client_is_online = process_future_result(
@@ -321,7 +357,7 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 					&mut target_go_offline_future,
 					async_std::task::sleep,
 					|| format!("Error retrieving existence status from {} node", P::TARGET_NAME),
-					).is_ok();
+				).fail_if_connection_error(FailedClient::Target)?;
 			},
 			submitted_headers = target_submit_header_future => {
 				// following line helps Rust understand the type of `submitted_headers` :/
@@ -349,7 +385,7 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 					&mut target_go_offline_future,
 					async_std::task::sleep,
 					|| format!("Error submitting headers to {} node", P::TARGET_NAME),
-					).is_ok();
+				).fail_if_connection_error(FailedClient::Target)?;
 
 				log::debug!(target: "bridge", "Header submit result: {}", submitted_headers_str);
 
@@ -370,7 +406,7 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 					&mut target_go_offline_future,
 					async_std::task::sleep,
 					|| format!("Error completing headers at {}", P::TARGET_NAME),
-					).is_ok();
+				).fail_if_connection_error(FailedClient::Target)?;
 			},
 			target_extra_check_result = target_extra_check_future => {
 				target_client_is_online = process_future_result(
@@ -382,9 +418,9 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 					&mut target_go_offline_future,
 					async_std::task::sleep,
 					|| format!("Error retrieving receipts requirement from {} node", P::TARGET_NAME),
-					).is_ok();
+				).fail_if_connection_error(FailedClient::Target)?;
 			},
-				target_client = target_go_offline_future => {
+			_ = target_go_offline_future => {
 				target_client_is_online = true;
 			},
 			_ = target_tick_stream.next() => {
@@ -396,13 +432,15 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 				maintain_required = true;
 			},
 			_ = exit_signal => {
-					return;
+				return Ok(());
 			}
 		}
 
 		// update metrics
-			if metrics_enabled {
-				metrics_global.update();
+		if let Some(ref metrics_global) = metrics_global {
+			metrics_global.update().await;
+		}
+		if let Some(ref metrics_sync) = metrics_sync {
 			metrics_sync.update(&sync);
 		}
 
@@ -559,7 +597,7 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 						P::SOURCE_NAME,
 						P::TARGET_NAME,
 					);
-						return;
+					return Ok(());
 				}
 
 				log::debug!(
@@ -584,7 +622,6 @@ pub fn run<P: HeadersSyncPipeline, TC: TargetClient<P>>(
 			}
 		}
 	}
-	});
 }
 
 /// Print synchronization progress.

bridges/relays/headers-relay/src/sync_loop_metrics.rs

@@ -23,6 +23,7 @@ use num_traits::Zero;
 use relay_utils::metrics::{register, GaugeVec, Metrics, Opts, Registry, U64};
 
 /// Headers sync metrics.
+#[derive(Clone)]
 pub struct SyncLoopMetrics {
 	/// Best syncing headers at "source" and "target" nodes.
 	best_block_numbers: GaugeVec<U64>,
@@ -57,7 +58,7 @@ impl Default for SyncLoopMetrics {
 
 impl SyncLoopMetrics {
 	/// Update metrics.
-	pub fn update<P: HeadersSyncPipeline>(&mut self, sync: &HeadersSync<P>) {
+	pub fn update<P: HeadersSyncPipeline>(&self, sync: &HeadersSync<P>) {
 		let headers = sync.headers();
 		let source_best_number = sync.source_best_number().unwrap_or_else(Zero::zero);
 		let target_best_number = sync.target_best_header().map(|id| id.0).unwrap_or_else(Zero::zero);

bridges/relays/headers-relay/src/sync_loop_tests.rs

@@ -23,7 +23,9 @@ use async_trait::async_trait;
 use backoff::backoff::Backoff;
 use futures::{future::FutureExt, stream::StreamExt};
 use parking_lot::Mutex;
-use relay_utils::{process_future_result, retry_backoff, HeaderId, MaybeConnectionError};
+use relay_utils::{
+	process_future_result, relay_loop::Client as RelayClient, retry_backoff, HeaderId, MaybeConnectionError,
+};
 use std::{
 	collections::{HashMap, HashSet},
 	sync::Arc,
@@ -89,8 +91,9 @@ enum SourceMethod {
 	HeaderExtra(TestHeaderId, TestQueuedHeader),
 }
 
+#[derive(Clone)]
 struct Source {
-	data: Mutex<SourceData>,
+	data: Arc<Mutex<SourceData>>,
 	on_method_call: Arc<dyn Fn(SourceMethod, &mut SourceData) + Send + Sync>,
 }
 
@@ -109,7 +112,7 @@ impl Source {
 		on_method_call: impl Fn(SourceMethod, &mut SourceData) + Send + Sync + 'static,
 	) -> Self {
 		Source {
-			data: Mutex::new(SourceData {
+			data: Arc::new(Mutex::new(SourceData {
 				best_block_number: Ok(best_block_id.0),
 				header_by_hash: headers
 					.iter()
@@ -127,35 +130,42 @@ impl Source {
 					.collect(),
 				provides_completion: true,
 				provides_extra: true,
-			}),
+			})),
 			on_method_call: Arc::new(on_method_call),
 		}
 	}
 }
 
 #[async_trait]
-impl SourceClient<TestHeadersSyncPipeline> for Source {
+impl RelayClient for Source {
 	type Error = TestError;
 
-	async fn best_block_number(&self) -> Result<TestNumber, Self::Error> {
+	async fn reconnect(&mut self) -> Result<(), TestError> {
+		unimplemented!()
+	}
+}
+
+#[async_trait]
+impl SourceClient<TestHeadersSyncPipeline> for Source {
+	async fn best_block_number(&self) -> Result<TestNumber, TestError> {
 		let mut data = self.data.lock();
 		(self.on_method_call)(SourceMethod::BestBlockNumber, &mut *data);
 		data.best_block_number.clone()
 	}
 
-	async fn header_by_hash(&self, hash: TestHash) -> Result<TestHeader, Self::Error> {
+	async fn header_by_hash(&self, hash: TestHash) -> Result<TestHeader, TestError> {
 		let mut data = self.data.lock();
 		(self.on_method_call)(SourceMethod::HeaderByHash(hash), &mut *data);
 		data.header_by_hash.get(&hash).cloned().ok_or(TestError(false))
 	}
 
-	async fn header_by_number(&self, number: TestNumber) -> Result<TestHeader, Self::Error> {
+	async fn header_by_number(&self, number: TestNumber) -> Result<TestHeader, TestError> {
 		let mut data = self.data.lock();
 		(self.on_method_call)(SourceMethod::HeaderByNumber(number), &mut *data);
 		data.header_by_number.get(&number).cloned().ok_or(TestError(false))
 	}
 
-	async fn header_completion(&self, id: TestHeaderId) -> Result<(TestHeaderId, Option<TestCompletion>), Self::Error> {
+	async fn header_completion(&self, id: TestHeaderId) -> Result<(TestHeaderId, Option<TestCompletion>), TestError> {
 		let mut data = self.data.lock();
 		(self.on_method_call)(SourceMethod::HeaderCompletion(id), &mut *data);
 		if data.provides_completion {
@@ -169,7 +179,7 @@ impl SourceClient<TestHeadersSyncPipeline> for Source {
 		&self,
 		id: TestHeaderId,
 		header: TestQueuedHeader,
-	) -> Result<(TestHeaderId, TestExtra), Self::Error> {
+	) -> Result<(TestHeaderId, TestExtra), TestError> {
 		let mut data = self.data.lock();
 		(self.on_method_call)(SourceMethod::HeaderExtra(id, header), &mut *data);
 		if data.provides_extra {
@@ -189,8 +199,9 @@ enum TargetMethod {
 	RequiresExtra(TestQueuedHeader),
 }
 
+#[derive(Clone)]
 struct Target {
-	data: Mutex<TargetData>,
+	data: Arc<Mutex<TargetData>>,
 	on_method_call: Arc<dyn Fn(TargetMethod, &mut TargetData) + Send + Sync>,
 }
 
@@ -211,7 +222,7 @@ impl Target {
 		on_method_call: impl Fn(TargetMethod, &mut TargetData) + Send + Sync + 'static,
 	) -> Self {
 		Target {
-			data: Mutex::new(TargetData {
+			data: Arc::new(Mutex::new(TargetData {
 				best_header_id: Ok(best_header_id),
 				is_known_header_by_hash: headers.iter().map(|header| (header.1, true)).collect(),
 				submitted_headers: HashMap::new(),
@@ -219,16 +230,23 @@ impl Target {
 				completed_headers: HashMap::new(),
 				requires_completion: false,
 				requires_extra: false,
-			}),
+			})),
 			on_method_call: Arc::new(on_method_call),
 		}
 	}
 }
 
 #[async_trait]
-impl TargetClient<TestHeadersSyncPipeline> for Target {
+impl RelayClient for Target {
 	type Error = TestError;
 
+	async fn reconnect(&mut self) -> Result<(), TestError> {
+		unimplemented!()
+	}
+}
+
+#[async_trait]
+impl TargetClient<TestHeadersSyncPipeline> for Target {
 	async fn best_header_id(&self) -> Result<TestHeaderId, TestError> {
 		let mut data = self.data.lock();
 		(self.on_method_call)(TargetMethod::BestHeaderId, &mut *data);

bridges/relays/messages-relay/src/message_lane_loop.rs

@@ -35,7 +35,9 @@ use futures::{channel::mpsc::unbounded, future::FutureExt, stream::StreamExt};
 use relay_utils::{
 	interval,
 	metrics::{start as metrics_start, GlobalMetrics, MetricsParams},
-	process_future_result, retry_backoff, FailedClient, MaybeConnectionError,
+	process_future_result,
+	relay_loop::Client as RelayClient,
+	retry_backoff, FailedClient,
 };
 use std::{collections::BTreeMap, fmt::Debug, future::Future, ops::RangeInclusive, time::Duration};
 
@@ -98,13 +100,7 @@ pub struct MessageProofParameters {
 
 /// Source client trait.
 #[async_trait]
-pub trait SourceClient<P: MessageLane>: Clone + Send + Sync {
-	/// Type of error this clients returns.
-	type Error: std::fmt::Debug + MaybeConnectionError;
-
-	/// Try to reconnect to source node.
-	async fn reconnect(self) -> Result<Self, Self::Error>;
-
+pub trait SourceClient<P: MessageLane>: RelayClient {
 	/// Returns state of the client.
 	async fn state(&self) -> Result<SourceClientState<P>, Self::Error>;
 
@@ -147,13 +143,7 @@ pub trait SourceClient<P: MessageLane>: Clone + Send + Sync {
 
 /// Target client trait.
 #[async_trait]
-pub trait TargetClient<P: MessageLane>: Clone + Send + Sync {
-	/// Type of error this clients returns.
-	type Error: std::fmt::Debug + MaybeConnectionError;
-
-	/// Try to reconnect to source node.
-	async fn reconnect(self) -> Result<Self, Self::Error>;
-
+pub trait TargetClient<P: MessageLane>: RelayClient {
 	/// Returns state of the client.
 	async fn state(&self) -> Result<TargetClientState<P>, Self::Error>;
 
@@ -218,16 +208,13 @@ pub struct ClientsState<P: MessageLane> {
 /// Run message lane service loop.
 pub fn run<P: MessageLane>(
 	params: Params,
-	mut source_client: impl SourceClient<P>,
-	mut target_client: impl TargetClient<P>,
+	source_client: impl SourceClient<P>,
+	target_client: impl TargetClient<P>,
 	metrics_params: Option<MetricsParams>,
 	exit_signal: impl Future<Output = ()>,
 ) {
-	let mut local_pool = futures::executor::LocalPool::new();
 	let exit_signal = exit_signal.shared();
-
-	local_pool.run_until(async move {
-		let mut metrics_global = GlobalMetrics::default();
+	let metrics_global = GlobalMetrics::default();
 	let metrics_msg = MessageLaneLoopMetrics::default();
 	let metrics_enabled = metrics_params.is_some();
 	metrics_start(
@@ -242,13 +229,17 @@ pub fn run<P: MessageLane>(
 		&metrics_msg,
 	);
 
-		loop {
-			let result = run_until_connection_lost(
+	relay_utils::relay_loop::run(
+		params.reconnect_delay,
+		source_client,
+		target_client,
+		|source_client, target_client| {
+			run_until_connection_lost(
 				params.clone(),
-				source_client.clone(),
-				target_client.clone(),
+				source_client,
+				target_client,
 				if metrics_enabled {
-					Some(&mut metrics_global)
+					Some(metrics_global.clone())
 				} else {
 					None
 				},
@@ -259,55 +250,8 @@ pub fn run<P: MessageLane>(
 				},
 				exit_signal.clone(),
 			)
-			.await;
-
-			match result {
-				Ok(()) => break,
-				Err(failed_client) => loop {
-					async_std::task::sleep(params.reconnect_delay).await;
-					if failed_client == FailedClient::Both || failed_client == FailedClient::Source {
-						source_client = match source_client.clone().reconnect().await {
-							Ok(source_client) => source_client,
-							Err(error) => {
-								log::warn!(
-									target: "bridge",
-									"Failed to reconnect {}. Going to retry in {}s: {:?}",
-									P::SOURCE_NAME,
-									params.reconnect_delay.as_secs(),
-									error,
-								);
-								continue;
-							}
-						}
-					}
-					if failed_client == FailedClient::Both || failed_client == FailedClient::Target {
-						target_client = match target_client.clone().reconnect().await {
-							Ok(target_client) => target_client,
-							Err(error) => {
-								log::warn!(
-									target: "bridge",
-									"Failed to reconnect {}. Going to retry in {}s: {:?}",
-									P::TARGET_NAME,
-									params.reconnect_delay.as_secs(),
-									error,
-								);
-								continue;
-							}
-						}
-					}
-
-					break;
 		},
-			}
-
-			log::debug!(
-				target: "bridge",
-				"Restarting lane {} -> {}",
-				P::SOURCE_NAME,
-				P::TARGET_NAME,
 	);
-		}
-	});
 }
 
 /// Run one-way message delivery loop until connection with target or source node is lost, or exit signal is received.
@@ -315,7 +259,7 @@ async fn run_until_connection_lost<P: MessageLane, SC: SourceClient<P>, TC: Targ
 	params: Params,
 	source_client: SC,
 	target_client: TC,
-	mut metrics_global: Option<&mut GlobalMetrics>,
+	metrics_global: Option<GlobalMetrics>,
 	metrics_msg: Option<MessageLaneLoopMetrics>,
 	exit_signal: impl Future<Output = ()>,
 ) -> Result<(), FailedClient> {
@@ -459,8 +403,8 @@ async fn run_until_connection_lost<P: MessageLane, SC: SourceClient<P>, TC: Targ
 			}
 		}
 
-		if let Some(metrics_global) = metrics_global.as_mut() {
-			metrics_global.update();
+		if let Some(ref metrics_global) = metrics_global {
+			metrics_global.update().await;
 		}
 
 		if source_client_is_online && source_state_required {
@@ -482,7 +426,7 @@ pub(crate) mod tests {
 	use super::*;
 	use futures::stream::StreamExt;
 	use parking_lot::Mutex;
-	use relay_utils::HeaderId;
+	use relay_utils::{HeaderId, MaybeConnectionError};
 	use std::sync::Arc;
 
 	pub fn header_id(number: TestSourceHeaderNumber) -> TestSourceHeaderId {
@@ -550,19 +494,22 @@ pub(crate) mod tests {
 	}
 
 	#[async_trait]
-	impl SourceClient<TestMessageLane> for TestSourceClient {
+	impl RelayClient for TestSourceClient {
 		type Error = TestError;
 
-		async fn reconnect(self) -> Result<Self, Self::Error> {
+		async fn reconnect(&mut self) -> Result<(), TestError> {
 			{
 				let mut data = self.data.lock();
 				(self.tick)(&mut *data);
 				data.is_source_reconnected = true;
 			}
-			Ok(self)
+			Ok(())
+		}
 	}
 
-		async fn state(&self) -> Result<SourceClientState<TestMessageLane>, Self::Error> {
+	#[async_trait]
+	impl SourceClient<TestMessageLane> for TestSourceClient {
+		async fn state(&self) -> Result<SourceClientState<TestMessageLane>, TestError> {
 			let mut data = self.data.lock();
 			(self.tick)(&mut *data);
 			if data.is_source_fails {
@@ -574,7 +521,7 @@ pub(crate) mod tests {
 		async fn latest_generated_nonce(
 			&self,
 			id: SourceHeaderIdOf<TestMessageLane>,
-		) -> Result<(SourceHeaderIdOf<TestMessageLane>, MessageNonce), Self::Error> {
+		) -> Result<(SourceHeaderIdOf<TestMessageLane>, MessageNonce), TestError> {
 			let mut data = self.data.lock();
 			(self.tick)(&mut *data);
 			if data.is_source_fails {
@@ -586,7 +533,7 @@ pub(crate) mod tests {
 		async fn latest_confirmed_received_nonce(
 			&self,
 			id: SourceHeaderIdOf<TestMessageLane>,
-		) -> Result<(SourceHeaderIdOf<TestMessageLane>, MessageNonce), Self::Error> {
+		) -> Result<(SourceHeaderIdOf<TestMessageLane>, MessageNonce), TestError> {
 			let mut data = self.data.lock();
 			(self.tick)(&mut *data);
 			Ok((id, data.source_latest_confirmed_received_nonce))
@@ -596,7 +543,7 @@ pub(crate) mod tests {
 			&self,
 			_id: SourceHeaderIdOf<TestMessageLane>,
 			nonces: RangeInclusive<MessageNonce>,
-		) -> Result<MessageWeightsMap, Self::Error> {
+		) -> Result<MessageWeightsMap, TestError> {
 			Ok(nonces
 				.map(|nonce| (nonce, MessageWeights { weight: 1, size: 1 }))
 				.collect())
@@ -613,7 +560,7 @@ pub(crate) mod tests {
 				RangeInclusive<MessageNonce>,
 				TestMessagesProof,
 			),
-			Self::Error,
+			TestError,
 		> {
 			let mut data = self.data.lock();
 			(self.tick)(&mut *data);
@@ -635,7 +582,7 @@ pub(crate) mod tests {
 			&self,
 			_generated_at_block: TargetHeaderIdOf<TestMessageLane>,
 			proof: TestMessagesReceivingProof,
-		) -> Result<(), Self::Error> {
+		) -> Result<(), TestError> {
 			let mut data = self.data.lock();
 			(self.tick)(&mut *data);
 			data.submitted_messages_receiving_proofs.push(proof);
@@ -651,19 +598,22 @@ pub(crate) mod tests {
 	}
 
 	#[async_trait]
-	impl TargetClient<TestMessageLane> for TestTargetClient {
+	impl RelayClient for TestTargetClient {
 		type Error = TestError;
 
-		async fn reconnect(self) -> Result<Self, Self::Error> {
+		async fn reconnect(&mut self) -> Result<(), TestError> {
 			{
 				let mut data = self.data.lock();
 				(self.tick)(&mut *data);
 				data.is_target_reconnected = true;
 			}
-			Ok(self)
+			Ok(())
+		}
 	}
 
-		async fn state(&self) -> Result<TargetClientState<TestMessageLane>, Self::Error> {
+	#[async_trait]
+	impl TargetClient<TestMessageLane> for TestTargetClient {
+		async fn state(&self) -> Result<TargetClientState<TestMessageLane>, TestError> {
 			let mut data = self.data.lock();
 			(self.tick)(&mut *data);
 			if data.is_target_fails {
@@ -675,7 +625,7 @@ pub(crate) mod tests {
 		async fn latest_received_nonce(
 			&self,
 			id: TargetHeaderIdOf<TestMessageLane>,
-		) -> Result<(TargetHeaderIdOf<TestMessageLane>, MessageNonce), Self::Error> {
+		) -> Result<(TargetHeaderIdOf<TestMessageLane>, MessageNonce), TestError> {
 			let mut data = self.data.lock();
 			(self.tick)(&mut *data);
 			if data.is_target_fails {
@@ -687,7 +637,7 @@ pub(crate) mod tests {
 		async fn unrewarded_relayers_state(
 			&self,
 			id: TargetHeaderIdOf<TestMessageLane>,
-		) -> Result<(TargetHeaderIdOf<TestMessageLane>, UnrewardedRelayersState), Self::Error> {
+		) -> Result<(TargetHeaderIdOf<TestMessageLane>, UnrewardedRelayersState), TestError> {
 			Ok((
 				id,
 				UnrewardedRelayersState {
@@ -701,7 +651,7 @@ pub(crate) mod tests {
 		async fn latest_confirmed_received_nonce(
 			&self,
 			id: TargetHeaderIdOf<TestMessageLane>,
-		) -> Result<(TargetHeaderIdOf<TestMessageLane>, MessageNonce), Self::Error> {
+		) -> Result<(TargetHeaderIdOf<TestMessageLane>, MessageNonce), TestError> {
 			let mut data = self.data.lock();
 			(self.tick)(&mut *data);
 			if data.is_target_fails {
@@ -713,7 +663,7 @@ pub(crate) mod tests {
 		async fn prove_messages_receiving(
 			&self,
 			id: TargetHeaderIdOf<TestMessageLane>,
-		) -> Result<(TargetHeaderIdOf<TestMessageLane>, TestMessagesReceivingProof), Self::Error> {
+		) -> Result<(TargetHeaderIdOf<TestMessageLane>, TestMessagesReceivingProof), TestError> {
 			Ok((id, self.data.lock().target_latest_received_nonce))
 		}
 
@@ -722,7 +672,7 @@ pub(crate) mod tests {
 			_generated_at_header: SourceHeaderIdOf<TestMessageLane>,
 			nonces: RangeInclusive<MessageNonce>,
 			proof: TestMessagesProof,
-		) -> Result<RangeInclusive<MessageNonce>, Self::Error> {
+		) -> Result<RangeInclusive<MessageNonce>, TestError> {
 			let mut data = self.data.lock();
 			(self.tick)(&mut *data);
 			if data.is_target_fails {

bridges/relays/substrate-client/src/client.rs

@@ -89,12 +89,9 @@ impl<C: Chain> Client<C> {
 	}
 
 	/// Reopen client connection.
-	pub async fn reconnect(self) -> Result<Self> {
-		Ok(Self {
-			params: self.params.clone(),
-			client: Self::build_client(self.params).await?,
-			genesis_hash: self.genesis_hash,
-		})
+	pub async fn reconnect(&mut self) -> Result<()> {
+		self.client = Self::build_client(self.params.clone()).await?;
+		Ok(())
 	}
 
 	/// Build client to use in connection.

bridges/relays/substrate-client/src/headers_source.rs

@@ -25,6 +25,7 @@ use headers_relay::{
 	sync_loop::SourceClient,
 	sync_types::{HeaderIdOf, HeadersSyncPipeline, QueuedHeader, SourceHeader},
 };
+use relay_utils::relay_loop::Client as RelayClient;
 use sp_runtime::{traits::Header as HeaderT, Justification};
 use std::marker::PhantomData;
 
@@ -44,6 +45,24 @@ impl<C: Chain, P> HeadersSource<C, P> {
 	}
 }
 
+impl<C: Chain, P> Clone for HeadersSource<C, P> {
+	fn clone(&self) -> Self {
+		HeadersSource {
+			client: self.client.clone(),
+			_phantom: Default::default(),
+		}
+	}
+}
+
+#[async_trait]
+impl<C: Chain, P: HeadersSyncPipeline> RelayClient for HeadersSource<C, P> {
+	type Error = Error;
+
+	async fn reconnect(&mut self) -> Result<(), Error> {
+		self.client.reconnect().await
+	}
+}
+
 #[async_trait]
 impl<C, P> SourceClient<P> for HeadersSource<C, P>
 where
@@ -53,13 +72,11 @@ where
 	P: HeadersSyncPipeline<Extra = (), Completion = Justification, Hash = C::Hash, Number = C::BlockNumber>,
 	P::Header: SourceHeader<C::Hash, C::BlockNumber>,
 {
-	type Error = Error;
-
-	async fn best_block_number(&self) -> Result<P::Number, Self::Error> {
+	async fn best_block_number(&self) -> Result<P::Number, Error> {
 		Ok(*self.client.best_header().await?.number())
 	}
 
-	async fn header_by_hash(&self, hash: P::Hash) -> Result<P::Header, Self::Error> {
+	async fn header_by_hash(&self, hash: P::Hash) -> Result<P::Header, Error> {
 		self.client
 			.header_by_hash(hash)
 			.await
@@ -67,7 +84,7 @@ where
 			.map_err(Into::into)
 	}
 
-	async fn header_by_number(&self, number: P::Number) -> Result<P::Header, Self::Error> {
+	async fn header_by_number(&self, number: P::Number) -> Result<P::Header, Error> {
 		self.client
 			.header_by_number(number)
 			.await
@@ -75,10 +92,7 @@ where
 			.map_err(Into::into)
 	}
 
-	async fn header_completion(
-		&self,
-		id: HeaderIdOf<P>,
-	) -> Result<(HeaderIdOf<P>, Option<P::Completion>), Self::Error> {
+	async fn header_completion(&self, id: HeaderIdOf<P>) -> Result<(HeaderIdOf<P>, Option<P::Completion>), Error> {
 		let hash = id.1;
 		let signed_block = self.client.get_block(Some(hash)).await?;
 		let grandpa_justification = signed_block.justification().cloned();
@@ -86,11 +100,7 @@ where
 		Ok((id, grandpa_justification))
 	}
 
-	async fn header_extra(
-		&self,
-		id: HeaderIdOf<P>,
-		_header: QueuedHeader<P>,
-	) -> Result<(HeaderIdOf<P>, ()), Self::Error> {
+	async fn header_extra(&self, id: HeaderIdOf<P>, _header: QueuedHeader<P>) -> Result<(HeaderIdOf<P>, ()), Error> {
 		Ok((id, ()))
 	}
 }

bridges/relays/substrate/src/headers_maintain.rs

@@ -80,6 +80,20 @@ impl<P: SubstrateHeadersSyncPipeline, SourceChain, TargetChain: Chain>
 	}
 }
 
+#[async_trait]
+impl<P: SubstrateHeadersSyncPipeline, SourceChain, TargetChain: Chain> Clone
+	for SubstrateHeadersToSubstrateMaintain<P, SourceChain, TargetChain>
+{
+	fn clone(&self) -> Self {
+		SubstrateHeadersToSubstrateMaintain {
+			pipeline: self.pipeline.clone(),
+			target_client: self.target_client.clone(),
+			justifications: self.justifications.clone(),
+			_marker: Default::default(),
+		}
+	}
+}
+
 #[async_trait]
 impl<P, SourceChain, TargetChain> SyncMaintain<P> for SubstrateHeadersToSubstrateMaintain<P, SourceChain, TargetChain>
 where

bridges/relays/substrate/src/headers_target.rs

@@ -28,7 +28,7 @@ use headers_relay::{
 	sync_types::{HeaderIdOf, QueuedHeader, SubmittedHeaders},
 };
 use relay_substrate_client::{Chain, Client, Error as SubstrateError};
-use relay_utils::HeaderId;
+use relay_utils::{relay_loop::Client as RelayClient, HeaderId};
 use sp_core::Bytes;
 use sp_runtime::Justification;
 use std::collections::HashSet;
@@ -46,6 +46,24 @@ impl<C: Chain, P> SubstrateHeadersTarget<C, P> {
 	}
 }
 
+impl<C: Chain, P: SubstrateHeadersSyncPipeline> Clone for SubstrateHeadersTarget<C, P> {
+	fn clone(&self) -> Self {
+		SubstrateHeadersTarget {
+			client: self.client.clone(),
+			pipeline: self.pipeline.clone(),
+		}
+	}
+}
+
+#[async_trait]
+impl<C: Chain, P: SubstrateHeadersSyncPipeline> RelayClient for SubstrateHeadersTarget<C, P> {
+	type Error = SubstrateError;
+
+	async fn reconnect(&mut self) -> Result<(), SubstrateError> {
+		self.client.reconnect().await
+	}
+}
+
 #[async_trait]
 impl<C, P> TargetClient<P> for SubstrateHeadersTarget<C, P>
 where
@@ -54,9 +72,7 @@ where
 	P::Hash: Decode + Encode,
 	P: SubstrateHeadersSyncPipeline<Completion = Justification, Extra = ()>,
 {
-	type Error = SubstrateError;
-
-	async fn best_header_id(&self) -> Result<HeaderIdOf<P>, Self::Error> {
+	async fn best_header_id(&self) -> Result<HeaderIdOf<P>, SubstrateError> {
 		let call = P::BEST_BLOCK_METHOD.into();
 		let data = Bytes(Vec::new());
 
@@ -72,7 +88,7 @@ where
 			.map(|(num, hash)| HeaderId(*num, *hash))
 	}
 
-	async fn is_known_header(&self, id: HeaderIdOf<P>) -> Result<(HeaderIdOf<P>, bool), Self::Error> {
+	async fn is_known_header(&self, id: HeaderIdOf<P>) -> Result<(HeaderIdOf<P>, bool), SubstrateError> {
 		let call = P::IS_KNOWN_BLOCK_METHOD.into();
 		let data = Bytes(id.1.encode());
 
@@ -83,7 +99,10 @@ where
 		Ok((id, is_known_block))
 	}
 
-	async fn submit_headers(&self, mut headers: Vec<QueuedHeader<P>>) -> SubmittedHeaders<HeaderIdOf<P>, Self::Error> {
+	async fn submit_headers(
+		&self,
+		mut headers: Vec<QueuedHeader<P>>,
+	) -> SubmittedHeaders<HeaderIdOf<P>, SubstrateError> {
 		debug_assert_eq!(
 			headers.len(),
 			1,
@@ -114,7 +133,7 @@ where
 		}
 	}
 
-	async fn incomplete_headers_ids(&self) -> Result<HashSet<HeaderIdOf<P>>, Self::Error> {
+	async fn incomplete_headers_ids(&self) -> Result<HashSet<HeaderIdOf<P>>, SubstrateError> {
 		let call = P::INCOMPLETE_HEADERS_METHOD.into();
 		let data = Bytes(Vec::new());
 
@@ -133,13 +152,13 @@ where
 		&self,
 		id: HeaderIdOf<P>,
 		completion: Justification,
-	) -> Result<HeaderIdOf<P>, Self::Error> {
+	) -> Result<HeaderIdOf<P>, SubstrateError> {
 		let tx = self.pipeline.make_complete_header_transaction(id, completion).await?;
 		self.client.submit_extrinsic(Bytes(tx.encode())).await?;
 		Ok(id)
 	}
 
-	async fn requires_extra(&self, header: QueuedHeader<P>) -> Result<(HeaderIdOf<P>, bool), Self::Error> {
+	async fn requires_extra(&self, header: QueuedHeader<P>) -> Result<(HeaderIdOf<P>, bool), SubstrateError> {
 		Ok((header.id(), false))
 	}
 }

bridges/relays/substrate/src/messages_source.rs

@@ -32,7 +32,7 @@ use messages_relay::{
 	},
 };
 use relay_substrate_client::{Chain, Client, Error as SubstrateError, HashOf, HeaderIdOf};
-use relay_utils::{BlockNumberBase, HeaderId};
+use relay_utils::{relay_loop::Client as RelayClient, BlockNumberBase, HeaderId};
 use sp_core::Bytes;
 use sp_runtime::{traits::Header as HeaderT, DeserializeOwned};
 use sp_trie::StorageProof;
@@ -74,6 +74,15 @@ impl<C: Chain, P: SubstrateMessageLane> Clone for SubstrateMessagesSource<C, P>
 	}
 }
 
+#[async_trait]
+impl<C: Chain, P: SubstrateMessageLane> RelayClient for SubstrateMessagesSource<C, P> {
+	type Error = SubstrateError;
+
+	async fn reconnect(&mut self) -> Result<(), SubstrateError> {
+		self.client.reconnect().await
+	}
+}
+
 #[async_trait]
 impl<C, P> SourceClient<P> for SubstrateMessagesSource<C, P>
 where
@@ -89,15 +98,7 @@ where
 	P::TargetHeaderNumber: Decode,
 	P::TargetHeaderHash: Decode,
 {
-	type Error = SubstrateError;
-
-	async fn reconnect(mut self) -> Result<Self, Self::Error> {
-		let new_client = self.client.clone().reconnect().await?;
-		self.client = new_client;
-		Ok(self)
-	}
-
-	async fn state(&self) -> Result<SourceClientState<P>, Self::Error> {
+	async fn state(&self) -> Result<SourceClientState<P>, SubstrateError> {
 		read_client_state::<_, P::TargetHeaderHash, P::TargetHeaderNumber>(
 			&self.client,
 			P::BEST_FINALIZED_TARGET_HEADER_ID_AT_SOURCE,
@@ -108,7 +109,7 @@ where
 	async fn latest_generated_nonce(
 		&self,
 		id: SourceHeaderIdOf<P>,
-	) -> Result<(SourceHeaderIdOf<P>, MessageNonce), Self::Error> {
+	) -> Result<(SourceHeaderIdOf<P>, MessageNonce), SubstrateError> {
 		let encoded_response = self
 			.client
 			.state_call(
@@ -125,7 +126,7 @@ where
 	async fn latest_confirmed_received_nonce(
 		&self,
 		id: SourceHeaderIdOf<P>,
-	) -> Result<(SourceHeaderIdOf<P>, MessageNonce), Self::Error> {
+	) -> Result<(SourceHeaderIdOf<P>, MessageNonce), SubstrateError> {
 		let encoded_response = self
 			.client
 			.state_call(
@@ -143,7 +144,7 @@ where
 		&self,
 		id: SourceHeaderIdOf<P>,
 		nonces: RangeInclusive<MessageNonce>,
-	) -> Result<MessageWeightsMap, Self::Error> {
+	) -> Result<MessageWeightsMap, SubstrateError> {
 		let encoded_response = self
 			.client
 			.state_call(
@@ -164,7 +165,7 @@ where
 		id: SourceHeaderIdOf<P>,
 		nonces: RangeInclusive<MessageNonce>,
 		proof_parameters: MessageProofParameters,
-	) -> Result<(SourceHeaderIdOf<P>, RangeInclusive<MessageNonce>, P::MessagesProof), Self::Error> {
+	) -> Result<(SourceHeaderIdOf<P>, RangeInclusive<MessageNonce>, P::MessagesProof), SubstrateError> {
 		let proof = self
 			.client
 			.prove_messages(
@@ -183,7 +184,7 @@ where
 		&self,
 		generated_at_block: TargetHeaderIdOf<P>,
 		proof: P::MessagesReceivingProof,
-	) -> Result<(), Self::Error> {
+	) -> Result<(), SubstrateError> {
 		let tx = self
 			.lane
 			.make_messages_receiving_proof_transaction(generated_at_block, proof)

bridges/relays/substrate/src/messages_target.rs

@@ -30,7 +30,7 @@ use messages_relay::{
 	message_lane_loop::{TargetClient, TargetClientState},
 };
 use relay_substrate_client::{Chain, Client, Error as SubstrateError, HashOf};
-use relay_utils::BlockNumberBase;
+use relay_utils::{relay_loop::Client as RelayClient, BlockNumberBase};
 use sp_core::Bytes;
 use sp_runtime::{traits::Header as HeaderT, DeserializeOwned};
 use sp_trie::StorageProof;
@@ -70,6 +70,15 @@ impl<C: Chain, P: SubstrateMessageLane> Clone for SubstrateMessagesTarget<C, P>
 	}
 }
 
+#[async_trait]
+impl<C: Chain, P: SubstrateMessageLane> RelayClient for SubstrateMessagesTarget<C, P> {
+	type Error = SubstrateError;
+
+	async fn reconnect(&mut self) -> Result<(), SubstrateError> {
+		self.client.reconnect().await
+	}
+}
+
 #[async_trait]
 impl<C, P> TargetClient<P> for SubstrateMessagesTarget<C, P>
 where
@@ -85,15 +94,7 @@ where
 	P::SourceHeaderNumber: Decode,
 	P::SourceHeaderHash: Decode,
 {
-	type Error = SubstrateError;
-
-	async fn reconnect(mut self) -> Result<Self, Self::Error> {
-		let new_client = self.client.clone().reconnect().await?;
-		self.client = new_client;
-		Ok(self)
-	}
-
-	async fn state(&self) -> Result<TargetClientState<P>, Self::Error> {
+	async fn state(&self) -> Result<TargetClientState<P>, SubstrateError> {
 		read_client_state::<_, P::SourceHeaderHash, P::SourceHeaderNumber>(
 			&self.client,
 			P::BEST_FINALIZED_SOURCE_HEADER_ID_AT_TARGET,
@@ -104,7 +105,7 @@ where
 	async fn latest_received_nonce(
 		&self,
 		id: TargetHeaderIdOf<P>,
-	) -> Result<(TargetHeaderIdOf<P>, MessageNonce), Self::Error> {
+	) -> Result<(TargetHeaderIdOf<P>, MessageNonce), SubstrateError> {
 		let encoded_response = self
 			.client
 			.state_call(
@@ -121,7 +122,7 @@ where
 	async fn latest_confirmed_received_nonce(
 		&self,
 		id: TargetHeaderIdOf<P>,
-	) -> Result<(TargetHeaderIdOf<P>, MessageNonce), Self::Error> {
+	) -> Result<(TargetHeaderIdOf<P>, MessageNonce), SubstrateError> {
 		let encoded_response = self
 			.client
 			.state_call(
@@ -138,7 +139,7 @@ where
 	async fn unrewarded_relayers_state(
 		&self,
 		id: TargetHeaderIdOf<P>,
-	) -> Result<(TargetHeaderIdOf<P>, UnrewardedRelayersState), Self::Error> {
+	) -> Result<(TargetHeaderIdOf<P>, UnrewardedRelayersState), SubstrateError> {
 		let encoded_response = self
 			.client
 			.state_call(
@@ -155,7 +156,7 @@ where
 	async fn prove_messages_receiving(
 		&self,
 		id: TargetHeaderIdOf<P>,
-	) -> Result<(TargetHeaderIdOf<P>, P::MessagesReceivingProof), Self::Error> {
+	) -> Result<(TargetHeaderIdOf<P>, P::MessagesReceivingProof), SubstrateError> {
 		let (id, relayers_state) = self.unrewarded_relayers_state(id).await?;
 		let proof = self
 			.client
@@ -170,7 +171,7 @@ where
 		generated_at_header: SourceHeaderIdOf<P>,
 		nonces: RangeInclusive<MessageNonce>,
 		proof: P::MessagesProof,
-	) -> Result<RangeInclusive<MessageNonce>, Self::Error> {
+	) -> Result<RangeInclusive<MessageNonce>, SubstrateError> {
 		let tx = self
 			.lane
 			.make_messages_delivery_transaction(generated_at_header, nonces.clone(), proof)

bridges/relays/substrate/src/millau_messages_to_rialto.rs

@@ -105,7 +105,6 @@ pub fn run(
 	lane_id: LaneId,
 	metrics_params: Option<MetricsParams>,
 ) {
-	let reconnect_delay = Duration::from_secs(10);
 	let stall_timeout = Duration::from_secs(5 * 60);
 	let relayer_id_at_millau = millau_sign.signer.public().as_array_ref().clone().into();
 
@@ -135,7 +134,7 @@ pub fn run(
 			lane: lane_id,
 			source_tick: Millau::AVERAGE_BLOCK_INTERVAL,
 			target_tick: Rialto::AVERAGE_BLOCK_INTERVAL,
-			reconnect_delay,
+			reconnect_delay: relay_utils::relay_loop::RECONNECT_DELAY,
 			stall_timeout,
 			delivery_params: messages_relay::message_lane_loop::MessageDeliveryParams {
 				max_unrewarded_relayer_entries_at_target: bp_rialto::MAX_UNREWARDED_RELAYER_ENTRIES_AT_INBOUND_LANE,

bridges/relays/substrate/src/rialto_messages_to_millau.rs

@@ -105,7 +105,6 @@ pub fn run(
 	lane_id: LaneId,
 	metrics_params: Option<MetricsParams>,
 ) {
-	let reconnect_delay = Duration::from_secs(10);
 	let stall_timeout = Duration::from_secs(5 * 60);
 	let relayer_id_at_rialto = rialto_sign.signer.public().as_array_ref().clone().into();
 
@@ -134,7 +133,7 @@ pub fn run(
 			lane: lane_id,
 			source_tick: Rialto::AVERAGE_BLOCK_INTERVAL,
 			target_tick: Millau::AVERAGE_BLOCK_INTERVAL,
-			reconnect_delay,
+			reconnect_delay: relay_utils::relay_loop::RECONNECT_DELAY,
 			stall_timeout,
 			delivery_params: messages_relay::message_lane_loop::MessageDeliveryParams {
 				max_unrewarded_relayer_entries_at_target: bp_millau::MAX_UNREWARDED_RELAYER_ENTRIES_AT_INBOUND_LANE,

bridges/relays/utils/Cargo.toml

@@ -8,6 +8,7 @@ license = "GPL-3.0-or-later WITH Classpath-exception-2.0"
 [dependencies]
 ansi_term = "0.12"
 async-std = "1.6.5"
+async-trait = "0.1.40"
 backoff = "0.2"
 env_logger = "0.8.2"
 futures = "0.3.5"

bridges/relays/utils/src/lib.rs

@@ -29,6 +29,7 @@ pub const CONNECTION_ERROR_DELAY: Duration = Duration::from_secs(10);
 
 pub mod initialize;
 pub mod metrics;
+pub mod relay_loop;
 
 /// Block number traits shared by all chains that relay is able to serve.
 pub trait BlockNumberBase:

bridges/relays/utils/src/metrics.rs

@@ -16,6 +16,7 @@
 
 pub use substrate_prometheus_endpoint::{register, Counter, CounterVec, Gauge, GaugeVec, Opts, Registry, F64, U64};
 
+use async_std::sync::{Arc, Mutex};
 use std::net::SocketAddr;
 use substrate_prometheus_endpoint::init_prometheus;
 use sysinfo::{ProcessExt, RefreshKind, System, SystemExt};
@@ -36,9 +37,9 @@ pub trait Metrics {
 }
 
 /// Global Prometheus metrics.
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 pub struct GlobalMetrics {
-	system: System,
+	system: Arc<Mutex<System>>,
 	system_average_load: GaugeVec<F64>,
 	process_cpu_usage_percentage: Gauge<F64>,
 	process_memory_usage_bytes: Gauge<U64>,
@@ -110,7 +111,7 @@ impl Metrics for GlobalMetrics {
 impl Default for GlobalMetrics {
 	fn default() -> Self {
 		GlobalMetrics {
-			system: System::new_with_specifics(RefreshKind::everything()),
+			system: Arc::new(Mutex::new(System::new_with_specifics(RefreshKind::everything()))),
 			system_average_load: GaugeVec::new(Opts::new("system_average_load", "System load average"), &["over"])
 				.expect("metric is static and thus valid; qed"),
 			process_cpu_usage_percentage: Gauge::new("process_cpu_usage_percentage", "Process CPU usage")
@@ -126,9 +127,10 @@ impl Default for GlobalMetrics {
 
 impl GlobalMetrics {
 	/// Update metrics.
-	pub fn update(&mut self) {
+	pub async fn update(&self) {
 		// update system-wide metrics
-		let load = self.system.get_load_average();
+		let mut system = self.system.lock().await;
+		let load = system.get_load_average();
 		self.system_average_load.with_label_values(&["1min"]).set(load.one);
 		self.system_average_load.with_label_values(&["5min"]).set(load.five);
 		self.system_average_load.with_label_values(&["15min"]).set(load.fifteen);
@@ -139,8 +141,8 @@ impl GlobalMetrics {
 				relay is not supposed to run in such MetricsParamss;\
 				qed",
 		);
-		let is_process_refreshed = self.system.refresh_process(pid);
-		match (is_process_refreshed, self.system.get_process(pid)) {
+		let is_process_refreshed = system.refresh_process(pid);
+		match (is_process_refreshed, system.get_process(pid)) {
 			(true, Some(process_info)) => {
 				let cpu_usage = process_info.cpu_usage() as f64;
 				let memory_usage = process_info.memory() * 1024;

bridges/relays/utils/src/relay_loop.rs

@@ -0,0 +1,95 @@
+// Copyright 2019-2020 Parity Technologies (UK) Ltd.
+// This file is part of Parity Bridges Common.
+
+// Parity Bridges Common is free software: you can redistribute it and/or modify
+// it under the terms of the GNU General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+
+// Parity Bridges Common is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU General Public License for more details.
+
+// You should have received a copy of the GNU General Public License
+// along with Parity Bridges Common.  If not, see <http://www.gnu.org/licenses/>.
+
+use crate::{FailedClient, MaybeConnectionError};
+
+use async_trait::async_trait;
+use std::{fmt::Debug, future::Future, time::Duration};
+
+/// Default pause between reconnect attempts.
+pub const RECONNECT_DELAY: Duration = Duration::from_secs(10);
+
+/// Basic blockchain client from relay perspective.
+#[async_trait]
+pub trait Client: Clone + Send + Sync {
+	/// Type of error this clients returns.
+	type Error: Debug + MaybeConnectionError;
+
+	/// Try to reconnect to source node.
+	async fn reconnect(&mut self) -> Result<(), Self::Error>;
+}
+
+/// Run relay loop.
+///
+/// This function represents an outer loop, which in turn calls provided `loop_run` function to do
+/// actual job. When `loop_run` returns, this outer loop reconnects to failed client (source,
+/// target or both) and calls `loop_run` again.
+pub fn run<SC: Client, TC: Client, R, F>(
+	reconnect_delay: Duration,
+	mut source_client: SC,
+	mut target_client: TC,
+	loop_run: R,
+) where
+	R: Fn(SC, TC) -> F,
+	F: Future<Output = Result<(), FailedClient>>,
+{
+	let mut local_pool = futures::executor::LocalPool::new();
+
+	local_pool.run_until(async move {
+		loop {
+			let result = loop_run(source_client.clone(), target_client.clone()).await;
+
+			match result {
+				Ok(()) => break,
+				Err(failed_client) => loop {
+					async_std::task::sleep(reconnect_delay).await;
+					if failed_client == FailedClient::Both || failed_client == FailedClient::Source {
+						match source_client.reconnect().await {
+							Ok(()) => (),
+							Err(error) => {
+								log::warn!(
+									target: "bridge",
+									"Failed to reconnect to source client. Going to retry in {}s: {:?}",
+									reconnect_delay.as_secs(),
+									error,
+								);
+								continue;
+							}
+						}
+					}
+					if failed_client == FailedClient::Both || failed_client == FailedClient::Target {
+						match target_client.reconnect().await {
+							Ok(()) => (),
+							Err(error) => {
+								log::warn!(
+									target: "bridge",
+									"Failed to reconnect to target client. Going to retry in {}s: {:?}",
+									reconnect_delay.as_secs(),
+									error,
+								);
+								continue;
+							}
+						}
+					}
+
+					break;
+				},
+			}
+
+			log::debug!(target: "bridge", "Restarting relay loop");
+		}
+	});
+}