Do not run unneeded subsystems on collator and its alongside node (#3061)

Currently, collators and their alongside nodes spin up a full-scale
overseer running a bunch of subsystems that are not needed if the node
is not a validator. That was considered to be harmless; however, we've
got problems with unused subsystems getting stalled for a reason not
currently known, resulting in the overseer exiting and bringing down the
whole node.

This PR aims to only run needed subsystems on such nodes, replacing the
rest with `DummySubsystem`.

It also enables collator-optimized availability recovery subsystem
implementation.

Partially solves #1730.

cumulus/client/relay-chain-inprocess-interface/src/lib.rs

@@ -295,7 +295,7 @@ fn build_polkadot_full_node(
 			workers_path: None,
 			workers_names: None,
 
-			overseer_gen: polkadot_service::RealOverseerGen,
+			overseer_gen: polkadot_service::CollatorOverseerGen,
 			overseer_message_channel_capacity_override: None,
 			malus_finality_delay: None,
 			hwbench,

cumulus/test/service/src/lib.rs

@@ -271,6 +271,7 @@ async fn build_relay_chain_interface(
 				polkadot_service::IsParachainNode::Collator(CollatorPair::generate().0)
 			},
 			None,
+			polkadot_service::CollatorOverseerGen,
 		)
 		.map_err(|e| RelayChainError::Application(Box::new(e) as Box<_>))?,
 		cumulus_client_cli::RelayChainMode::ExternalRpc(rpc_target_urls) =>

polkadot/cli/src/command.rs

@@ -299,7 +299,7 @@ pub fn run() -> Result<()> {
 	match &cli.subcommand {
 		None => run_node_inner(
 			cli,
-			service::RealOverseerGen,
+			service::ValidatorOverseerGen,
 			None,
 			polkadot_node_metrics::logger_hook(),
 		),

polkadot/cli/src/lib.rs

@@ -29,7 +29,7 @@ mod error;
 pub use service::{self, Block, CoreApi, IdentifyVariant, ProvideRuntimeApi, TFullClient};
 
 #[cfg(feature = "malus")]
-pub use service::overseer::prepared_overseer_builder;
+pub use service::overseer::validator_overseer_builder;
 
 #[cfg(feature = "cli")]
 pub use cli::*;

polkadot/node/core/pvf-checker/src/lib.rs

@@ -49,29 +49,24 @@ use self::{
 
 /// PVF pre-checking subsystem.
 pub struct PvfCheckerSubsystem {
-	enabled: bool,
 	keystore: KeystorePtr,
 	metrics: Metrics,
 }
 
 impl PvfCheckerSubsystem {
-	pub fn new(enabled: bool, keystore: KeystorePtr, metrics: Metrics) -> Self {
+	pub fn new(keystore: KeystorePtr, metrics: Metrics) -> Self {
-		PvfCheckerSubsystem { enabled, keystore, metrics }
+		PvfCheckerSubsystem { keystore, metrics }
 	}
 }
 
 #[overseer::subsystem(PvfChecker, error=SubsystemError, prefix = self::overseer)]
 impl<Context> PvfCheckerSubsystem {
 	fn start(self, ctx: Context) -> SpawnedSubsystem {
-		if self.enabled {
 		let future = run(ctx, self.keystore, self.metrics)
 			.map_err(|e| SubsystemError::with_origin("pvf-checker", e))
 			.boxed();
 
 		SpawnedSubsystem { name: "pvf-checker-subsystem", future }
-		} else {
-			polkadot_overseer::DummySubsystem.start(ctx)
-		}
 	}
 }
 

polkadot/node/malus/src/variants/back_garbage_candidate.rs

@@ -19,13 +19,12 @@
 //! candidates.
 
 use polkadot_cli::{
-	prepared_overseer_builder,
 	service::{
-		AuthorityDiscoveryApi, AuxStore, BabeApi, Block, Error, HeaderBackend, Overseer,
+		AuthorityDiscoveryApi, AuxStore, BabeApi, Block, Error, ExtendedOverseerGenArgs,
-		OverseerConnector, OverseerGen, OverseerGenArgs, OverseerHandle, ParachainHost,
+		HeaderBackend, Overseer, OverseerConnector, OverseerGen, OverseerGenArgs, OverseerHandle,
-		ProvideRuntimeApi,
+		ParachainHost, ProvideRuntimeApi,
 	},
-	Cli,
+	validator_overseer_builder, Cli,
 };
 use polkadot_node_subsystem::SpawnGlue;
 use polkadot_node_subsystem_types::DefaultSubsystemClient;
@@ -63,6 +62,7 @@ impl OverseerGen for BackGarbageCandidates {
 		&self,
 		connector: OverseerConnector,
 		args: OverseerGenArgs<'_, Spawner, RuntimeClient>,
+		ext_args: Option<ExtendedOverseerGenArgs>,
 	) -> Result<
 		(Overseer<SpawnGlue<Spawner>, Arc<DefaultSubsystemClient<RuntimeClient>>>, OverseerHandle),
 		Error,
@@ -80,7 +80,10 @@ impl OverseerGen for BackGarbageCandidates {
 			SpawnGlue(spawner),
 		);
 
-		prepared_overseer_builder(args)?
+		validator_overseer_builder(
+			args,
+			ext_args.expect("Extended arguments required to build validator overseer are provided"),
+		)?
 		.replace_candidate_validation(move |cv_subsystem| {
 			InterceptedSubsystem::new(cv_subsystem, validation_filter)
 		})

polkadot/node/malus/src/variants/dispute_finalized_candidates.rs

@@ -33,13 +33,12 @@
 
 use futures::channel::oneshot;
 use polkadot_cli::{
-	prepared_overseer_builder,
 	service::{
-		AuthorityDiscoveryApi, AuxStore, BabeApi, Block, Error, HeaderBackend, Overseer,
+		AuthorityDiscoveryApi, AuxStore, BabeApi, Block, Error, ExtendedOverseerGenArgs,
-		OverseerConnector, OverseerGen, OverseerGenArgs, OverseerHandle, ParachainHost,
+		HeaderBackend, Overseer, OverseerConnector, OverseerGen, OverseerGenArgs, OverseerHandle,
-		ProvideRuntimeApi,
+		ParachainHost, ProvideRuntimeApi,
 	},
-	Cli,
+	validator_overseer_builder, Cli,
 };
 use polkadot_node_subsystem::{messages::ApprovalVotingMessage, SpawnGlue};
 use polkadot_node_subsystem_types::{DefaultSubsystemClient, OverseerSignal};
@@ -237,6 +236,7 @@ impl OverseerGen for DisputeFinalizedCandidates {
 		&self,
 		connector: OverseerConnector,
 		args: OverseerGenArgs<'_, Spawner, RuntimeClient>,
+		ext_args: Option<ExtendedOverseerGenArgs>,
 	) -> Result<
 		(Overseer<SpawnGlue<Spawner>, Arc<DefaultSubsystemClient<RuntimeClient>>>, OverseerHandle),
 		Error,
@@ -257,7 +257,10 @@ impl OverseerGen for DisputeFinalizedCandidates {
 			dispute_offset: self.dispute_offset,
 		};
 
-		prepared_overseer_builder(args)?
+		validator_overseer_builder(
+			args,
+			ext_args.expect("Extended arguments required to build validator overseer are provided"),
+		)?
 		.replace_approval_voting(move |cb| InterceptedSubsystem::new(cb, ancestor_disputer))
 		.build_with_connector(connector)
 		.map_err(|e| e.into())

polkadot/node/malus/src/variants/dispute_valid_candidates.rs

@@ -23,13 +23,12 @@
 #![allow(missing_docs)]
 
 use polkadot_cli::{
-	prepared_overseer_builder,
 	service::{
-		AuthorityDiscoveryApi, AuxStore, BabeApi, Block, Error, HeaderBackend, Overseer,
+		AuthorityDiscoveryApi, AuxStore, BabeApi, Block, Error, ExtendedOverseerGenArgs,
-		OverseerConnector, OverseerGen, OverseerGenArgs, OverseerHandle, ParachainHost,
+		HeaderBackend, Overseer, OverseerConnector, OverseerGen, OverseerGenArgs, OverseerHandle,
-		ProvideRuntimeApi,
+		ParachainHost, ProvideRuntimeApi,
 	},
-	Cli,
+	validator_overseer_builder, Cli,
 };
 use polkadot_node_subsystem::SpawnGlue;
 use polkadot_node_subsystem_types::DefaultSubsystemClient;
@@ -80,6 +79,7 @@ impl OverseerGen for DisputeValidCandidates {
 		&self,
 		connector: OverseerConnector,
 		args: OverseerGenArgs<'_, Spawner, RuntimeClient>,
+		ext_args: Option<ExtendedOverseerGenArgs>,
 	) -> Result<
 		(Overseer<SpawnGlue<Spawner>, Arc<DefaultSubsystemClient<RuntimeClient>>>, OverseerHandle),
 		Error,
@@ -97,7 +97,10 @@ impl OverseerGen for DisputeValidCandidates {
 			SpawnGlue(spawner.clone()),
 		);
 
-		prepared_overseer_builder(args)?
+		validator_overseer_builder(
+			args,
+			ext_args.expect("Extended arguments required to build validator overseer are provided"),
+		)?
 		.replace_candidate_validation(move |cv_subsystem| {
 			InterceptedSubsystem::new(cv_subsystem, validation_filter)
 		})

polkadot/node/malus/src/variants/suggest_garbage_candidate.rs

@@ -23,13 +23,12 @@
 #![allow(missing_docs)]
 
 use polkadot_cli::{
-	prepared_overseer_builder,
 	service::{
-		AuthorityDiscoveryApi, AuxStore, BabeApi, Block, Error, HeaderBackend, Overseer,
+		AuthorityDiscoveryApi, AuxStore, BabeApi, Block, Error, ExtendedOverseerGenArgs,
-		OverseerConnector, OverseerGen, OverseerGenArgs, OverseerHandle, ParachainHost,
+		HeaderBackend, Overseer, OverseerConnector, OverseerGen, OverseerGenArgs, OverseerHandle,
-		ProvideRuntimeApi,
+		ParachainHost, ProvideRuntimeApi,
 	},
-	Cli,
+	validator_overseer_builder, Cli,
 };
 use polkadot_node_core_candidate_validation::find_validation_data;
 use polkadot_node_primitives::{AvailableData, BlockData, PoV};
@@ -266,6 +265,7 @@ impl OverseerGen for SuggestGarbageCandidates {
 		&self,
 		connector: OverseerConnector,
 		args: OverseerGenArgs<'_, Spawner, RuntimeClient>,
+		ext_args: Option<ExtendedOverseerGenArgs>,
 	) -> Result<
 		(Overseer<SpawnGlue<Spawner>, Arc<DefaultSubsystemClient<RuntimeClient>>>, OverseerHandle),
 		Error,
@@ -293,11 +293,12 @@ impl OverseerGen for SuggestGarbageCandidates {
 			SpawnGlue(args.spawner.clone()),
 		);
 
-		prepared_overseer_builder(args)?
+		validator_overseer_builder(
+			args,
+			ext_args.expect("Extended arguments required to build validator overseer are provided"),
+		)?
 		.replace_candidate_backing(move |cb| InterceptedSubsystem::new(cb, note_candidate))
-			.replace_candidate_validation(move |cb| {
+		.replace_candidate_validation(move |cb| InterceptedSubsystem::new(cb, validation_filter))
-				InterceptedSubsystem::new(cb, validation_filter)
-			})
 		.build_with_connector(connector)
 		.map_err(|e| e.into())
 	}

polkadot/node/malus/src/variants/support_disabled.rs

@@ -18,13 +18,12 @@
 //! to always return an empty set of disabled validators.
 
 use polkadot_cli::{
-	prepared_overseer_builder,
 	service::{
-		AuthorityDiscoveryApi, AuxStore, BabeApi, Block, Error, HeaderBackend, Overseer,
+		AuthorityDiscoveryApi, AuxStore, BabeApi, Block, Error, ExtendedOverseerGenArgs,
-		OverseerConnector, OverseerGen, OverseerGenArgs, OverseerHandle, ParachainHost,
+		HeaderBackend, Overseer, OverseerConnector, OverseerGen, OverseerGenArgs, OverseerHandle,
-		ProvideRuntimeApi,
+		ParachainHost, ProvideRuntimeApi,
 	},
-	Cli,
+	validator_overseer_builder, Cli,
 };
 use polkadot_node_subsystem::SpawnGlue;
 use polkadot_node_subsystem_types::DefaultSubsystemClient;
@@ -50,6 +49,7 @@ impl OverseerGen for SupportDisabled {
 		&self,
 		connector: OverseerConnector,
 		args: OverseerGenArgs<'_, Spawner, RuntimeClient>,
+		ext_args: Option<ExtendedOverseerGenArgs>,
 	) -> Result<
 		(Overseer<SpawnGlue<Spawner>, Arc<DefaultSubsystemClient<RuntimeClient>>>, OverseerHandle),
 		Error,
@@ -59,7 +59,10 @@ impl OverseerGen for SupportDisabled {
 		RuntimeClient::Api: ParachainHost<Block> + BabeApi<Block> + AuthorityDiscoveryApi<Block>,
 		Spawner: 'static + SpawnNamed + Clone + Unpin,
 	{
-		prepared_overseer_builder(args)?
+		validator_overseer_builder(
+			args,
+			ext_args.expect("Extended arguments required to build validator overseer are provided"),
+		)?
 		.replace_runtime_api(move |ra_subsystem| {
 			InterceptedSubsystem::new(ra_subsystem, IgnoreDisabled)
 		})

polkadot/node/service/src/lib.rs

@@ -31,7 +31,10 @@ pub mod overseer;
 pub mod workers;
 
 #[cfg(feature = "full-node")]
-pub use self::overseer::{OverseerGen, OverseerGenArgs, RealOverseerGen};
+pub use self::overseer::{
+	CollatorOverseerGen, ExtendedOverseerGenArgs, OverseerGen, OverseerGenArgs,
+	ValidatorOverseerGen,
+};
 
 #[cfg(test)]
 mod tests;
@@ -775,8 +778,6 @@ pub fn new_full<OverseerGenerator: OverseerGen>(
 
 	let keystore = basics.keystore_container.local_keystore();
 	let auth_or_collator = role.is_authority() || is_parachain_node.is_collator();
-	// We only need to enable the pvf checker when this is a validator.
-	let pvf_checker_enabled = role.is_authority();
 
 	let select_chain = if auth_or_collator {
 		let metrics =
@@ -867,10 +868,6 @@ pub fn new_full<OverseerGenerator: OverseerGen>(
 
 	let req_protocol_names = ReqProtocolNames::new(&genesis_hash, config.chain_spec.fork_id());
 
-	let (pov_req_receiver, cfg) = IncomingRequest::get_config_receiver(&req_protocol_names);
-	net_config.add_request_response_protocol(cfg);
-	let (chunk_req_receiver, cfg) = IncomingRequest::get_config_receiver(&req_protocol_names);
-	net_config.add_request_response_protocol(cfg);
 	let (collation_req_v1_receiver, cfg) =
 		IncomingRequest::get_config_receiver(&req_protocol_names);
 	net_config.add_request_response_protocol(cfg);
@@ -880,12 +877,9 @@ pub fn new_full<OverseerGenerator: OverseerGen>(
 	let (available_data_req_receiver, cfg) =
 		IncomingRequest::get_config_receiver(&req_protocol_names);
 	net_config.add_request_response_protocol(cfg);
-	let (statement_req_receiver, cfg) = IncomingRequest::get_config_receiver(&req_protocol_names);
+	let (pov_req_receiver, cfg) = IncomingRequest::get_config_receiver(&req_protocol_names);
 	net_config.add_request_response_protocol(cfg);
-	let (candidate_req_v2_receiver, cfg) =
+	let (chunk_req_receiver, cfg) = IncomingRequest::get_config_receiver(&req_protocol_names);
-		IncomingRequest::get_config_receiver(&req_protocol_names);
-	net_config.add_request_response_protocol(cfg);
-	let (dispute_req_receiver, cfg) = IncomingRequest::get_config_receiver(&req_protocol_names);
 	net_config.add_request_response_protocol(cfg);
 
 	let grandpa_hard_forks = if config.chain_spec.is_kusama() {
@@ -900,6 +894,69 @@ pub fn new_full<OverseerGenerator: OverseerGen>(
 		grandpa_hard_forks,
 	));
 
+	let ext_overseer_args = if is_parachain_node.is_running_alongside_parachain_node() {
+		None
+	} else {
+		let parachains_db = open_database(&config.database)?;
+		let candidate_validation_config = if role.is_authority() {
+			let (prep_worker_path, exec_worker_path) = workers::determine_workers_paths(
+				workers_path,
+				workers_names,
+				node_version.clone(),
+			)?;
+			log::info!("🚀 Using prepare-worker binary at: {:?}", prep_worker_path);
+			log::info!("🚀 Using execute-worker binary at: {:?}", exec_worker_path);
+
+			Some(CandidateValidationConfig {
+				artifacts_cache_path: config
+					.database
+					.path()
+					.ok_or(Error::DatabasePathRequired)?
+					.join("pvf-artifacts"),
+				node_version,
+				secure_validator_mode,
+				prep_worker_path,
+				exec_worker_path,
+			})
+		} else {
+			None
+		};
+		let (statement_req_receiver, cfg) =
+			IncomingRequest::get_config_receiver(&req_protocol_names);
+		net_config.add_request_response_protocol(cfg);
+		let (candidate_req_v2_receiver, cfg) =
+			IncomingRequest::get_config_receiver(&req_protocol_names);
+		net_config.add_request_response_protocol(cfg);
+		let (dispute_req_receiver, cfg) = IncomingRequest::get_config_receiver(&req_protocol_names);
+		net_config.add_request_response_protocol(cfg);
+		let approval_voting_config = ApprovalVotingConfig {
+			col_approval_data: parachains_db::REAL_COLUMNS.col_approval_data,
+			slot_duration_millis: slot_duration.as_millis() as u64,
+		};
+		let dispute_coordinator_config = DisputeCoordinatorConfig {
+			col_dispute_data: parachains_db::REAL_COLUMNS.col_dispute_coordinator_data,
+		};
+		let chain_selection_config = ChainSelectionConfig {
+			col_data: parachains_db::REAL_COLUMNS.col_chain_selection_data,
+			stagnant_check_interval: Default::default(),
+			stagnant_check_mode: chain_selection_subsystem::StagnantCheckMode::PruneOnly,
+		};
+		Some(ExtendedOverseerGenArgs {
+			keystore,
+			parachains_db,
+			candidate_validation_config,
+			availability_config: AVAILABILITY_CONFIG,
+			pov_req_receiver,
+			chunk_req_receiver,
+			statement_req_receiver,
+			candidate_req_v2_receiver,
+			approval_voting_config,
+			dispute_req_receiver,
+			dispute_coordinator_config,
+			chain_selection_config,
+		})
+	};
+
 	let (network, system_rpc_tx, tx_handler_controller, network_starter, sync_service) =
 		service::build_network(service::BuildNetworkParams {
 			config: &config,
@@ -936,44 +993,6 @@ pub fn new_full<OverseerGenerator: OverseerGen>(
 		);
 	}
 
-	let parachains_db = open_database(&config.database)?;
-
-	let approval_voting_config = ApprovalVotingConfig {
-		col_approval_data: parachains_db::REAL_COLUMNS.col_approval_data,
-		slot_duration_millis: slot_duration.as_millis() as u64,
-	};
-
-	let candidate_validation_config = if role.is_authority() {
-		let (prep_worker_path, exec_worker_path) =
-			workers::determine_workers_paths(workers_path, workers_names, node_version.clone())?;
-		log::info!("🚀 Using prepare-worker binary at: {:?}", prep_worker_path);
-		log::info!("🚀 Using execute-worker binary at: {:?}", exec_worker_path);
-
-		Some(CandidateValidationConfig {
-			artifacts_cache_path: config
-				.database
-				.path()
-				.ok_or(Error::DatabasePathRequired)?
-				.join("pvf-artifacts"),
-			node_version,
-			secure_validator_mode,
-			prep_worker_path,
-			exec_worker_path,
-		})
-	} else {
-		None
-	};
-
-	let chain_selection_config = ChainSelectionConfig {
-		col_data: parachains_db::REAL_COLUMNS.col_chain_selection_data,
-		stagnant_check_interval: Default::default(),
-		stagnant_check_mode: chain_selection_subsystem::StagnantCheckMode::PruneOnly,
-	};
-
-	let dispute_coordinator_config = DisputeCoordinatorConfig {
-		col_dispute_data: parachains_db::REAL_COLUMNS.col_dispute_coordinator_data,
-	};
-
 	let rpc_handlers = service::spawn_tasks(service::SpawnTasksParams {
 		config,
 		backend: backend.clone(),
@@ -1067,29 +1086,16 @@ pub fn new_full<OverseerGenerator: OverseerGen>(
 			.generate::<service::SpawnTaskHandle, FullClient>(
 				overseer_connector,
 				OverseerGenArgs {
-					keystore,
 					runtime_client: overseer_client.clone(),
-					parachains_db,
 					network_service: network.clone(),
 					sync_service: sync_service.clone(),
 					authority_discovery_service,
-					pov_req_receiver,
-					chunk_req_receiver,
 					collation_req_v1_receiver,
 					collation_req_v2_receiver,
 					available_data_req_receiver,
-					statement_req_receiver,
-					candidate_req_v2_receiver,
-					dispute_req_receiver,
 					registry: prometheus_registry.as_ref(),
 					spawner,
 					is_parachain_node,
-					approval_voting_config,
-					availability_config: AVAILABILITY_CONFIG,
-					candidate_validation_config,
-					chain_selection_config,
-					dispute_coordinator_config,
-					pvf_checker_enabled,
 					overseer_message_channel_capacity_override,
 					req_protocol_names,
 					peerset_protocol_names,
@@ -1098,6 +1104,7 @@ pub fn new_full<OverseerGenerator: OverseerGen>(
 					),
 					notification_services,
 				},
+				ext_overseer_args,
 			)
 			.map_err(|e| {
 				gum::error!("Failed to init overseer: {}", e);

polkadot/node/service/src/overseer.rs

@@ -16,6 +16,7 @@
 
 use super::{AuthorityDiscoveryApi, Block, Error, Hash, IsParachainNode, Registry};
 use polkadot_node_subsystem_types::DefaultSubsystemClient;
+use polkadot_overseer::{DummySubsystem, InitializedOverseerBuilder, SubsystemError};
 use sc_transaction_pool_api::OffchainTransactionPoolFactory;
 use sp_core::traits::SpawnNamed;
 
@@ -32,13 +33,10 @@ use polkadot_node_network_protocol::{
 	},
 };
 #[cfg(any(feature = "malus", test))]
-pub use polkadot_overseer::{
+pub use polkadot_overseer::{dummy::dummy_overseer_builder, HeadSupportsParachains};
-	dummy::{dummy_overseer_builder, DummySubsystem},
-	HeadSupportsParachains,
-};
 use polkadot_overseer::{
-	metrics::Metrics as OverseerMetrics, InitializedOverseerBuilder, MetricsTrait, Overseer,
+	metrics::Metrics as OverseerMetrics, MetricsTrait, Overseer, OverseerConnector, OverseerHandle,
-	OverseerConnector, OverseerHandle, SpawnGlue,
+	SpawnGlue,
 };
 
 use parking_lot::Mutex;
@@ -86,22 +84,14 @@ where
 	RuntimeClient::Api: ParachainHost<Block> + BabeApi<Block> + AuthorityDiscoveryApi<Block>,
 	Spawner: 'static + SpawnNamed + Clone + Unpin,
 {
-	/// The keystore to use for i.e. validator keys.
-	pub keystore: Arc<LocalKeystore>,
 	/// Runtime client generic, providing the `ProvieRuntimeApi` trait besides others.
 	pub runtime_client: Arc<RuntimeClient>,
-	/// The underlying key value store for the parachains.
-	pub parachains_db: Arc<dyn polkadot_node_subsystem_util::database::Database>,
 	/// Underlying network service implementation.
 	pub network_service: Arc<sc_network::NetworkService<Block, Hash>>,
 	/// Underlying syncing service implementation.
 	pub sync_service: Arc<sc_network_sync::SyncingService<Block>>,
 	/// Underlying authority discovery service.
 	pub authority_discovery_service: AuthorityDiscoveryService,
-	/// POV request receiver.
-	pub pov_req_receiver: IncomingRequestReceiver<request_v1::PoVFetchingRequest>,
-	/// Erasure chunks request receiver.
-	pub chunk_req_receiver: IncomingRequestReceiver<request_v1::ChunkFetchingRequest>,
 	/// Collations request receiver for network protocol v1.
 	pub collation_req_v1_receiver: IncomingRequestReceiver<request_v1::CollationFetchingRequest>,
 	/// Collations request receiver for network protocol v2.
@@ -109,30 +99,12 @@ where
 	/// Receiver for available data requests.
 	pub available_data_req_receiver:
 		IncomingRequestReceiver<request_v1::AvailableDataFetchingRequest>,
-	/// Receiver for incoming large statement requests.
-	pub statement_req_receiver: IncomingRequestReceiver<request_v1::StatementFetchingRequest>,
-	/// Receiver for incoming candidate requests.
-	pub candidate_req_v2_receiver: IncomingRequestReceiver<request_v2::AttestedCandidateRequest>,
-	/// Receiver for incoming disputes.
-	pub dispute_req_receiver: IncomingRequestReceiver<request_v1::DisputeRequest>,
 	/// Prometheus registry, commonly used for production systems, less so for test.
 	pub registry: Option<&'a Registry>,
 	/// Task spawner to be used throughout the overseer and the APIs it provides.
 	pub spawner: Spawner,
 	/// Determines the behavior of the collator.
 	pub is_parachain_node: IsParachainNode,
-	/// Configuration for the approval voting subsystem.
-	pub approval_voting_config: ApprovalVotingConfig,
-	/// Configuration for the availability store subsystem.
-	pub availability_config: AvailabilityConfig,
-	/// Configuration for the candidate validation subsystem.
-	pub candidate_validation_config: Option<CandidateValidationConfig>,
-	/// Configuration for the chain selection subsystem.
-	pub chain_selection_config: ChainSelectionConfig,
-	/// Configuration for the dispute coordinator subsystem.
-	pub dispute_coordinator_config: DisputeCoordinatorConfig,
-	/// Enable PVF pre-checking
-	pub pvf_checker_enabled: bool,
 	/// Overseer channel capacity override.
 	pub overseer_message_channel_capacity_override: Option<usize>,
 	/// Request-response protocol names source.
@@ -145,39 +117,66 @@ where
 	pub notification_services: HashMap<PeerSet, Box<dyn NotificationService>>,
 }
 
-/// Obtain a prepared `OverseerBuilder`, that is initialized
+pub struct ExtendedOverseerGenArgs {
-/// with all default values.
+	/// The keystore to use for i.e. validator keys.
-pub fn prepared_overseer_builder<Spawner, RuntimeClient>(
+	pub keystore: Arc<LocalKeystore>,
+	/// The underlying key value store for the parachains.
+	pub parachains_db: Arc<dyn polkadot_node_subsystem_util::database::Database>,
+	/// Configuration for the candidate validation subsystem.
+	pub candidate_validation_config: Option<CandidateValidationConfig>,
+	/// Configuration for the availability store subsystem.
+	pub availability_config: AvailabilityConfig,
+	/// POV request receiver.
+	pub pov_req_receiver: IncomingRequestReceiver<request_v1::PoVFetchingRequest>,
+	/// Erasure chunks request receiver.
+	pub chunk_req_receiver: IncomingRequestReceiver<request_v1::ChunkFetchingRequest>,
+	/// Receiver for incoming large statement requests.
+	pub statement_req_receiver: IncomingRequestReceiver<request_v1::StatementFetchingRequest>,
+	/// Receiver for incoming candidate requests.
+	pub candidate_req_v2_receiver: IncomingRequestReceiver<request_v2::AttestedCandidateRequest>,
+	/// Configuration for the approval voting subsystem.
+	pub approval_voting_config: ApprovalVotingConfig,
+	/// Receiver for incoming disputes.
+	pub dispute_req_receiver: IncomingRequestReceiver<request_v1::DisputeRequest>,
+	/// Configuration for the dispute coordinator subsystem.
+	pub dispute_coordinator_config: DisputeCoordinatorConfig,
+	/// Configuration for the chain selection subsystem.
+	pub chain_selection_config: ChainSelectionConfig,
+}
+
+/// Obtain a prepared validator `Overseer`, that is initialized with all default values.
+pub fn validator_overseer_builder<Spawner, RuntimeClient>(
 	OverseerGenArgs {
-		keystore,
 		runtime_client,
-		parachains_db,
 		network_service,
 		sync_service,
 		authority_discovery_service,
-		pov_req_receiver,
+		collation_req_v1_receiver: _,
-		chunk_req_receiver,
+		collation_req_v2_receiver: _,
-		collation_req_v1_receiver,
-		collation_req_v2_receiver,
 		available_data_req_receiver,
-		statement_req_receiver,
-		candidate_req_v2_receiver,
-		dispute_req_receiver,
 		registry,
 		spawner,
 		is_parachain_node,
-		approval_voting_config,
-		availability_config,
-		candidate_validation_config,
-		chain_selection_config,
-		dispute_coordinator_config,
-		pvf_checker_enabled,
 		overseer_message_channel_capacity_override,
 		req_protocol_names,
 		peerset_protocol_names,
 		offchain_transaction_pool_factory,
 		notification_services,
 	}: OverseerGenArgs<Spawner, RuntimeClient>,
+	ExtendedOverseerGenArgs {
+		keystore,
+		parachains_db,
+		candidate_validation_config,
+		availability_config,
+		pov_req_receiver,
+		chunk_req_receiver,
+		statement_req_receiver,
+		candidate_req_v2_receiver,
+		approval_voting_config,
+		dispute_req_receiver,
+		dispute_coordinator_config,
+		chain_selection_config,
+	}: ExtendedOverseerGenArgs,
 ) -> Result<
 	InitializedOverseerBuilder<
 		SpawnGlue<Spawner>,
@@ -280,23 +279,15 @@ where
 			Metrics::register(registry)?, // candidate-validation metrics
 			Metrics::register(registry)?, // validation host metrics
 		))
-		.pvf_checker(PvfCheckerSubsystem::new(
+		.pvf_checker(PvfCheckerSubsystem::new(keystore.clone(), Metrics::register(registry)?))
-			pvf_checker_enabled,
-			keystore.clone(),
-			Metrics::register(registry)?,
-		))
 		.chain_api(ChainApiSubsystem::new(runtime_client.clone(), Metrics::register(registry)?))
 		.collation_generation(CollationGenerationSubsystem::new(Metrics::register(registry)?))
 		.collator_protocol({
 			let side = match is_parachain_node {
-				IsParachainNode::Collator(collator_pair) => ProtocolSide::Collator {
+				IsParachainNode::Collator(_) | IsParachainNode::FullNode =>
-					peer_id: network_service.local_peer_id(),
+					return Err(Error::Overseer(SubsystemError::Context(
-					collator_pair,
+						"build validator overseer for parachain node".to_owned(),
-					request_receiver_v1: collation_req_v1_receiver,
+					))),
-					request_receiver_v2: collation_req_v2_receiver,
-					metrics: Metrics::register(registry)?,
-				},
-				IsParachainNode::FullNode => ProtocolSide::None,
 				IsParachainNode::No => ProtocolSide::Validator {
 					keystore: keystore.clone(),
 					eviction_policy: Default::default(),
@@ -352,23 +343,173 @@ where
 		.metrics(metrics)
 		.spawner(spawner);
 
-	if let Some(capacity) = overseer_message_channel_capacity_override {
+	let builder = if let Some(capacity) = overseer_message_channel_capacity_override {
-		Ok(builder.message_channel_capacity(capacity))
+		builder.message_channel_capacity(capacity)
 	} else {
+		builder
+	};
+	Ok(builder)
+}
+
+/// Obtain a prepared collator `Overseer`, that is initialized with all default values.
+pub fn collator_overseer_builder<Spawner, RuntimeClient>(
+	OverseerGenArgs {
+		runtime_client,
+		network_service,
+		sync_service,
+		authority_discovery_service,
+		collation_req_v1_receiver,
+		collation_req_v2_receiver,
+		available_data_req_receiver,
+		registry,
+		spawner,
+		is_parachain_node,
+		overseer_message_channel_capacity_override,
+		req_protocol_names,
+		peerset_protocol_names,
+		offchain_transaction_pool_factory,
+		notification_services,
+	}: OverseerGenArgs<Spawner, RuntimeClient>,
+) -> Result<
+	InitializedOverseerBuilder<
+		SpawnGlue<Spawner>,
+		Arc<DefaultSubsystemClient<RuntimeClient>>,
+		DummySubsystem,
+		DummySubsystem,
+		DummySubsystem,
+		DummySubsystem,
+		DummySubsystem,
+		AvailabilityRecoverySubsystem,
+		DummySubsystem,
+		DummySubsystem,
+		DummySubsystem,
+		RuntimeApiSubsystem<DefaultSubsystemClient<RuntimeClient>>,
+		DummySubsystem,
+		NetworkBridgeRxSubsystem<
+			Arc<sc_network::NetworkService<Block, Hash>>,
+			AuthorityDiscoveryService,
+		>,
+		NetworkBridgeTxSubsystem<
+			Arc<sc_network::NetworkService<Block, Hash>>,
+			AuthorityDiscoveryService,
+		>,
+		ChainApiSubsystem<RuntimeClient>,
+		CollationGenerationSubsystem,
+		CollatorProtocolSubsystem,
+		DummySubsystem,
+		DummySubsystem,
+		DummySubsystem,
+		DummySubsystem,
+		DummySubsystem,
+		DummySubsystem,
+		ProspectiveParachainsSubsystem,
+	>,
+	Error,
+>
+where
+	RuntimeClient: 'static + ProvideRuntimeApi<Block> + HeaderBackend<Block> + AuxStore,
+	RuntimeClient::Api: ParachainHost<Block> + BabeApi<Block> + AuthorityDiscoveryApi<Block>,
+	Spawner: 'static + SpawnNamed + Clone + Unpin,
+{
+	use polkadot_node_subsystem_util::metrics::Metrics;
+
+	let metrics = <OverseerMetrics as MetricsTrait>::register(registry)?;
+	let notification_sinks = Arc::new(Mutex::new(HashMap::new()));
+
+	let spawner = SpawnGlue(spawner);
+
+	let network_bridge_metrics: NetworkBridgeMetrics = Metrics::register(registry)?;
+
+	let runtime_api_client = Arc::new(DefaultSubsystemClient::new(
+		runtime_client.clone(),
+		offchain_transaction_pool_factory,
+	));
+
+	let builder = Overseer::builder()
+		.network_bridge_tx(NetworkBridgeTxSubsystem::new(
+			network_service.clone(),
+			authority_discovery_service.clone(),
+			network_bridge_metrics.clone(),
+			req_protocol_names,
+			peerset_protocol_names.clone(),
+			notification_sinks.clone(),
+		))
+		.network_bridge_rx(NetworkBridgeRxSubsystem::new(
+			network_service.clone(),
+			authority_discovery_service.clone(),
+			Box::new(sync_service.clone()),
+			network_bridge_metrics,
+			peerset_protocol_names,
+			notification_services,
+			notification_sinks,
+		))
+		.availability_distribution(DummySubsystem)
+		.availability_recovery(AvailabilityRecoverySubsystem::for_collator(
+			available_data_req_receiver,
+			Metrics::register(registry)?,
+		))
+		.availability_store(DummySubsystem)
+		.bitfield_distribution(DummySubsystem)
+		.bitfield_signing(DummySubsystem)
+		.candidate_backing(DummySubsystem)
+		.candidate_validation(DummySubsystem)
+		.pvf_checker(DummySubsystem)
+		.chain_api(ChainApiSubsystem::new(runtime_client.clone(), Metrics::register(registry)?))
+		.collation_generation(CollationGenerationSubsystem::new(Metrics::register(registry)?))
+		.collator_protocol({
+			let side = match is_parachain_node {
+				IsParachainNode::No =>
+					return Err(Error::Overseer(SubsystemError::Context(
+						"build parachain node overseer for validator".to_owned(),
+					))),
+				IsParachainNode::Collator(collator_pair) => ProtocolSide::Collator {
+					peer_id: network_service.local_peer_id(),
+					collator_pair,
+					request_receiver_v1: collation_req_v1_receiver,
+					request_receiver_v2: collation_req_v2_receiver,
+					metrics: Metrics::register(registry)?,
+				},
+				IsParachainNode::FullNode => ProtocolSide::None,
+			};
+			CollatorProtocolSubsystem::new(side)
+		})
+		.provisioner(DummySubsystem)
+		.runtime_api(RuntimeApiSubsystem::new(
+			runtime_api_client.clone(),
+			Metrics::register(registry)?,
+			spawner.clone(),
+		))
+		.statement_distribution(DummySubsystem)
+		.approval_distribution(DummySubsystem)
+		.approval_voting(DummySubsystem)
+		.gossip_support(DummySubsystem)
+		.dispute_coordinator(DummySubsystem)
+		.dispute_distribution(DummySubsystem)
+		.chain_selection(DummySubsystem)
+		.prospective_parachains(ProspectiveParachainsSubsystem::new(Metrics::register(registry)?))
+		.activation_external_listeners(Default::default())
+		.span_per_active_leaf(Default::default())
+		.active_leaves(Default::default())
+		.supports_parachains(runtime_api_client)
+		.metrics(metrics)
+		.spawner(spawner);
+
+	let builder = if let Some(capacity) = overseer_message_channel_capacity_override {
+		builder.message_channel_capacity(capacity)
+	} else {
+		builder
+	};
 	Ok(builder)
-	}
 }
 
 /// Trait for the `fn` generating the overseer.
-///
-/// Default behavior is to create an unmodified overseer, as `RealOverseerGen`
-/// would do.
 pub trait OverseerGen {
 	/// Overwrite the full generation of the overseer, including the subsystems.
 	fn generate<Spawner, RuntimeClient>(
 		&self,
 		connector: OverseerConnector,
 		args: OverseerGenArgs<Spawner, RuntimeClient>,
+		ext_args: Option<ExtendedOverseerGenArgs>,
 	) -> Result<
 		(Overseer<SpawnGlue<Spawner>, Arc<DefaultSubsystemClient<RuntimeClient>>>, OverseerHandle),
 		Error,
@@ -376,24 +517,22 @@ pub trait OverseerGen {
 	where
 		RuntimeClient: 'static + ProvideRuntimeApi<Block> + HeaderBackend<Block> + AuxStore,
 		RuntimeClient::Api: ParachainHost<Block> + BabeApi<Block> + AuthorityDiscoveryApi<Block>,
-		Spawner: 'static + SpawnNamed + Clone + Unpin,
+		Spawner: 'static + SpawnNamed + Clone + Unpin;
-	{
+
-		let gen = RealOverseerGen;
-		RealOverseerGen::generate::<Spawner, RuntimeClient>(&gen, connector, args)
-	}
 	// It would be nice to make `create_subsystems` part of this trait,
 	// but the amount of generic arguments that would be required as
 	// as consequence make this rather annoying to implement and use.
 }
 
 /// The regular set of subsystems.
-pub struct RealOverseerGen;
+pub struct ValidatorOverseerGen;
 
-impl OverseerGen for RealOverseerGen {
+impl OverseerGen for ValidatorOverseerGen {
 	fn generate<Spawner, RuntimeClient>(
 		&self,
 		connector: OverseerConnector,
 		args: OverseerGenArgs<Spawner, RuntimeClient>,
+		ext_args: Option<ExtendedOverseerGenArgs>,
 	) -> Result<
 		(Overseer<SpawnGlue<Spawner>, Arc<DefaultSubsystemClient<RuntimeClient>>>, OverseerHandle),
 		Error,
@@ -403,7 +542,35 @@ impl OverseerGen for RealOverseerGen {
 		RuntimeClient::Api: ParachainHost<Block> + BabeApi<Block> + AuthorityDiscoveryApi<Block>,
 		Spawner: 'static + SpawnNamed + Clone + Unpin,
 	{
-		prepared_overseer_builder(args)?
+		let ext_args = ext_args.ok_or(Error::Overseer(SubsystemError::Context(
+			"create validator overseer as mandatory extended arguments were not provided"
+				.to_owned(),
+		)))?;
+		validator_overseer_builder(args, ext_args)?
+			.build_with_connector(connector)
+			.map_err(|e| e.into())
+	}
+}
+
+/// Reduced set of subsystems, to use in collator and collator's full node.
+pub struct CollatorOverseerGen;
+
+impl OverseerGen for CollatorOverseerGen {
+	fn generate<Spawner, RuntimeClient>(
+		&self,
+		connector: OverseerConnector,
+		args: OverseerGenArgs<Spawner, RuntimeClient>,
+		_ext_args: Option<ExtendedOverseerGenArgs>,
+	) -> Result<
+		(Overseer<SpawnGlue<Spawner>, Arc<DefaultSubsystemClient<RuntimeClient>>>, OverseerHandle),
+		Error,
+	>
+	where
+		RuntimeClient: 'static + ProvideRuntimeApi<Block> + HeaderBackend<Block> + AuxStore,
+		RuntimeClient::Api: ParachainHost<Block> + BabeApi<Block> + AuthorityDiscoveryApi<Block>,
+		Spawner: 'static + SpawnNamed + Clone + Unpin,
+	{
+		collator_overseer_builder(args)?
 			.build_with_connector(connector)
 			.map_err(|e| e.into())
 	}

polkadot/node/test/service/src/lib.rs

@@ -28,7 +28,9 @@ use polkadot_overseer::Handle;
 use polkadot_primitives::{Balance, CollatorPair, HeadData, Id as ParaId, ValidationCode};
 use polkadot_runtime_common::BlockHashCount;
 use polkadot_runtime_parachains::paras::{ParaGenesisArgs, ParaKind};
-use polkadot_service::{Error, FullClient, IsParachainNode, NewFull, PrometheusConfig};
+use polkadot_service::{
+	Error, FullClient, IsParachainNode, NewFull, OverseerGen, PrometheusConfig,
+};
 use polkadot_test_runtime::{
 	ParasCall, ParasSudoWrapperCall, Runtime, SignedExtra, SignedPayload, SudoCall,
 	UncheckedExtrinsic, VERSION,
@@ -69,10 +71,11 @@ pub use polkadot_service::{FullBackend, GetLastTimestamp};
 
 /// Create a new full node.
 #[sc_tracing::logging::prefix_logs_with(config.network.node_name.as_str())]
-pub fn new_full(
+pub fn new_full<OverseerGenerator: OverseerGen>(
 	config: Configuration,
 	is_parachain_node: IsParachainNode,
 	workers_path: Option<PathBuf>,
+	overseer_gen: OverseerGenerator,
 ) -> Result<NewFull, Error> {
 	let workers_path = Some(workers_path.unwrap_or_else(get_relative_workers_path_for_test));
 
@@ -88,7 +91,7 @@ pub fn new_full(
 			secure_validator_mode: false,
 			workers_path,
 			workers_names: None,
-			overseer_gen: polkadot_service::RealOverseerGen,
+			overseer_gen,
 			overseer_message_channel_capacity_override: None,
 			malus_finality_delay: None,
 			hwbench: None,
@@ -207,8 +210,12 @@ pub fn run_validator_node(
 	worker_program_path: Option<PathBuf>,
 ) -> PolkadotTestNode {
 	let multiaddr = config.network.listen_addresses[0].clone();
-	let NewFull { task_manager, client, network, rpc_handlers, overseer_handle, .. } =
+	let NewFull { task_manager, client, network, rpc_handlers, overseer_handle, .. } = new_full(
-		new_full(config, IsParachainNode::No, worker_program_path)
+		config,
+		IsParachainNode::No,
+		worker_program_path,
+		polkadot_service::ValidatorOverseerGen,
+	)
 	.expect("could not create Polkadot test service");
 
 	let overseer_handle = overseer_handle.expect("test node must have an overseer handle");
@@ -239,8 +246,12 @@ pub fn run_collator_node(
 ) -> PolkadotTestNode {
 	let config = node_config(storage_update_func, tokio_handle, key, boot_nodes, false);
 	let multiaddr = config.network.listen_addresses[0].clone();
-	let NewFull { task_manager, client, network, rpc_handlers, overseer_handle, .. } =
+	let NewFull { task_manager, client, network, rpc_handlers, overseer_handle, .. } = new_full(
-		new_full(config, IsParachainNode::Collator(collator_pair), None)
+		config,
+		IsParachainNode::Collator(collator_pair),
+		None,
+		polkadot_service::CollatorOverseerGen,
+	)
 	.expect("could not create Polkadot test service");
 
 	let overseer_handle = overseer_handle.expect("test node must have an overseer handle");

polkadot/parachain/test-parachains/adder/collator/src/main.rs

@@ -91,7 +91,7 @@ fn main() -> Result<()> {
 						workers_path: None,
 						workers_names: None,
 
-						overseer_gen: polkadot_service::RealOverseerGen,
+						overseer_gen: polkadot_service::CollatorOverseerGen,
 						overseer_message_channel_capacity_override: None,
 						malus_finality_delay: None,
 						hwbench: None,

polkadot/parachain/test-parachains/undying/collator/src/main.rs

@@ -93,7 +93,7 @@ fn main() -> Result<()> {
 						workers_path: None,
 						workers_names: None,
 
-						overseer_gen: polkadot_service::RealOverseerGen,
+						overseer_gen: polkadot_service::CollatorOverseerGen,
 						overseer_message_channel_capacity_override: None,
 						malus_finality_delay: None,
 						hwbench: None,

prdoc/pr_3061.prdoc

@@ -0,0 +1,15 @@
+# Schema: Polkadot SDK PRDoc Schema (prdoc) v1.0.0
+# See doc at https://raw.githubusercontent.com/paritytech/polkadot-sdk/master/prdoc/schema_user.json
+
+title: Do not run unneeded subsystems on collator and its alongside node
+
+doc:
+  - audience: Node Dev
+    description: |
+      Optimizes overseer building strategy to only include subsystems needed to run the given
+      type of node. Reduces overseer overhead and also solves the problem with unused subsystems
+      getting stalled from time to time.
+
+crates:
+  - name: polkadot-overseer
+  - name: polkadot-service