Rework telemetry to replace the use of tracing with an object we pass around (#8143)

polkadot companion: paritytech/polkadot#2535

substrate/bin/node/bench/src/construct.rs

@@ -151,6 +151,7 @@ impl core::Benchmark for ConstructionBenchmark {
 			context.client.clone(),
 			self.transactions.clone().into(),
 			None,
+			None,
 		);
 		let inherent_data_providers = sp_inherents::InherentDataProviders::new();
 		inherent_data_providers

substrate/bin/node/cli/Cargo.toml

@@ -105,6 +105,7 @@ try-runtime-cli = { version = "0.9.0", optional = true, path = "../../../utils/f
 wasm-bindgen = { version = "0.2.57", optional = true }
 wasm-bindgen-futures = { version = "0.4.18", optional = true }
 browser-utils = { package = "substrate-browser-utils", path = "../../../utils/browser", optional = true, version = "0.9.0"}
+libp2p-wasm-ext = { version = "0.27", features = ["websocket"], optional = true }
 
 [target.'cfg(target_arch="x86_64")'.dependencies]
 node-executor = { version = "2.0.0", path = "../executor", features = [ "wasmtime" ] }
@@ -148,6 +149,7 @@ browser = [
 	"browser-utils",
 	"wasm-bindgen",
 	"wasm-bindgen-futures",
+	"libp2p-wasm-ext",
 ]
 cli = [
 	"node-executor/wasmi-errno",

substrate/bin/node/cli/src/browser.rs

@@ -21,7 +21,7 @@ use log::info;
 use wasm_bindgen::prelude::*;
 use browser_utils::{
 	Client,
-	browser_configuration, init_logging_and_telemetry, set_console_error_panic_hook,
+	browser_configuration, init_logging, set_console_error_panic_hook,
 };
 
 /// Starts the client.
@@ -37,18 +37,14 @@ async fn start_inner(
 	log_directives: String,
 ) -> Result<Client, Box<dyn std::error::Error>> {
 	set_console_error_panic_hook();
-	let telemetry_worker = init_logging_and_telemetry(&log_directives)?;
+	init_logging(&log_directives)?;
 	let chain_spec = match chain_spec {
 		Some(chain_spec) => ChainSpec::from_json_bytes(chain_spec.as_bytes().to_vec())
 			.map_err(|e| format!("{:?}", e))?,
 		None => crate::chain_spec::development_config(),
 	};
 
-	let telemetry_handle = telemetry_worker.handle();
-	let config = browser_configuration(
-		chain_spec,
-		Some(telemetry_handle),
-	).await?;
+	let config = browser_configuration(chain_spec).await?;
 
 	info!("Substrate browser node");
 	info!("✌️  version {}", config.impl_version);
@@ -60,10 +56,8 @@ async fn start_inner(
 	// Create the service. This is the most heavy initialization step.
 	let (task_manager, rpc_handlers) =
 		crate::service::new_light_base(config)
-			.map(|(components, rpc_handlers, _, _, _, _)| (components, rpc_handlers))
+			.map(|(components, rpc_handlers, _, _, _)| (components, rpc_handlers))
 			.map_err(|e| format!("{:?}", e))?;
 
-	task_manager.spawn_handle().spawn("telemetry", telemetry_worker.run());
-
 	Ok(browser_utils::start_client(task_manager, rpc_handlers))
 }

substrate/bin/node/cli/src/chain_spec.rs

@@ -443,7 +443,7 @@ pub(crate) mod tests {
 				Ok(sc_service_test::TestNetComponents::new(task_manager, client, network, transaction_pool))
 			},
 			|config| {
-				let (keep_alive, _, _, client, network, transaction_pool) = new_light_base(config)?;
+				let (keep_alive, _, client, network, transaction_pool) = new_light_base(config)?;
 				Ok(sc_service_test::TestNetComponents::new(keep_alive, client, network, transaction_pool))
 			}
 		);

substrate/bin/node/cli/src/service.rs

@@ -33,7 +33,7 @@ use sp_runtime::traits::Block as BlockT;
 use futures::prelude::*;
 use sc_client_api::{ExecutorProvider, RemoteBackend};
 use node_executor::Executor;
-use sc_telemetry::{TelemetryConnectionNotifier, TelemetrySpan};
+use sc_telemetry::{Telemetry, TelemetryWorker};
 use sc_consensus_babe::SlotProportion;
 
 type FullClient = sc_service::TFullClient<Block, RuntimeApi, Executor>;
@@ -43,7 +43,9 @@ type FullGrandpaBlockImport =
 	grandpa::GrandpaBlockImport<FullBackend, Block, FullClient, FullSelectChain>;
 type LightClient = sc_service::TLightClient<Block, RuntimeApi, Executor>;
 
-pub fn new_partial(config: &Configuration) -> Result<sc_service::PartialComponents<
+pub fn new_partial(
+	config: &Configuration,
+) -> Result<sc_service::PartialComponents<
 	FullClient, FullBackend, FullSelectChain,
 	sp_consensus::DefaultImportQueue<Block, FullClient>,
 	sc_transaction_pool::FullPool<Block, FullClient>,
@@ -58,12 +60,31 @@ pub fn new_partial(config: &Configuration) -> Result<sc_service::PartialComponen
 			sc_consensus_babe::BabeLink<Block>,
 		),
 		grandpa::SharedVoterState,
+		Option<Telemetry>,
 	)
 >, ServiceError> {
+	let telemetry = config.telemetry_endpoints.clone()
+		.filter(|x| !x.is_empty())
+		.map(|endpoints| -> Result<_, sc_telemetry::Error> {
+			let worker = TelemetryWorker::new(16)?;
+			let telemetry = worker.handle().new_telemetry(endpoints);
+			Ok((worker, telemetry))
+		})
+		.transpose()?;
+
 	let (client, backend, keystore_container, task_manager) =
-		sc_service::new_full_parts::<Block, RuntimeApi, Executor>(&config)?;
+		sc_service::new_full_parts::<Block, RuntimeApi, Executor>(
+			&config,
+			telemetry.as_ref().map(|(_, telemetry)| telemetry.handle()),
+		)?;
 	let client = Arc::new(client);
 
+	let telemetry = telemetry
+		.map(|(worker, telemetry)| {
+			task_manager.spawn_handle().spawn("telemetry", worker.run());
+			telemetry
+		});
+
 	let select_chain = sc_consensus::LongestChain::new(backend.clone());
 
 	let transaction_pool = sc_transaction_pool::BasicPool::new_full(
@@ -75,7 +96,10 @@ pub fn new_partial(config: &Configuration) -> Result<sc_service::PartialComponen
 	);
 
 	let (grandpa_block_import, grandpa_link) = grandpa::block_import(
-		client.clone(), &(client.clone() as Arc<_>), select_chain.clone(),
+		client.clone(),
+		&(client.clone() as Arc<_>),
+		select_chain.clone(),
+		telemetry.as_ref().map(|x| x.handle()),
 	)?;
 	let justification_import = grandpa_block_import.clone();
 
@@ -97,6 +121,7 @@ pub fn new_partial(config: &Configuration) -> Result<sc_service::PartialComponen
 		&task_manager.spawn_essential_handle(),
 		config.prometheus_registry(),
 		sp_consensus::CanAuthorWithNativeVersion::new(client.executor().clone()),
+		telemetry.as_ref().map(|x| x.handle()),
 	)?;
 
 	let import_setup = (block_import, grandpa_link, babe_link);
@@ -159,7 +184,7 @@ pub fn new_partial(config: &Configuration) -> Result<sc_service::PartialComponen
 		import_queue,
 		transaction_pool,
 		inherent_data_providers,
-		other: (rpc_extensions_builder, import_setup, rpc_setup),
+		other: (rpc_extensions_builder, import_setup, rpc_setup, telemetry),
 	})
 }
 
@@ -189,7 +214,7 @@ pub fn new_full_base(
 		select_chain,
 		transaction_pool,
 		inherent_data_providers,
-		other: (rpc_extensions_builder, import_setup, rpc_setup),
+		other: (rpc_extensions_builder, import_setup, rpc_setup, mut telemetry),
 	} = new_partial(&config)?;
 
 	let shared_voter_state = rpc_setup;
@@ -231,10 +256,7 @@ pub fn new_full_base(
 	let enable_grandpa = !config.disable_grandpa;
 	let prometheus_registry = config.prometheus_registry().cloned();
 
-	let telemetry_span = TelemetrySpan::new();
-	let _telemetry_span_entered = telemetry_span.enter();
-
-	let (_rpc_handlers, telemetry_connection_notifier) = sc_service::spawn_tasks(
+	let _rpc_handlers = sc_service::spawn_tasks(
 		sc_service::SpawnTasksParams {
 			config,
 			backend: backend.clone(),
@@ -248,7 +270,7 @@ pub fn new_full_base(
 			remote_blockchain: None,
 			network_status_sinks: network_status_sinks.clone(),
 			system_rpc_tx,
-			telemetry_span: Some(telemetry_span.clone()),
+			telemetry: telemetry.as_mut(),
 		},
 	)?;
 
@@ -262,6 +284,7 @@ pub fn new_full_base(
 			client.clone(),
 			transaction_pool.clone(),
 			prometheus_registry.as_ref(),
+			telemetry.as_ref().map(|x| x.handle()),
 		);
 
 		let can_author_with =
@@ -280,6 +303,7 @@ pub fn new_full_base(
 			babe_link,
 			can_author_with,
 			block_proposal_slot_portion: SlotProportion::new(0.5),
+			telemetry: telemetry.as_ref().map(|x| x.handle()),
 		};
 
 		let babe = sc_consensus_babe::start_babe(babe_config)?;
@@ -323,6 +347,7 @@ pub fn new_full_base(
 		observer_enabled: false,
 		keystore,
 		is_authority: role.is_authority(),
+		telemetry: telemetry.as_ref().map(|x| x.handle()),
 	};
 
 	if enable_grandpa {
@@ -336,7 +361,7 @@ pub fn new_full_base(
 			config,
 			link: grandpa_link,
 			network: network.clone(),
-			telemetry_on_connect: telemetry_connection_notifier.map(|x| x.on_connect_stream()),
+			telemetry: telemetry.as_ref().map(|x| x.handle()),
 			voting_rule: grandpa::VotingRulesBuilder::default().build(),
 			prometheus_registry,
 			shared_voter_state,
@@ -362,20 +387,50 @@ pub fn new_full_base(
 }
 
 /// Builds a new service for a full client.
-pub fn new_full(config: Configuration)
--> Result<TaskManager, ServiceError> {
+pub fn new_full(
+	config: Configuration,
+) -> Result<TaskManager, ServiceError> {
 	new_full_base(config, |_, _| ()).map(|NewFullBase { task_manager, .. }| {
 		task_manager
 	})
 }
 
-pub fn new_light_base(mut config: Configuration) -> Result<(
-	TaskManager, RpcHandlers, Option<TelemetryConnectionNotifier>, Arc<LightClient>,
+pub fn new_light_base(
+	mut config: Configuration,
+) -> Result<(
+	TaskManager,
+	RpcHandlers,
+	Arc<LightClient>,
 	Arc<NetworkService<Block, <Block as BlockT>::Hash>>,
 	Arc<sc_transaction_pool::LightPool<Block, LightClient, sc_network::config::OnDemand<Block>>>
 ), ServiceError> {
+	let telemetry = config.telemetry_endpoints.clone()
+		.filter(|x| !x.is_empty())
+		.map(|endpoints| -> Result<_, sc_telemetry::Error> {
+			#[cfg(feature = "browser")]
+			let transport = Some(
+				sc_telemetry::ExtTransport::new(libp2p_wasm_ext::ffi::websocket_transport())
+			);
+			#[cfg(not(feature = "browser"))]
+			let transport = None;
+
+			let worker = TelemetryWorker::with_transport(16, transport)?;
+			let telemetry = worker.handle().new_telemetry(endpoints);
+			Ok((worker, telemetry))
+		})
+		.transpose()?;
+
 	let (client, backend, keystore_container, mut task_manager, on_demand) =
-		sc_service::new_light_parts::<Block, RuntimeApi, Executor>(&config)?;
+		sc_service::new_light_parts::<Block, RuntimeApi, Executor>(
+			&config,
+			telemetry.as_ref().map(|(_, telemetry)| telemetry.handle()),
+		)?;
+
+	let mut telemetry = telemetry
+		.map(|(worker, telemetry)| {
+			task_manager.spawn_handle().spawn("telemetry", worker.run());
+			telemetry
+		});
 
 	config.network.extra_sets.push(grandpa::grandpa_peers_set_config());
 
@@ -393,6 +448,7 @@ pub fn new_light_base(mut config: Configuration) -> Result<(
 		client.clone(),
 		&(client.clone() as Arc<_>),
 		select_chain.clone(),
+		telemetry.as_ref().map(|x| x.handle()),
 	)?;
 	let justification_import = grandpa_block_import.clone();
 
@@ -414,6 +470,7 @@ pub fn new_light_base(mut config: Configuration) -> Result<(
 		&task_manager.spawn_essential_handle(),
 		config.prometheus_registry(),
 		sp_consensus::NeverCanAuthor,
+		telemetry.as_ref().map(|x| x.handle()),
 	)?;
 
 	let (network, network_status_sinks, system_rpc_tx, network_starter) =
@@ -443,10 +500,7 @@ pub fn new_light_base(mut config: Configuration) -> Result<(
 
 	let rpc_extensions = node_rpc::create_light(light_deps);
 
-	let telemetry_span = TelemetrySpan::new();
-	let _telemetry_span_entered = telemetry_span.enter();
-
-	let (rpc_handlers, telemetry_connection_notifier) =
+	let rpc_handlers =
 		sc_service::spawn_tasks(sc_service::SpawnTasksParams {
 			on_demand: Some(on_demand),
 			remote_blockchain: Some(backend.remote_blockchain()),
@@ -457,13 +511,12 @@ pub fn new_light_base(mut config: Configuration) -> Result<(
 			config, backend, network_status_sinks, system_rpc_tx,
 			network: network.clone(),
 			task_manager: &mut task_manager,
-			telemetry_span: Some(telemetry_span.clone()),
+			telemetry: telemetry.as_mut(),
 		})?;
 
 	Ok((
 		task_manager,
 		rpc_handlers,
-		telemetry_connection_notifier,
 		client,
 		network,
 		transaction_pool,
@@ -471,8 +524,10 @@ pub fn new_light_base(mut config: Configuration) -> Result<(
 }
 
 /// Builds a new service for a light client.
-pub fn new_light(config: Configuration) -> Result<TaskManager, ServiceError> {
-	new_light_base(config).map(|(task_manager, _, _, _, _, _)| {
+pub fn new_light(
+	config: Configuration,
+) -> Result<TaskManager, ServiceError> {
+	new_light_base(config).map(|(task_manager, _, _, _, _)| {
 		task_manager
 	})
 }
@@ -553,7 +608,7 @@ mod tests {
 				Ok((node, (inherent_data_providers, setup_handles.unwrap())))
 			},
 			|config| {
-				let (keep_alive, _, _, client, network, transaction_pool) = new_light_base(config)?;
+				let (keep_alive, _, client, network, transaction_pool) = new_light_base(config)?;
 				Ok(sc_service_test::TestNetComponents::new(keep_alive, client, network, transaction_pool))
 			},
 			|service, &mut (ref inherent_data_providers, (ref mut block_import, ref babe_link))| {
@@ -580,6 +635,7 @@ mod tests {
 					service.client(),
 					service.transaction_pool(),
 					None,
+					None,
 				);
 
 				let epoch_descriptor = babe_link.epoch_changes().lock().epoch_descriptor_for_child_of(
@@ -708,7 +764,7 @@ mod tests {
 				Ok(sc_service_test::TestNetComponents::new(task_manager, client, network, transaction_pool))
 			},
 			|config| {
-				let (keep_alive, _, _, client, network, transaction_pool) = new_light_base(config)?;
+				let (keep_alive, _, client, network, transaction_pool) = new_light_base(config)?;
 				Ok(sc_service_test::TestNetComponents::new(keep_alive, client, network, transaction_pool))
 			},
 			vec![

substrate/bin/node/inspect/src/command.rs

@@ -34,7 +34,7 @@ impl InspectCmd {
 		RA: Send + Sync + 'static,
 		EX: NativeExecutionDispatch + 'static,
 	{
-		let client = new_full_client::<B, RA, EX>(&config)?;
+		let client = new_full_client::<B, RA, EX>(&config, None)?;
 		let inspect = Inspector::<B>::new(client);
 
 		match &self.command {

substrate/bin/node/testing/src/bench.rs

@@ -427,6 +427,7 @@ impl BenchDb {
 			ExecutionExtensions::new(profile.into_execution_strategies(), None, None),
 			Box::new(task_executor.clone()),
 			None,
+			None,
 			Default::default(),
 		).expect("Should not fail");