Add more metrics to prometheus (#5034)

* Add a few things

* Add finality_grandpa_round

* fix fg tests

* Nitpicks

* Nitpicks

* Fix name of prometheus crate

substrate/client/service/src/builder.rs

@@ -53,15 +53,15 @@ use sysinfo::{get_current_pid, ProcessExt, System, SystemExt};
 use sc_telemetry::{telemetry, SUBSTRATE_INFO};
 use sp_transaction_pool::{MaintainedTransactionPool, ChainEvent};
 use sp_blockchain;
-use substrate_prometheus_endpoint::{register, Gauge, U64, F64, Registry, PrometheusError, Opts, GaugeVec};
+use prometheus_endpoint::{register, Gauge, U64, F64, Registry, PrometheusError, Opts, GaugeVec};
 
 struct ServiceMetrics {
 	block_height_number: GaugeVec<U64>,
-	peers_count: Gauge<U64>,
 	ready_transactions_number: Gauge<U64>,
 	memory_usage_bytes: Gauge<U64>,
 	cpu_usage_percentage: Gauge<F64>,
 	network_per_sec_bytes: GaugeVec<U64>,
+	node_roles: Gauge<U64>,
 }
 
 impl ServiceMetrics {
@@ -71,9 +71,6 @@ impl ServiceMetrics {
 				Opts::new("block_height_number", "Height of the chain"),
 				&["status"]
 			)?, registry)?,
-			peers_count: register(Gauge::new(
-				"peers_count", "Number of network gossip peers",
-			)?, registry)?,
 			ready_transactions_number: register(Gauge::new(
 				"ready_transactions_number", "Number of transactions in the ready queue",
 			)?, registry)?,
@@ -87,6 +84,10 @@ impl ServiceMetrics {
 				Opts::new("network_per_sec_bytes", "Networking bytes per second"),
 				&["direction"]
 			)?, registry)?,
+			node_roles: register(Gauge::new(
+				"node_roles", "The roles the node is running as",
+			)?, registry)?,
+
 		})
 	}
 }
@@ -887,6 +888,14 @@ ServiceBuilder<
 		let block_announce_validator =
 			Box::new(sp_consensus::block_validation::DefaultBlockAnnounceValidator::new(client.clone()));
 
+		let prometheus_registry_and_port = match config.prometheus_port {
+			Some(port) => match prometheus_registry {
+				Some(registry) => Some((registry, port)),
+				None => Some((Registry::new_custom(Some("substrate".into()), None)?, port))
+			},
+			None => None
+		};
+
 		let network_params = sc_network::config::Params {
 			roles: config.roles,
 			executor: {
@@ -906,6 +915,7 @@ ServiceBuilder<
 			import_queue,
 			protocol_id,
 			block_announce_validator,
+			metrics_registry: prometheus_registry_and_port.as_ref().map(|(r, _)| r.clone())
 		};
 
 		let has_bootnodes = !network_params.network_config.boot_nodes.is_empty();
@@ -1020,17 +1030,14 @@ ServiceBuilder<
 			));
 		}
 
-		// Prometheus endpoint and metrics
-		let metrics = if let Some(port) = config.prometheus_port {
-			let registry = match prometheus_registry {
-				Some(registry) => registry,
-				None => Registry::new_custom(Some("substrate".into()), None)?
-			};
-
+		// Prometheus metrics
+		let metrics = if let Some((registry, port)) = prometheus_registry_and_port.clone() {
 			let metrics = ServiceMetrics::register(&registry)?;
 
+			metrics.node_roles.set(u64::from(config.roles.bits()));
+
 			let future = select(
-				substrate_prometheus_endpoint::init_prometheus(port, registry).boxed(),
+				prometheus_endpoint::init_prometheus(port, registry).boxed(),
 				exit.clone()
 			).map(drop);
 
@@ -1043,7 +1050,6 @@ ServiceBuilder<
 		} else {
 			None
 		};
-
 		// Periodically notify the telemetry.
 		let transaction_pool_ = transaction_pool.clone();
 		let client_ = client.clone();
@@ -1102,7 +1108,6 @@ ServiceBuilder<
 				metrics.memory_usage_bytes.set(memory);
 				metrics.cpu_usage_percentage.set(f64::from(cpu_usage));
 				metrics.ready_transactions_number.set(txpool_status.ready as u64);
-				metrics.peers_count.set(num_peers as u64);
 
 				metrics.network_per_sec_bytes.with_label_values(&["download"]).set(net_status.average_download_per_sec);
 				metrics.network_per_sec_bytes.with_label_values(&["upload"]).set(net_status.average_upload_per_sec);
@@ -1305,6 +1310,7 @@ ServiceBuilder<
 			_telemetry_on_connect_sinks: telemetry_connection_sinks.clone(),
 			keystore,
 			marker: PhantomData::<TBl>,
+			prometheus_registry: prometheus_registry_and_port.map(|(r, _)| r)
 		})
 	}
 }

substrate/client/service/src/error.rs

@@ -53,8 +53,8 @@ impl<'a> From<&'a str> for Error {
 	}
 }
 
-impl From<substrate_prometheus_endpoint::PrometheusError> for Error {
-	fn from(e: substrate_prometheus_endpoint::PrometheusError) -> Self {
+impl From<prometheus_endpoint::PrometheusError> for Error {
+	fn from(e: prometheus_endpoint::PrometheusError) -> Self {
 		Error::Other(format!("Prometheus error: {}", e))
 	}
 }

substrate/client/service/src/lib.rs

@@ -114,6 +114,7 @@ pub struct Service<TBl, TCl, TSc, TNetStatus, TNet, TTxPool, TOc> {
 	_offchain_workers: Option<Arc<TOc>>,
 	keystore: sc_keystore::KeyStorePtr,
 	marker: PhantomData<TBl>,
+	prometheus_registry: Option<prometheus_endpoint::Registry>,
 }
 
 impl<TBl, TCl, TSc, TNetStatus, TNet, TTxPool, TOc> Unpin for Service<TBl, TCl, TSc, TNetStatus, TNet, TTxPool, TOc> {}
@@ -225,6 +226,9 @@ pub trait AbstractService: 'static + Future<Output = Result<(), Error>> +
 
 	/// Get a handle to a future that will resolve on exit.
 	fn on_exit(&self) -> ::exit_future::Exit;
+
+	/// Get the prometheus metrics registry, if available.
+	fn prometheus_registry(&self) -> Option<prometheus_endpoint::Registry>;
 }
 
 impl<TBl, TBackend, TExec, TRtApi, TSc, TExPool, TOc> AbstractService for
@@ -328,6 +332,10 @@ where
 	fn on_exit(&self) -> exit_future::Exit {
 		self.exit.clone()
 	}
+
+	fn prometheus_registry(&self) -> Option<prometheus_endpoint::Registry> {
+		self.prometheus_registry.clone()
+	}
 }
 
 impl<TBl, TCl, TSc, TNetStatus, TNet, TTxPool, TOc> Future for