chore: convert health checks & metrics to services

packages/common/health-checks/src/lib.rs

@@ -20,7 +20,7 @@ pub enum HealthCheckError {
 }
 
 #[tracing::instrument(skip_all)]
-pub fn spawn_standalone(config: Config) -> GlobalResult<()> {
+pub async fn run_standalone(config: Config) -> GlobalResult<()> {
 	let config = Arc::new(config);
 	let host = config.config.server()?.rivet.health.host();
 	let port = config.config.server()?.rivet.health.port();
@@ -39,12 +39,8 @@ pub fn spawn_standalone(config: Config) -> GlobalResult<()> {
 
 	let server = Server::try_bind(&addr)?.serve(make_service);
 
-	tokio::spawn(async move {
-		tracing::info!(?host, ?port, "started health server");
-		if let Err(err) = server.await {
-			tracing::error!(?err, "health server error");
-		}
-	});
+	tracing::info!(?host, ?port, "started health server");
+	server.await?;
 
 	Ok(())
 }

packages/common/metrics/src/lib.rs

@@ -5,4 +5,4 @@ mod server;
 pub use buckets::{BUCKETS, PROVISION_BUCKETS};
 pub use prometheus;
 pub use registry::REGISTRY;
-pub use server::spawn_standalone;
+pub use server::run_standalone;

packages/common/metrics/src/server.rs

@@ -10,7 +10,7 @@ use std::net::SocketAddr;
 // TODO: Record extra labels
 
 #[tracing::instrument(skip_all)]
-pub fn spawn_standalone(config: rivet_config::Config) -> GlobalResult<()> {
+pub async fn run_standalone(config: rivet_config::Config) -> GlobalResult<()> {
 	let host = config.server()?.rivet.metrics.host();
 	let port = config.server()?.rivet.metrics.port();
 	let addr = SocketAddr::from((host, port));
@@ -19,12 +19,8 @@ pub fn spawn_standalone(config: rivet_config::Config) -> GlobalResult<()> {
 		Ok::<_, hyper::Error>(service_fn(serve_req))
 	}));
 
-	tokio::spawn(async move {
-		tracing::info!(?host, ?port, "started metrics server");
-		if let Err(err) = server.await {
-			tracing::error!(?err, "metrics server error");
-		}
-	});
+	tracing::info!(?host, ?port, "started metrics server");
+	server.await?;
 
 	Ok(())
 }

packages/common/server/src/lib.rs

@@ -39,14 +39,16 @@ pub enum ServiceKind {
 	Singleton,
 	Oneshot,
 	Cron,
+	/// Run no matter what.
+	Core,
 }
 
 impl ServiceKind {
 	fn behavior(&self) -> ServiceBehavior {
 		use ServiceKind::*;
 
 		match self {
-			Api | ApiInternal | Standalone | Singleton => ServiceBehavior::Service,
+			Api | ApiInternal | Standalone | Singleton | Core => ServiceBehavior::Service,
 			Oneshot | Cron => ServiceBehavior::Oneshot,
 		}
 	}
@@ -72,11 +74,27 @@ enum ServiceBehavior {
 pub async fn start(
 	config: rivet_config::Config,
 	pools: rivet_pools::Pools,
-	services: Vec<Service>,
+	mut services: Vec<Service>,
 ) -> Result<()> {
-	tracing::info!(services = ?services.len(), "starting server");
+	// Inject metrics and health services
+	services.push(Service::new(
+		"health_checks",
+		ServiceKind::Core,
+		|config, pools| {
+			rivet_health_checks::run_standalone(rivet_health_checks::Config {
+				config,
+				pools: Some(pools),
+			})
+		},
+	));
+	services.push(Service::new(
+		"metrics",
+		ServiceKind::Core,
+		|config, _pools| rivet_metrics::run_standalone(config),
+	));
 
 	// Spawn services
+	tracing::info!(services = ?services.len(), "starting server");
 	let mut join_set = tokio::task::JoinSet::new();
 	for service in services {
 		tracing::info!(name = %service.name, kind = ?service.kind, "server starting service");
@@ -138,15 +156,6 @@ pub async fn start(
 		}
 	}
 
-	// Run health & metrics servers
-	rivet_health_checks::spawn_standalone(rivet_health_checks::Config {
-		config: config.clone(),
-		pools: Some(pools.clone()),
-	})
-	.map_err(|err| anyhow!("failed to spawn health checks: {err}"))?;
-	rivet_metrics::spawn_standalone(config.clone())
-		.map_err(|err| anyhow!("failed to spawn metrics: {err}"))?;
-
 	// Wait for services
 	join_set.join_all().await;
 	tracing::info!("all services finished");