fix(guard): add metrics

packages/common/logs/src/lib.rs

@@ -25,10 +25,12 @@ impl Logs {
 }
 
 impl Logs {
-	pub async fn start(self) -> Result<tokio::task::JoinHandle<()>> {
+	pub async fn start(mut self) -> Result<tokio::task::JoinHandle<()>> {
 		// Create logs dir if it does not exist
 		fs::create_dir_all(&self.path).await?;
 
+		self.rotate().await?;
+
 		Ok(tokio::spawn(self.run()))
 	}
 
@@ -112,10 +114,12 @@ impl Logs {
 }
 
 impl Logs {
-	pub fn start_sync(self) -> Result<std::thread::JoinHandle<()>> {
+	pub fn start_sync(mut self) -> Result<std::thread::JoinHandle<()>> {
 		// Create logs dir if it does not exist
 		std::fs::create_dir_all(&self.path)?;
 
+		self.rotate_sync()?;
+
 		Ok(std::thread::spawn(|| self.run_sync()))
 	}
 

packages/common/server-cli/src/commands/start.rs

@@ -54,7 +54,6 @@ impl Opts {
 			.and_then(|x| x.rivet.edge.as_ref())
 			.and_then(|x| x.redirect_logs_dir.as_ref())
 		{
-			std::fs::create_dir_all(logs_dir)?;
 			rivet_logs::Logs::new(logs_dir.clone(), LOGS_RETENTION)
 				.start()
 				.await?;

packages/core/services/cluster/src/workflows/server/install/install_scripts/mod.rs

@@ -229,6 +229,14 @@ pub async fn gen_initialize(
 			)?);
 			script.push(components::rivet::guard::fetch_tls(server_token)?);
 			script.push(components::rivet::guard::configure(config)?);
+
+			prometheus_targets.insert(
+				"guard".into(),
+				components::vector::PrometheusTarget {
+					endpoint: "http://127.0.0.1:8091".into(),
+					scrape_interval: 15,
+				},
+			);
 		}
 	}
 

packages/edge/infra/guard/core/src/metrics.rs

@@ -1,29 +1,80 @@
 use lazy_static::lazy_static;
-use rivet_metrics::prometheus::*;
+use rivet_metrics::{prometheus::*, REGISTRY};
 
 lazy_static! {
-	pub static ref ACTOR_REQUEST_TOTAL: IntCounterVec = register_int_counter_vec!(
-		"actor_request_total",
+	// MARK: Internal
+	pub static ref ROUTE_CACHE_SIZE: IntGauge = register_int_gauge_with_registry!(
+		"guard_route_cache_size",
+		"Number of entries in the route cache",
+		*REGISTRY,
+	).unwrap();
+	pub static ref RATE_LIMITER_COUNT: IntGauge = register_int_gauge_with_registry!(
+		"guard_rate_limiter_count",
+		"Number of active rate limiters",
+		*REGISTRY,
+	).unwrap();
+	pub static ref IN_FLIGHT_COUNTER_COUNT: IntGauge = register_int_gauge_with_registry!(
+		"guard_in_flight_counter_count",
+		"Number of active in-flight counters",
+		*REGISTRY,
+	)
+	.unwrap();
+
+	// MARK: TCP
+	pub static ref TCP_CONNECTION_TOTAL: IntCounter = register_int_counter_with_registry!(
+		"guard_tcp_connection_total",
+		"Total number of TCP connections ever",
+		*REGISTRY,
+	)
+	.unwrap();
+	pub static ref TCP_CONNECTION_PENDING: IntGauge = register_int_gauge_with_registry!(
+		"guard_tcp_connection_pending",
+		"Total number of open TCP connections",
+		*REGISTRY,
+	)
+	.unwrap();
+	pub static ref TCP_CONNECTION_DURATION: Histogram = register_histogram_with_registry!(
+		"guard_tcp_connection_duration",
+		"TCP connection duration in seconds",
+		*REGISTRY,
+	)
+	.unwrap();
+
+	// MARK: Pre-proxy
+	pub static ref RESOLVE_ROUTE_DURATION: Histogram = register_histogram_with_registry!(
+		"guard_resolve_route_duration",
+		"Time to resolve request route in seconds",
+		*REGISTRY,
+	)
+	.unwrap();
+
+	// MARK: Proxy requests
+	pub static ref PROXY_REQUEST_TOTAL: IntCounterVec = register_int_counter_vec_with_registry!(
+		"guard_proxy_request_total",
 		"Total number of requests to actor",
-		&["actor_id", "server_id", "method", "path"]
+		&["actor_id", "server_id", "method", "path"],
+		*REGISTRY,
 	)
 	.unwrap();
-	pub static ref ACTOR_REQUEST_PENDING: IntGaugeVec = register_int_gauge_vec!(
-		"actor_request_pending",
+	pub static ref PROXY_REQUEST_PENDING: IntGaugeVec = register_int_gauge_vec_with_registry!(
+		"guard_proxy_request_pending",
 		"Number of pending requests to actor",
-		&["actor_id", "server_id", "method", "path"]
+		&["actor_id", "server_id", "method", "path"],
+		*REGISTRY,
 	)
 	.unwrap();
-	pub static ref ACTOR_REQUEST_DURATION: HistogramVec = register_histogram_vec!(
-		"actor_request_duration_seconds",
+	pub static ref PROXY_REQUEST_DURATION: HistogramVec = register_histogram_vec_with_registry!(
+		"guard_proxy_request_duration",
 		"Request duration in seconds",
-		&["actor_id", "server_id", "status"]
+		&["actor_id", "server_id", "status"],
+		*REGISTRY,
 	)
 	.unwrap();
-	pub static ref ACTOR_REQUEST_ERRORS: IntCounterVec = register_int_counter_vec!(
-		"actor_request_errors_total",
+	pub static ref PROXY_REQUEST_ERROR: IntCounterVec = register_int_counter_vec_with_registry!(
+		"guard_proxy_request_errors_total",
 		"Total number of errors when proxying requests to actor",
-		&["actor_id", "server_id", "error_type"]
+		&["actor_id", "server_id", "error_type"],
+		*REGISTRY,
 	)
 	.unwrap();
 }