fix(guard): add more tokio runtime metrics, remove labels from metrics

Author: MasterPtato

packages/common/runtime/src/lib.rs

@@ -67,12 +67,17 @@ fn build_tokio_runtime_builder() -> tokio::runtime::Builder {
 		metrics::TOKIO_THREAD_COUNT.dec();
 	});
 
+	rt_builder.on_task_spawn(move |_| {
+		metrics::TOKIO_TASK_TOTAL.inc();
+	});
+
 	if env::var("TOKIO_RUNTIME_METRICS").is_ok() {
 		rt_builder.on_before_task_poll(|_| {
 			let metrics = tokio::runtime::Handle::current().metrics();
 			let buckets = metrics.poll_time_histogram_num_buckets();
 
 			metrics::TOKIO_GLOBAL_QUEUE_DEPTH.set(metrics.global_queue_depth() as i64);
+			metrics::TOKIO_ACTIVE_TASK_COUNT.set(metrics.num_alive_tasks() as i64);
 
 			for worker in 0..metrics.num_workers() {
 				metrics::TOKIO_WORKER_OVERFLOW_COUNT

packages/common/runtime/src/metrics.rs

@@ -14,6 +14,18 @@ lazy_static::lazy_static! {
 			"Number of pending tasks in the global queue.",
 			*REGISTRY
 		).unwrap();
+	pub static ref TOKIO_TASK_TOTAL: IntCounter =
+		register_int_counter_with_registry!(
+			"tokio_task_total",
+			"Total number of spawned tasks.",
+			*REGISTRY
+		).unwrap();
+	pub static ref TOKIO_ACTIVE_TASK_COUNT: IntGauge =
+		register_int_gauge_with_registry!(
+			"tokio_active_task_count",
+			"Total number of active (running or sleeping) tasks.",
+			*REGISTRY
+		).unwrap();
 	pub static ref TOKIO_WORKER_OVERFLOW_COUNT: IntGaugeVec = register_int_gauge_vec_with_registry!(
 			"tokio_worker_overflow_count",
 			"Number of times the given worker thread saturated its local queue.",

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

@@ -49,31 +49,29 @@ lazy_static! {
 	.unwrap();
 
 	// MARK: Proxy requests
-	pub static ref PROXY_REQUEST_TOTAL: IntCounterVec = register_int_counter_vec_with_registry!(
+	pub static ref PROXY_REQUEST_TOTAL: IntCounter = register_int_counter_with_registry!(
 		"guard_proxy_request_total",
 		"Total number of requests to actor",
-		&["actor_id", "server_id", "method", "path"],
 		*REGISTRY,
 	)
 	.unwrap();
-	pub static ref PROXY_REQUEST_PENDING: IntGaugeVec = register_int_gauge_vec_with_registry!(
+	pub static ref PROXY_REQUEST_PENDING: IntGauge = register_int_gauge_with_registry!(
 		"guard_proxy_request_pending",
 		"Number of pending requests to actor",
-		&["actor_id", "server_id", "method", "path"],
 		*REGISTRY,
 	)
 	.unwrap();
 	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"],
+		&["status"],
 		*REGISTRY,
 	)
 	.unwrap();
 	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"],
+		&["error_type"],
 		*REGISTRY,
 	)
 	.unwrap();

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

@@ -613,8 +613,6 @@ impl ProxyService {
 			.path_and_query()
 			.map(|x| x.to_string())
 			.unwrap_or_else(|| req.uri().path().to_string());
-		let method = req.method().clone();
-		let method_str = method.as_str();
 
 		let start_time = Instant::now();
 
@@ -642,11 +640,6 @@ impl ProxyService {
 		};
 
 		let actor_id = target.actor_id;
-		let server_id = target.server_id;
-
-		// Convert UUIDs to strings for metrics, handling Optional fields
-		let actor_id_str = actor_id.map_or_else(|| "none".to_string(), |id| id.to_string());
-		let server_id_str = server_id.map_or_else(|| "none".to_string(), |id| id.to_string());
 
 		// Extract IP address from remote_addr
 		let client_ip = self.remote_addr.ip();
@@ -666,13 +659,9 @@ impl ProxyService {
 				.map_err(Into::into)
 		} else {
 			// Increment metrics
-			metrics::PROXY_REQUEST_PENDING
-				.with_label_values(&[&actor_id_str, &server_id_str, method_str, &path])
-				.inc();
+			metrics::PROXY_REQUEST_PENDING.inc();
 
-			metrics::PROXY_REQUEST_TOTAL
-				.with_label_values(&[&actor_id_str, &server_id_str, method_str, &path])
-				.inc();
+			metrics::PROXY_REQUEST_TOTAL.inc();
 
 			// Prepare to release in-flight counter when done
 			let state_clone = self.state.clone();
@@ -695,18 +684,22 @@ impl ProxyService {
 
 		let status = match &res {
 			Ok(resp) => resp.status().as_u16().to_string(),
-			Err(_) => "error".to_string(),
+			Err(err) => {
+				metrics::PROXY_REQUEST_ERROR
+					.with_label_values(&[&err.to_string()])
+					.inc();
+
+				"error".to_string()
+			}
 		};
 
 		// Record metrics
 		let duration = start_time.elapsed();
 		metrics::PROXY_REQUEST_DURATION
-			.with_label_values(&[&actor_id_str, &server_id_str, &status])
+			.with_label_values(&[&status])
 			.observe(duration.as_secs_f64());
 
-		metrics::PROXY_REQUEST_PENDING
-			.with_label_values(&[&actor_id_str, &server_id_str, method_str, &path])
-			.dec();
+		metrics::PROXY_REQUEST_PENDING.dec();
 
 		res
 	}