refactor(metrics): split server metrics crate

Author: NathanFlurry

Cargo.lock

@@ -29,6 +29,7 @@ members = [
   "engine/packages/guard-core",
   "engine/packages/logs",
   "engine/packages/metrics",
+  "engine/packages/metrics-server",
   "engine/packages/namespace",
   "engine/packages/pegboard",
   "engine/packages/pegboard-envoy",
@@ -452,6 +453,9 @@ members = [
     [workspace.dependencies.rivet-metrics]
     path = "engine/packages/metrics"
 
+    [workspace.dependencies.rivet-metrics-server]
+    path = "engine/packages/metrics-server"
+
     [workspace.dependencies.namespace]
     path = "engine/packages/namespace"
 

Cargo.toml

@@ -0,0 +1,20 @@
+[package]
+name = "rivet-metrics-server"
+version.workspace = true
+authors.workspace = true
+license.workspace = true
+edition.workspace = true
+
+[dependencies]
+anyhow.workspace = true
+console-subscriber.workspace = true
+hyper.workspace = true
+opentelemetry.workspace = true
+opentelemetry-otlp.workspace = true
+opentelemetry-semantic-conventions.workspace = true
+opentelemetry_sdk.workspace = true
+rivet-config.workspace = true
+rivet-env.workspace = true
+rivet-metrics.workspace = true
+tokio.workspace = true
+tracing.workspace = true

engine/packages/metrics-server/Cargo.toml

@@ -0,0 +1,5 @@
+mod providers;
+mod server;
+
+pub use providers::{OtelProviderGuard, init_otel_providers, set_sampler_ratio};
+pub use server::run_standalone;

engine/packages/metrics-server/src/lib.rs

@@ -1,6 +1,3 @@
-// Based off of https://github.com/tokio-rs/tracing-opentelemetry/blob/v0.1.x/examples/opentelemetry-otlp.rs
-// Based off of https://github.com/tokio-rs/tracing-opentelemetry/blob/v0.1.x/examples/opentelemetry-otlp.rs
-
 use opentelemetry::KeyValue;
 use opentelemetry::trace::{SamplingResult, SpanKind};
 use opentelemetry_otlp::WithExportConfig;
@@ -11,7 +8,7 @@ use opentelemetry_sdk::{
 use opentelemetry_semantic_conventions::{SCHEMA_URL, attribute::SERVICE_VERSION};
 use std::sync::{Arc, OnceLock, RwLock};
 
-/// Dynamic sampler that can be updated at runtime
+/// Dynamic sampler that can be updated at runtime.
 #[derive(Clone, Debug)]
 struct DynamicSampler {
 	ratio: Arc<RwLock<f64>>,
@@ -43,7 +40,6 @@ impl opentelemetry_sdk::trace::ShouldSample for DynamicSampler {
 	) -> SamplingResult {
 		let ratio = self.ratio.read().ok().map(|r| *r).unwrap_or(0.001);
 
-		// Use TraceIdRatioBased sampling logic
 		let sampler = Sampler::TraceIdRatioBased(ratio);
 		sampler.should_sample(
 			parent_context,
@@ -58,7 +54,7 @@ impl opentelemetry_sdk::trace::ShouldSample for DynamicSampler {
 
 static SAMPLER: OnceLock<DynamicSampler> = OnceLock::new();
 
-/// Update the sampler ratio at runtime
+/// Update the sampler ratio at runtime.
 pub fn set_sampler_ratio(ratio: f64) -> anyhow::Result<()> {
 	let sampler = SAMPLER
 		.get()
@@ -94,21 +90,17 @@ fn init_tracer_provider() -> SdkTracerProvider {
 		.build()
 		.unwrap();
 
-	// Create dynamic sampler with initial ratio from env
 	let initial_ratio = std::env::var("RIVET_OTEL_SAMPLER_RATIO")
 		.ok()
 		.and_then(|s| s.parse::<f64>().ok())
 		.unwrap_or(0.001);
 
 	let dynamic_sampler = DynamicSampler::new(initial_ratio);
 
-	// Store sampler globally for later updates
 	let _ = SAMPLER.set(dynamic_sampler.clone());
 
 	SdkTracerProvider::builder()
-		// Customize sampling strategy with parent-based sampling using our dynamic sampler
 		.with_sampler(Sampler::ParentBased(Box::new(dynamic_sampler)))
-		// If export trace to AWS X-Ray, you can use XrayIdGenerator
 		.with_id_generator(RandomIdGenerator::default())
 		.with_resource(resource())
 		.with_batch_exporter(exporter)
@@ -117,7 +109,6 @@ fn init_tracer_provider() -> SdkTracerProvider {
 
 /// Initialize OtelProviderGuard for opentelemetry-related termination processing.
 pub fn init_otel_providers() -> Option<OtelProviderGuard> {
-	// Check if otel is enabled
 	let enable_otel = std::env::var("RIVET_OTEL_ENABLED").map_or(false, |x| x == "1");
 
 	if enable_otel {
@@ -137,7 +128,7 @@ pub struct OtelProviderGuard {
 impl Drop for OtelProviderGuard {
 	fn drop(&mut self) {
 		if let Err(err) = self.tracer_provider.shutdown() {
-			eprintln!("{err:?}");
+			tracing::error!(?err, "failed to shut down otel tracer provider");
 		}
 	}
 }

engine/packages/metrics/src/providers.rs …packages/metrics-server/src/providers.rsengine/packages/metrics/src/providers.rs renamed to engine/packages/metrics-server/src/providers.rs engine/packages/metrics/src/providers.rs renamed to engine/packages/metrics-server/src/providers.rs

@@ -6,9 +6,7 @@ use hyper::{
 	header::CONTENT_TYPE,
 	service::{make_service_fn, service_fn},
 };
-use prometheus::{Encoder, TextEncoder};
-
-// TODO: Record extra labels
+use rivet_metrics::prometheus::{Encoder, TextEncoder};
 
 #[tracing::instrument(skip_all)]
 pub async fn run_standalone(config: rivet_config::Config) -> Result<()> {
@@ -21,8 +19,7 @@ pub async fn run_standalone(config: rivet_config::Config) -> Result<()> {
 		Err(err) => {
 			tracing::error!(?host, ?port, ?err, "failed to bind metrics server");
 
-			// TODO: Find cleaner way of crashing entire program
-			// Hard crash program since a server failing to bind is critical
+			// Hard crash the program since a server failing to bind is critical.
 			std::process::exit(1);
 		}
 	};
@@ -41,7 +38,7 @@ pub async fn run_standalone(config: rivet_config::Config) -> Result<()> {
 async fn serve_req(_req: Request<Body>) -> Result<Response<Body>, hyper::Error> {
 	let encoder = TextEncoder::new();
 
-	let metric_families = crate::registry::REGISTRY.gather();
+	let metric_families = rivet_metrics::REGISTRY.gather();
 	let mut buffer = Vec::new();
 	encoder
 		.encode(&metric_families, &mut buffer)

engine/packages/metrics/src/server.rs …ne/packages/metrics-server/src/server.rsengine/packages/metrics/src/server.rs renamed to engine/packages/metrics-server/src/server.rs engine/packages/metrics/src/server.rs renamed to engine/packages/metrics-server/src/server.rs

@@ -6,16 +6,5 @@ license.workspace = true
 edition.workspace = true
 
 [dependencies]
-anyhow.workspace = true
-console-subscriber.workspace = true
-hyper.workspace = true
 lazy_static.workspace = true
-opentelemetry.workspace = true
-opentelemetry-otlp.workspace = true
-opentelemetry-semantic-conventions.workspace = true
-opentelemetry_sdk.workspace = true
 prometheus.workspace = true
-rivet-config.workspace = true
-rivet-env.workspace = true
-tokio.workspace = true
-tracing.workspace = true

engine/packages/metrics/Cargo.toml

@@ -1,11 +1,6 @@
-mod providers;
-
 mod buckets;
 mod registry;
-mod server;
 
 pub use buckets::{BUCKETS, MICRO_BUCKETS, TASK_POLL_BUCKETS};
 pub use prometheus;
-pub use providers::{OtelProviderGuard, init_otel_providers, set_sampler_ratio};
 pub use registry::REGISTRY;
-pub use server::run_standalone;

engine/packages/metrics/src/lib.rs

@@ -17,6 +17,7 @@ opentelemetry.workspace = true
 reqwest.workspace = true
 rivet-env.workspace = true
 rivet-metrics.workspace = true
+rivet-metrics-server.workspace = true
 sentry.workspace = true
 serde_json.workspace = true
 serde.workspace = true

engine/packages/runtime/Cargo.toml

@@ -1,6 +1,6 @@
 use std::{env, future::Future, sync::Arc, time::Duration};
 
-use rivet_metrics::init_otel_providers;
+use rivet_metrics_server::init_otel_providers;
 use tokio::sync::{Notify, OnceCell};
 mod metrics;
 mod traces;