feat(metrics): Record metrics with OpenTelemetry and export over OTLP

src/main/Cargo.toml

@@ -131,9 +131,7 @@ media_thumbnail = [
     "conduwuit-service/media_thumbnail",
 ]
 otlp_telemetry = [
-    "dep:opentelemetry",
     "dep:tracing-opentelemetry",
-    "dep:opentelemetry_sdk",
     "dep:opentelemetry-otlp",
 ]
 perf_measurements = [
@@ -221,12 +219,12 @@ console-subscriber.optional = true
 console-subscriber.workspace = true
 const-str.workspace = true
 log.workspace = true
-opentelemetry.optional = true
 opentelemetry.workspace = true
 opentelemetry-otlp.optional = true
 opentelemetry-otlp.workspace = true
-opentelemetry_sdk.optional = true
+opentelemetry-prometheus.workspace = true
 opentelemetry_sdk.workspace = true
+prometheus.workspace = true
 sentry-tower.optional = true
 sentry-tower.workspace = true
 sentry-tracing.optional = true

src/main/metrics.rs

@@ -0,0 +1,76 @@
+use conduwuit_core::{Result, config::Config, err, metrics::Metrics};
+use opentelemetry::{KeyValue, metrics::MeterProvider};
+use opentelemetry_sdk::{
+	Resource,
+	metrics::SdkMeterProvider,
+	resource::{EnvResourceDetector, TelemetryResourceDetector},
+};
+use tokio::runtime;
+
+/// The returned provider must outlive the server so pending OTLP exports are
+/// flushed on drop.
+pub(crate) fn init(
+	config: &Config,
+	runtime: Option<&runtime::Handle>,
+) -> Result<(Metrics, SdkMeterProvider)> {
+	// Defaults first so OTEL_RESOURCE_ATTRIBUTES and OTEL_SERVICE_NAME win.
+	let mut resource = Resource::builder_empty()
+		.with_detector(Box::new(TelemetryResourceDetector))
+		.with_service_name(conduwuit_core::BRANDING)
+		.with_attributes([
+			KeyValue::new("service.version", conduwuit_core::version()),
+			KeyValue::new("service.instance.id", config.server_name.to_string()),
+		])
+		.with_detector(Box::new(EnvResourceDetector::new()));
+	if let Some(name) = std::env::var("OTEL_SERVICE_NAME")
+		.ok()
+		.filter(|name| !name.is_empty())
+	{
+		resource = resource.with_service_name(name);
+	}
+	let resource = resource.build();
+
+	let mut builder = SdkMeterProvider::builder().with_resource(resource);
+
+	let registry = config
+		.allow_prometheus_metrics
+		.then(prometheus::Registry::new);
+	if let Some(registry) = &registry {
+		let exporter = opentelemetry_prometheus::exporter()
+			.with_registry(registry.clone())
+			.scope_info_enabled(false)
+			.build()
+			.map_err(|e| err!(Config("allow_prometheus_metrics", "{e}")))?;
+
+		builder = builder.with_reader(exporter);
+	}
+
+	#[cfg(feature = "otlp_telemetry")]
+	if config.allow_otlp_metrics {
+		use opentelemetry_otlp::{MetricExporter, Protocol, WithExportConfig};
+
+		let exporter = match config.otlp_protocol.as_str() {
+			| "grpc" => MetricExporter::builder()
+				.with_tonic()
+				.with_protocol(Protocol::Grpc)
+				.build(),
+			| _ => MetricExporter::builder().with_http().build(),
+		}
+		.map_err(|e| err!(Config("allow_otlp_metrics", "{e}")))?;
+
+		builder = builder.with_periodic_exporter(exporter);
+	}
+
+	#[cfg(not(feature = "otlp_telemetry"))]
+	if config.allow_otlp_metrics {
+		conduwuit_core::warn!(
+			"allow_otlp_metrics is set but this build does not include the otlp_telemetry \
+			 feature"
+		);
+	}
+
+	let provider = builder.build();
+	let meter = provider.meter(conduwuit_core::BRANDING);
+
+	Ok((Metrics::new(runtime.cloned(), meter, registry), provider))
+}

src/main/mod.rs

@@ -9,6 +9,7 @@ conduwuit_macros::introspect_crate! {}
 mod clap;
 mod deadlock;
 mod logging;
+mod metrics;
 mod mods;
 mod panic;
 mod restart;

src/main/server.rs

@@ -23,6 +23,8 @@ pub(crate) struct Server {
 
 	_tracing_flame_guard: TracingFlameGuard,
 
+	_meter_provider: opentelemetry_sdk::metrics::SdkMeterProvider,
+
 	#[cfg(feature = "sentry_telemetry")]
 	_sentry_guard: Option<::sentry::ClientInitGuard>,
 
@@ -56,6 +58,8 @@ impl Server {
 		sys::maximize_fd_limit()
 			.expect("Unable to increase maximum soft and hard file descriptor limit");
 
+		let (metrics, meter_provider) = crate::metrics::init(&config, runtime)?;
+
 		let (_old_width, _new_width) = stream::set_width(config.stream_width_default);
 		let (_old_amp, _new_amp) = stream::set_amplification(config.stream_amplification);
 
@@ -68,15 +72,19 @@ impl Server {
 		);
 
 		Ok(Arc::new(Self {
-			server: Arc::new(conduwuit_core::Server::new(config, runtime.cloned(), Log {
-				reload: tracing_reload_handle,
-				capture,
-			})),
+			server: Arc::new(conduwuit_core::Server::new(
+				config,
+				runtime.cloned(),
+				Log { reload: tracing_reload_handle, capture },
+				metrics,
+			)),
 
 			services: None.into(),
 
 			_tracing_flame_guard: tracing_flame_guard,
 
+			_meter_provider: meter_provider,
+
 			#[cfg(feature = "sentry_telemetry")]
 			_sentry_guard: sentry_guard,