Skip to main content

aether_telemetry/
telemetry_runtime.rs

1use crate::error::{TelemetryInitError, TelemetryShutdownError};
2use crate::gen_ai_metrics::GenAiMetrics;
3use crate::genai_constants;
4use crate::otel_observer::OtelInstrumentation;
5use opentelemetry::KeyValue;
6use opentelemetry::metrics::MeterProvider as _;
7use opentelemetry::trace::TracerProvider as _;
8use opentelemetry_otlp::{MetricExporter, SpanExporter, WithExportConfig, WithHttpConfig};
9use opentelemetry_sdk::Resource;
10use opentelemetry_sdk::metrics::SdkMeterProvider;
11use opentelemetry_sdk::metrics::periodic_reader_with_async_runtime::PeriodicReader;
12use opentelemetry_sdk::runtime::Tokio;
13use opentelemetry_sdk::trace::span_processor_with_async_runtime::BatchSpanProcessor;
14use opentelemetry_sdk::trace::{Sampler, SdkTracerProvider};
15use std::collections::HashMap;
16
17pub struct TelemetryRuntime {
18    tracer_provider: SdkTracerProvider,
19    meter_provider: SdkMeterProvider,
20    instrumentation: OtelInstrumentation,
21}
22
23/// Fully-resolved telemetry configuration. Every field is required: defaults
24/// and settings-layer merging live in `TelemetrySettings`, so a value here is
25/// always deliberate.
26pub struct TelemetryConfig {
27    /// Base URL for the OTLP/HTTP collector, required whenever a signal needs an exporter.
28    pub endpoint: Option<String>,
29    /// Exact OTLP/HTTP trace export URL. When set, this takes precedence over the
30    /// trace URL derived from `endpoint`.
31    pub traces_endpoint: Option<String>,
32    /// Exact OTLP/HTTP metric export URL. When set, this takes precedence over the
33    /// metric URL derived from `endpoint`.
34    pub metrics_endpoint: Option<String>,
35    pub headers: HashMap<String, String>,
36    pub service_name: String,
37    pub service_version: String,
38    pub sample_ratio: f64,
39    pub capture_content: bool,
40    pub traces_enabled: bool,
41    pub metrics_enabled: bool,
42}
43
44impl TelemetryRuntime {
45    pub fn new(config: &TelemetryConfig) -> Result<Self, TelemetryInitError> {
46        if !(0.0..=1.0).contains(&config.sample_ratio) {
47            return Err(TelemetryInitError::InvalidSampleRatio(config.sample_ratio));
48        }
49
50        let http_client = build_http_client(&config.headers)?;
51        let resource = Resource::builder()
52            .with_service_name(config.service_name.clone())
53            .with_attribute(KeyValue::new(genai_constants::SERVICE_VERSION, config.service_version.clone()))
54            .build();
55        let scope = genai_constants::genai_instrumentation_scope(config.service_version.clone());
56        let tracer_provider = build_tracer_provider(config, resource.clone(), http_client.clone())?;
57        let meter_provider = build_meter_provider(config, resource, http_client)?;
58        let instrumentation = OtelInstrumentation {
59            tracer: tracer_provider.tracer_with_scope(scope.clone()),
60            metrics: GenAiMetrics::new(&meter_provider.meter_with_scope(scope)),
61            capture_content: config.capture_content,
62        };
63
64        Ok(Self { tracer_provider, meter_provider, instrumentation })
65    }
66
67    pub fn observer_factory(&self) -> aether_core::events::ObserverFactory {
68        let instrumentation = self.instrumentation.clone();
69        std::sync::Arc::new(move || Box::new(crate::OtelObserver::new(instrumentation.clone())))
70    }
71
72    /// Flushes and shuts down, logging any failure. For callers whose only
73    /// recovery is logging.
74    pub fn shutdown_or_log(&self) {
75        if let Err(error) = self.shutdown() {
76            tracing::error!("Failed to shutdown telemetry: {error}");
77        }
78    }
79
80    /// Flushes and shuts down every signal. Both providers are always attempted;
81    /// providers for disabled signals have no exporters.
82    pub fn shutdown(&self) -> Result<(), TelemetryShutdownError> {
83        let traces = self.tracer_provider.shutdown().map_err(TelemetryShutdownError::Trace);
84        let metrics = self.meter_provider.shutdown().map_err(TelemetryShutdownError::Metric);
85        traces.and(metrics)
86    }
87}
88
89impl TelemetryConfig {
90    fn signal_endpoint(&self, signal: &str) -> Result<String, TelemetryInitError> {
91        let signal_specific_endpoint = match signal {
92            "traces" => self.traces_endpoint.as_deref(),
93            "metrics" => self.metrics_endpoint.as_deref(),
94            _ => None,
95        };
96        if let Some(endpoint) = signal_specific_endpoint.filter(|endpoint| !endpoint.is_empty()) {
97            return Ok(endpoint.to_string());
98        }
99
100        Ok(format!("{}/v1/{signal}", self.required_endpoint()?.trim_end_matches('/')))
101    }
102
103    fn required_endpoint(&self) -> Result<&str, TelemetryInitError> {
104        self.endpoint.as_deref().filter(|endpoint| !endpoint.is_empty()).ok_or(TelemetryInitError::MissingOtlpEndpoint)
105    }
106}
107
108fn build_tracer_provider(
109    config: &TelemetryConfig,
110    resource: Resource,
111    http_client: reqwest::Client,
112) -> Result<SdkTracerProvider, TelemetryInitError> {
113    let builder = SdkTracerProvider::builder().with_resource(resource);
114    if !config.traces_enabled {
115        return Ok(builder.with_sampler(Sampler::AlwaysOff).build());
116    }
117
118    let endpoint = config.signal_endpoint("traces")?;
119    let exporter = SpanExporter::builder()
120        .with_http()
121        .with_endpoint(endpoint)
122        .with_http_client(http_client)
123        .build()
124        .map_err(TelemetryInitError::TraceExporter)?;
125
126    Ok(builder
127        .with_sampler(Sampler::TraceIdRatioBased(config.sample_ratio))
128        .with_span_processor(BatchSpanProcessor::builder(exporter, Tokio).build())
129        .build())
130}
131
132fn build_meter_provider(
133    config: &TelemetryConfig,
134    resource: Resource,
135    http_client: reqwest::Client,
136) -> Result<SdkMeterProvider, TelemetryInitError> {
137    let builder = SdkMeterProvider::builder().with_resource(resource);
138    if !config.metrics_enabled {
139        return Ok(builder.build());
140    }
141
142    let endpoint = config.signal_endpoint("metrics")?;
143    let exporter = MetricExporter::builder()
144        .with_http()
145        .with_endpoint(endpoint)
146        .with_http_client(http_client)
147        .build()
148        .map_err(TelemetryInitError::MetricExporter)?;
149
150    Ok(builder.with_reader(PeriodicReader::builder(exporter, Tokio).build()).build())
151}
152
153fn build_http_client(headers: &HashMap<String, String>) -> Result<reqwest::Client, TelemetryInitError> {
154    let mut parsed = reqwest::header::HeaderMap::new();
155    for (key, value) in headers {
156        let name = reqwest::header::HeaderName::from_bytes(key.as_bytes())
157            .map_err(|_| TelemetryInitError::InvalidHeaderName(key.clone()))?;
158        let value = reqwest::header::HeaderValue::from_str(value)
159            .map_err(|_| TelemetryInitError::InvalidHeaderValue(key.clone()))?;
160        parsed.insert(name, value);
161    }
162    reqwest::Client::builder().default_headers(parsed).build().map_err(TelemetryInitError::HttpClient)
163}