|
| 1 | +//! HTTP server for metrics collection and exposition |
| 2 | +
|
| 3 | +use parking_lot::Mutex; |
| 4 | +use std::net::SocketAddr; |
| 5 | +use std::sync::Arc; |
| 6 | + |
| 7 | +use axum::{ |
| 8 | + Router, |
| 9 | + extract::State, |
| 10 | + http::{StatusCode, header}, |
| 11 | + response::IntoResponse, |
| 12 | + routing::get, |
| 13 | +}; |
| 14 | +use opentelemetry_sdk::metrics::{ManualReader, data::ResourceMetrics, reader::MetricReader as _}; |
| 15 | +use tokio::net::TcpListener; |
| 16 | +use tracing::error; |
| 17 | + |
| 18 | +use crate::prometheus::{MetricContainer, convert_to_prometheus, encode_registry}; |
| 19 | + |
| 20 | +/// Start a metrics server that binds synchronously and serves asynchronously. |
| 21 | +/// Returns the bound socket address after successful binding. |
| 22 | +/// The server continues running in the spawned task. |
| 23 | +pub(crate) async fn start_metrics_server( |
| 24 | + address: &str, |
| 25 | + reader: Arc<ManualReader>, |
| 26 | +) -> anyhow::Result<SocketAddr> { |
| 27 | + let addr: SocketAddr = address.parse().map_err(|err| { |
| 28 | + anyhow::anyhow!( |
| 29 | + "Failed to parse metrics listen address '{}': {}", |
| 30 | + address, |
| 31 | + err |
| 32 | + ) |
| 33 | + })?; |
| 34 | + |
| 35 | + let app = Router::new() |
| 36 | + .route("/metrics", get(manual_metrics_handler)) |
| 37 | + .with_state(reader); |
| 38 | + |
| 39 | + // Bind synchronously to catch binding errors immediately |
| 40 | + let listener = TcpListener::bind(addr) |
| 41 | + .await |
| 42 | + .map_err(|err| anyhow::anyhow!("Failed to bind to {}: {}", addr, err))?; |
| 43 | + |
| 44 | + let bound_addr = listener |
| 45 | + .local_addr() |
| 46 | + .map_err(|err| anyhow::anyhow!("Failed to get local address: {}", err))?; |
| 47 | + |
| 48 | + // Spawn the server task to run asynchronously |
| 49 | + tokio::spawn(async move { |
| 50 | + if let Err(err) = axum::serve(listener, app).await { |
| 51 | + error!("Metrics server error: {}", err); |
| 52 | + } |
| 53 | + }); |
| 54 | + |
| 55 | + Ok(bound_addr) |
| 56 | +} |
| 57 | + |
| 58 | +/// Handler for the ManualReader-based /metrics endpoint |
| 59 | +/// This collects metrics on-demand from `OpenTelemetry's` `ManualReader` |
| 60 | +async fn manual_metrics_handler(State(reader): State<Arc<ManualReader>>) -> impl IntoResponse { |
| 61 | + // This handler is picking up data from telemetry SDK's ManualReader, |
| 62 | + // this is a temporary solution to expose metrics in different ways |
| 63 | + // (pull and push). |
| 64 | + // This is to be replaced in the future with a less complex solution, |
| 65 | + // using only a single approach. |
| 66 | + let mut resource_metrics = ResourceMetrics::default(); |
| 67 | + |
| 68 | + // Collect metrics from ManualReader |
| 69 | + match reader.collect(&mut resource_metrics) { |
| 70 | + Ok(_) => { |
| 71 | + let metrics = Arc::new(Mutex::new(MetricContainer::new())); |
| 72 | + |
| 73 | + // Convert ResourceMetrics to Prometheus metrics and get the registry |
| 74 | + let registry = convert_to_prometheus(&resource_metrics, &metrics); |
| 75 | + |
| 76 | + // Encode metrics to Prometheus text format |
| 77 | + match encode_registry(®istry) { |
| 78 | + Ok(buffer) => ( |
| 79 | + StatusCode::OK, |
| 80 | + [(header::CONTENT_TYPE, "text/plain; version=0.0.4")], |
| 81 | + buffer, |
| 82 | + ), |
| 83 | + Err(err) => { |
| 84 | + error!("Failed to encode metrics: {}", err); |
| 85 | + ( |
| 86 | + StatusCode::INTERNAL_SERVER_ERROR, |
| 87 | + [(header::CONTENT_TYPE, "text/plain")], |
| 88 | + format!("Failed to encode metrics: {err}"), |
| 89 | + ) |
| 90 | + } |
| 91 | + } |
| 92 | + } |
| 93 | + Err(err) => { |
| 94 | + error!("Failed to collect metrics from ManualReader: {}", err); |
| 95 | + ( |
| 96 | + StatusCode::INTERNAL_SERVER_ERROR, |
| 97 | + [(header::CONTENT_TYPE, "text/plain")], |
| 98 | + format!("Failed to collect metrics: {err}"), |
| 99 | + ) |
| 100 | + } |
| 101 | + } |
| 102 | +} |
0 commit comments