12 releases (6 breaking)

0.7.0 Jul 20, 2024
0.6.1 Jan 23, 2024
0.5.0 Nov 27, 2023
0.4.0 Jul 24, 2023
0.2.0 Nov 25, 2022

#139 in Network programming

Download history 7752/week @ 2024-08-05 7364/week @ 2024-08-12 8184/week @ 2024-08-19 6984/week @ 2024-08-26 7452/week @ 2024-09-02 9343/week @ 2024-09-09 7375/week @ 2024-09-16 8272/week @ 2024-09-23 9313/week @ 2024-09-30 8225/week @ 2024-10-07 9647/week @ 2024-10-14 9493/week @ 2024-10-21 9241/week @ 2024-10-28 9124/week @ 2024-11-04 10053/week @ 2024-11-11 14491/week @ 2024-11-18

43,325 downloads per month
Used in 9 crates (8 directly)

MIT license

76KB
982 lines

Axum-Prometheus

A middleware to collect HTTP metrics for Axum applications.

axum-prometheus relies on metrics.rs and its ecosystem to collect and export metrics - for instance for Prometheus, metrics_exporter_prometheus is used as a backend to interact with Prometheus.

Metrics

By default three HTTP metrics are tracked

  • axum_http_requests_total (labels: endpoint, method, status): the total number of HTTP requests handled (counter)
  • axum_http_requests_duration_seconds (labels: endpoint, method, status): the request duration for all HTTP requests handled (histogram)
  • axum_http_requests_pending (labels: endpoint, method): the number of currently in-flight requests (gauge)

This crate also allows to track response body sizes as a histogram — see PrometheusMetricLayerBuilder::enable_response_body_size.

Renaming Metrics

These metrics can be renamed by specifying environmental variables at compile time:

  • AXUM_HTTP_REQUESTS_TOTAL
  • AXUM_HTTP_REQUESTS_DURATION_SECONDS
  • AXUM_HTTP_REQUESTS_PENDING
  • AXUM_HTTP_RESPONSE_BODY_SIZE (if body size tracking is enabled)

These environmental variables can be set in your .cargo/config.toml since Cargo 1.56:

[env]
AXUM_HTTP_REQUESTS_TOTAL = "my_app_requests_total"
AXUM_HTTP_REQUESTS_DURATION_SECONDS = "my_app_requests_duration_seconds"
AXUM_HTTP_REQUESTS_PENDING = "my_app_requests_pending"
AXUM_HTTP_RESPONSE_BODY_SIZE = "my_app_response_body_size"

..or optionally use PrometheusMetricLayerBuilder::with_prefix function.

Compatibility

Axum Version Crate Version
0.5 0.1
0.6 0.2, 0.3, 0.4
0.7 0.5, 0.6, 0.7

MSRV

This crate's current MSRV is 1.70.

Usage

For more elaborate use-cases, see the builder example.

Add axum-prometheus to your Cargo.toml.

[dependencies]
axum-prometheus = "0.7.0"

Then you instantiate the prometheus middleware:

use std::{net::SocketAddr, time::Duration};
use axum::{routing::get, Router};
use axum_prometheus::PrometheusMetricLayer;

#[tokio::main]
async fn main() {
    let (prometheus_layer, metric_handle) = PrometheusMetricLayer::pair();
    let app = Router::<()>::new()
        .route("/fast", get(|| async {}))
        .route(
            "/slow",
            get(|| async {
                tokio::time::sleep(Duration::from_secs(1)).await;
            }),
        )
        .route("/metrics", get(|| async move { metric_handle.render() }))
        .layer(prometheus_layer);


    let listener = tokio::net::TcpListener::bind(SocketAddr::from(([127, 0, 0, 1], 3000)))
        .await
        .unwrap();
    axum::serve(listener, app).await.unwrap();
}

Note that the /metrics endpoint is not automatically exposed, so you need to add that as a route manually. Calling the /metrics endpoint will expose your metrics:

axum_http_requests_total{method="GET",endpoint="/metrics",status="200"} 5
axum_http_requests_pending{method="GET",endpoint="/metrics"} 1
axum_http_requests_duration_seconds_bucket{method="GET",status="200",endpoint="/metrics",le="0.005"} 4
axum_http_requests_duration_seconds_bucket{method="GET",status="200",endpoint="/metrics",le="0.01"} 4
axum_http_requests_duration_seconds_bucket{method="GET",status="200",endpoint="/metrics",le="0.025"} 4
axum_http_requests_duration_seconds_bucket{method="GET",status="200",endpoint="/metrics",le="0.05"} 4
axum_http_requests_duration_seconds_bucket{method="GET",status="200",endpoint="/metrics",le="0.1"} 4
axum_http_requests_duration_seconds_bucket{method="GET",status="200",endpoint="/metrics",le="0.25"} 4
axum_http_requests_duration_seconds_bucket{method="GET",status="200",endpoint="/metrics",le="0.5"} 4
axum_http_requests_duration_seconds_bucket{method="GET",status="200",endpoint="/metrics",le="1"} 4
axum_http_requests_duration_seconds_bucket{method="GET",status="200",endpoint="/metrics",le="2.5"} 4
axum_http_requests_duration_seconds_bucket{method="GET",status="200",endpoint="/metrics",le="5"} 4
axum_http_requests_duration_seconds_bucket{method="GET",status="200",endpoint="/metrics",le="10"} 4
axum_http_requests_duration_seconds_bucket{method="GET",status="200",endpoint="/metrics",le="+Inf"} 4
axum_http_requests_duration_seconds_sum{method="GET",status="200",endpoint="/metrics"} 0.001997171
axum_http_requests_duration_seconds_count{method="GET",status="200",endpoint="/metrics"} 4

Let's note that since metrics-exporter-prometheus = "0.13" that crate introduced the push-gateway default feature, that requires openssl support. The axum_prometheus crate does not rely on, nor enable this feature by default — if you need it, you may enable it through the "push-gateway" feature in axum_prometheus.

Prometheus push gateway feature

This crate currently has no higher level API for the push-gateway feature. If you plan to use it, enable the push-gateway feature in axum-prometheus, use BaseMetricLayer, and setup your recorder manually, similar to the base-metric-layer-example.

Using a different exporter than Prometheus

This crate may be used with other exporters than Prometheus. First, disable the default features:

axum-prometheus = { version = "0.7.0", default-features = false }

Then implement the MakeDefaultHandle for the provider you'd like to use. For StatsD:

use metrics_exporter_statsd::StatsdBuilder;
use axum_prometheus::{MakeDefaultHandle, GenericMetricLayer};

// The custom StatsD exporter struct. It may take fields as well.
struct Recorder { port: u16 }

// In order to use this with `axum_prometheus`, we must implement `MakeDefaultHandle`.
impl MakeDefaultHandle for Recorder {
    // We don't need to return anything meaningful from here (unlike PrometheusHandle)
    // Let's just return an empty tuple.
    type Out = ();

    fn make_default_handle(self) -> Self::Out {
        // The regular setup for StatsD. Notice that `self` is passed in by value.
        let recorder = StatsdBuilder::from("127.0.0.1", self.port)
            .with_queue_size(5000)
            .with_buffer_size(1024)
            .build(Some("prefix"))
            .expect("Could not create StatsdRecorder");

        metrics::set_boxed_recorder(Box::new(recorder)).unwrap();
    }
}

fn main() {
    // Use `GenericMetricLayer` instead of `PrometheusMetricLayer`.
    // Generally `GenericMetricLayer::pair_from` is what you're looking for.
    // It lets you pass in a concrete initialized `Recorder`.
    let (metric_layer, _handle) = GenericMetricLayer::pair_from(Recorder { port: 8125 });
}

It's also possible to use GenericMetricLayer::pair, however it's only callable if the recorder struct implements Default as well.

use metrics_exporter_statsd::StatsdBuilder;
use axum_prometheus::{MakeDefaultHandle, GenericMetricLayer};

#[derive(Default)]
struct Recorder { port: u16 }

impl MakeDefaultHandle for Recorder {
   /* .. same as before .. */
}

fn main() {
    // This will internally call `Recorder::make_default_handle(Recorder::default)`.
    let (metric_layer, _handle) = GenericMetricLayer::<_, Recorder>::pair();
}

This crate is similar to (and takes inspiration from) actix-web-prom and rocket_prometheus, and also builds on top of davidpdrsn's earlier work with LifeCycleHooks in tower-http.

Dependencies

~7–15MB
~189K SLoC