18 releases

new 0.9.0 Sep 27, 2024
0.8.0 Feb 10, 2024
0.7.0 Jul 21, 2023
0.6.0 Feb 25, 2022
0.1.2 May 13, 2019

#66 in Web programming

Download history 7152/week @ 2024-06-14 7748/week @ 2024-06-21 7061/week @ 2024-06-28 7085/week @ 2024-07-05 7324/week @ 2024-07-12 8078/week @ 2024-07-19 7644/week @ 2024-07-26 7545/week @ 2024-08-02 7381/week @ 2024-08-09 6483/week @ 2024-08-16 7307/week @ 2024-08-23 7430/week @ 2024-08-30 7891/week @ 2024-09-06 6459/week @ 2024-09-13 7040/week @ 2024-09-20 5512/week @ 2024-09-27

28,526 downloads per month
Used in 7 crates

MIT license

72KB
1K SLoC

actix-web-prom

CI Status docs.rs crates.io MIT licensed

Prometheus instrumentation for actix-web. This middleware is heavily influenced by the work in sd2k/rocket_prometheus. We track the same default metrics and allow for adding user defined metrics.

By default two metrics are tracked (this assumes the namespace actix_web_prom):

  • actix_web_prom_http_requests_total (labels: endpoint, method, status): the total number of HTTP requests handled by the actix HttpServer.

  • actix_web_prom_http_requests_duration_seconds (labels: endpoint, method, status): the request duration for all HTTP requests handled by the actix HttpServer.

Usage

First add actix-web-prom to your Cargo.toml:

[dependencies]
actix-web-prom = "0.9.0"

You then instantiate the prometheus middleware and pass it to .wrap():

use std::collections::HashMap;

use actix_web::{web, App, HttpResponse, HttpServer};
use actix_web_prom::{PrometheusMetrics, PrometheusMetricsBuilder};

async fn health() -> HttpResponse {
    HttpResponse::Ok().finish()
}

#[actix_web::main]
async fn main() -> std::io::Result<()> {
    let mut labels = HashMap::new();
    labels.insert("label1".to_string(), "value1".to_string());
    let prometheus = PrometheusMetricsBuilder::new("api")
        .endpoint("/metrics")
        .const_labels(labels)
        .build()
        .unwrap();

        HttpServer::new(move || {
            App::new()
                .wrap(prometheus.clone())
                .service(web::resource("/health").to(health))
        })
        .bind("127.0.0.1:8080")?
        .run()
        .await?;
    Ok(())
}

Using the above as an example, a few things are worth mentioning:

  • api is the metrics namespace
  • /metrics will be auto exposed (GET requests only) with Content-Type header content-type: text/plain; version=0.0.4; charset=utf-8
  • Some(labels) is used to add fixed labels to the metrics; None can be passed instead if no additional labels are necessary.

A call to the /metrics endpoint will expose your metrics:

$ curl http://localhost:8080/metrics
# HELP api_http_requests_duration_seconds HTTP request duration in seconds for all requests
# TYPE api_http_requests_duration_seconds histogram
api_http_requests_duration_seconds_bucket{endpoint="/metrics",label1="value1",method="GET",status="200",le="0.005"} 1
api_http_requests_duration_seconds_bucket{endpoint="/metrics",label1="value1",method="GET",status="200",le="0.01"} 1
api_http_requests_duration_seconds_bucket{endpoint="/metrics",label1="value1",method="GET",status="200",le="0.025"} 1
api_http_requests_duration_seconds_bucket{endpoint="/metrics",label1="value1",method="GET",status="200",le="0.05"} 1
api_http_requests_duration_seconds_bucket{endpoint="/metrics",label1="value1",method="GET",status="200",le="0.1"} 1
api_http_requests_duration_seconds_bucket{endpoint="/metrics",label1="value1",method="GET",status="200",le="0.25"} 1
api_http_requests_duration_seconds_bucket{endpoint="/metrics",label1="value1",method="GET",status="200",le="0.5"} 1
api_http_requests_duration_seconds_bucket{endpoint="/metrics",label1="value1",method="GET",status="200",le="1"} 1
api_http_requests_duration_seconds_bucket{endpoint="/metrics",label1="value1",method="GET",status="200",le="2.5"} 1
api_http_requests_duration_seconds_bucket{endpoint="/metrics",label1="value1",method="GET",status="200",le="5"} 1
api_http_requests_duration_seconds_bucket{endpoint="/metrics",label1="value1",method="GET",status="200",le="10"} 1
api_http_requests_duration_seconds_bucket{endpoint="/metrics",label1="value1",method="GET",status="200",le="+Inf"} 1
api_http_requests_duration_seconds_sum{endpoint="/metrics",label1="value1",method="GET",status="200"} 0.00003
api_http_requests_duration_seconds_count{endpoint="/metrics",label1="value1",method="GET",status="200"} 1
# HELP api_http_requests_total Total number of HTTP requests
# TYPE api_http_requests_total counter
api_http_requests_total{endpoint="/metrics",label1="value1",method="GET",status="200"} 1

Features

If you enable process feature of this crate, default process metrics will also be collected. Default process metrics

# HELP process_cpu_seconds_total Total user and system CPU time spent in seconds.
# TYPE process_cpu_seconds_total counter
process_cpu_seconds_total 0.22
# HELP process_max_fds Maximum number of open file descriptors.
# TYPE process_max_fds gauge
process_max_fds 1048576
# HELP process_open_fds Number of open file descriptors.
# TYPE process_open_fds gauge
process_open_fds 78
# HELP process_resident_memory_bytes Resident memory size in bytes.
# TYPE process_resident_memory_bytes gauge
process_resident_memory_bytes 17526784
# HELP process_start_time_seconds Start time of the process since unix epoch in seconds.
# TYPE process_start_time_seconds gauge
process_start_time_seconds 1628105774.92
# HELP process_virtual_memory_bytes Virtual memory size in bytes.
# TYPE process_virtual_memory_bytes gauge
process_virtual_memory_bytes 1893163008

Custom metrics

You instantiate PrometheusMetrics and then use its .registry to register your custom metric (in this case, we use a IntCounterVec).

Then you can pass this counter through .data() to have it available within the resource responder.

use actix_web::{web, App, HttpResponse, HttpServer};
use actix_web_prom::{PrometheusMetrics, PrometheusMetricsBuilder};
use prometheus::{opts, IntCounterVec};

async fn health(counter: web::Data<IntCounterVec>) -> HttpResponse {
    counter.with_label_values(&["endpoint", "method", "status"]).inc();
    HttpResponse::Ok().finish()
}

#[actix_web::main]
async fn main() -> std::io::Result<()> {
    let prometheus = PrometheusMetricsBuilder::new("api")
        .endpoint("/metrics")
        .build()
        .unwrap();

    let counter_opts = opts!("counter", "some random counter").namespace("api");
    let counter = IntCounterVec::new(counter_opts, &["endpoint", "method", "status"]).unwrap();
    prometheus
        .registry
        .register(Box::new(counter.clone()))
        .unwrap();

        HttpServer::new(move || {
            App::new()
                .wrap(prometheus.clone())
                .data(counter.clone())
                .service(web::resource("/health").to(health))
        })
        .bind("127.0.0.1:8080")?
        .run()
        .await?;
    Ok(())
}

Custom Registry

Some apps might have more than one actix_web::HttpServer. If that's the case, you might want to use your own registry:

use actix_web::{web, App, HttpResponse, HttpServer};
use actix_web_prom::{PrometheusMetrics, PrometheusMetricsBuilder};
use actix_web::rt::System;
use prometheus::Registry;
use std::thread;

async fn public_handler() -> HttpResponse {
    HttpResponse::Ok().body("Everyone can see it!")
}

async fn private_handler() -> HttpResponse {
    HttpResponse::Ok().body("This can be hidden behind a firewall")
}

fn main() -> std::io::Result<()> {
    let shared_registry = Registry::new();

    let private_metrics = PrometheusMetricsBuilder::new("private_api")
        .registry(shared_registry.clone())
        .endpoint("/metrics")
        .build()
        // It is safe to unwrap when __no other app has the same namespace__
        .unwrap();

    let public_metrics = PrometheusMetricsBuilder::new("public_api")
        .registry(shared_registry.clone())
        // Metrics should not be available from the outside
        // so no endpoint is registered
        .build()
        .unwrap();

    let private_thread = thread::spawn(move || {
        let mut sys = System::new();
        let srv = HttpServer::new(move || {
            App::new()
                .wrap(private_metrics.clone())
                .service(web::resource("/test").to(private_handler))
        })
        .bind("127.0.0.1:8081")
        .unwrap()
        .run();
        sys.block_on(srv).unwrap();
    });

    let public_thread = thread::spawn(|| {
        let mut sys = System::new();
        let srv = HttpServer::new(move || {
            App::new()
                .wrap(public_metrics.clone())
                .service(web::resource("/test").to(public_handler))
        })
        .bind("127.0.0.1:8082")
        .unwrap()
        .run();
        sys.block_on(srv).unwrap();
    });

    private_thread.join().unwrap();
    public_thread.join().unwrap();
    Ok(())
}

Configurable routes pattern cardinality

Let's say you have on your app a route to fetch posts by language and by slug GET /posts/{language}/{slug}. By default, actix-web-prom will provide metrics for the whole route with the label endpoint set to the pattern /posts/{language}/{slug}. This is great but you cannot differentiate metrics across languages (as there is only a limited set of them). Actix-web-prom can be configured to allow for more cardinality on some route params.

For that you need to add a middleware to pass some extensions data, specifically the MetricsConfig struct that contains the list of params you want to keep cardinality on.

use actix_web::{dev::Service, web, HttpMessage, HttpResponse};
use actix_web_prom::MetricsConfig;

async fn handler() -> HttpResponse {
    HttpResponse::Ok().finish()
}

web::resource("/posts/{language}/{slug}")
    .wrap_fn(|req, srv| {
        req.extensions_mut().insert::<MetricsConfig>(
            MetricsConfig { cardinality_keep_params: vec!["language".to_string()] }
        );
        srv.call(req)
    })
    .route(web::get().to(handler));

See the full example with_cardinality_on_params.rs.

Configurable metric names

If you want to rename the default metrics, you can use ActixMetricsConfiguration to do so.

use actix_web_prom::{PrometheusMetricsBuilder, ActixMetricsConfiguration};

PrometheusMetricsBuilder::new("api")
    .endpoint("/metrics")
    .metrics_configuration(
        ActixMetricsConfiguration::default()
        .http_requests_duration_seconds_name("my_http_request_duration"),
    )
    .build()
    .unwrap();

See full example configuring_default_metrics.rs.

Masking unknown paths

This is useful to avoid producting lots and lots of useless metrics due to bots on the internet.

What this does is transform a path that will never be found (404) into one single metric. So, if you want metrics about every single path that is hit, even if it doesn't exist, avoid this section altogether.

use actix_web_prom::PrometheusMetricsBuilder;

PrometheusMetricsBuilder::new("api")
    .endpoint("/metrics")
    .mask_unmatched_patterns("UNKNOWN")
    .build()
    .unwrap();

The above will convert all /<nonexistent-path> into UNKNOWN:

http_requests_duration_seconds_sum{endpoint="/favicon.ico",method="GET",status="400"} 0.000424898

becomes

http_requests_duration_seconds_sum{endpoint="UNKNOWN",method="GET",status="400"} 0.000424898

Dependencies

~17–30MB
~540K SLoC