Module tower_http::trace

This is supported on crate feature trace only.

Expand description

Middleware that adds high level tracing to a Service.

Example

Adding tracing to your service can be as simple as:

use http::{Request, Response};
use hyper::Body;
use tower::{ServiceBuilder, ServiceExt, Service};
use tower_http::trace::TraceLayer;
use std::convert::Infallible;

async fn handle(request: Request<Body>) -> Result<Response<Body>, Infallible> {
    Ok(Response::new(Body::from("foo")))
}

// Setup tracing
tracing_subscriber::fmt::init();

let mut service = ServiceBuilder::new()
    .layer(TraceLayer::new_for_http())
    .service_fn(handle);

let request = Request::new(Body::from("foo"));

let response = service
    .ready()
    .await?
    .call(request)
    .await?;

If you run this application with RUST_LOG=tower_http=trace cargo run you should see logs like:

Mar 05 20:50:28.523 DEBUG request{method=GET path="/foo"}: tower_http::trace::on_request: started processing request
Mar 05 20:50:28.524 DEBUG request{method=GET path="/foo"}: tower_http::trace::on_response: finished processing request latency=1 ms status=200

Customization

Trace comes with good defaults but also supports customizing many aspects of the output.

The default behaviour supports some customization:

use http::{Request, Response, HeaderMap, StatusCode};
use hyper::Body;
use bytes::Bytes;
use tower::ServiceBuilder;
use tracing::Level;
use tower_http::{
    LatencyUnit,
    trace::{TraceLayer, DefaultMakeSpan, DefaultOnRequest, DefaultOnResponse},
};
use std::time::Duration;

let service = ServiceBuilder::new()
    .layer(
        TraceLayer::new_for_http()
            .make_span_with(
                DefaultMakeSpan::new().include_headers(true)
            )
            .on_request(
                DefaultOnRequest::new().level(Level::INFO)
            )
            .on_response(
                DefaultOnResponse::new()
                    .level(Level::INFO)
                    .latency_unit(LatencyUnit::Micros)
            )
            // on so on for `on_eos`, `on_body_chunk`, and `on_failure`
    )
    .service_fn(handle);

However for maximum control you can provide callbacks:

use http::{Request, Response, HeaderMap, StatusCode};
use hyper::Body;
use bytes::Bytes;
use tower::ServiceBuilder;
use tower_http::{classify::ServerErrorsFailureClass, trace::TraceLayer};
use std::time::Duration;
use tracing::Span;

let service = ServiceBuilder::new()
    .layer(
        TraceLayer::new_for_http()
            .make_span_with(|request: &Request<Body>| {
                tracing::debug_span!("http-request")
            })
            .on_request(|request: &Request<Body>, _span: &Span| {
                tracing::debug!("started {} {}", request.method(), request.uri().path())
            })
            .on_response(|response: &Response<Body>, latency: Duration, _span: &Span| {
                tracing::debug!("response generated in {:?}", latency)
            })
            .on_body_chunk(|chunk: &Bytes, latency: Duration, _span: &Span| {
                tracing::debug!("sending {} bytes", chunk.len())
            })
            .on_eos(|trailers: Option<&HeaderMap>, stream_duration: Duration, _span: &Span| {
                tracing::debug!("stream closed after {:?}", stream_duration)
            })
            .on_failure(|error: ServerErrorsFailureClass, latency: Duration, _span: &Span| {
                tracing::debug!("something went wrong")
            })
    )
    .service_fn(handle);

Disabling something

Setting the behaviour to () will be disable that particular step:

use http::StatusCode;
use tower::ServiceBuilder;
use tower_http::{classify::ServerErrorsFailureClass, trace::TraceLayer};
use std::time::Duration;
use tracing::Span;

let service = ServiceBuilder::new()
    .layer(
        // This configuration will only emit events on failures
        TraceLayer::new_for_http()
            .on_request(())
            .on_response(())
            .on_body_chunk(())
            .on_eos(())
            .on_failure(|error: ServerErrorsFailureClass, latency: Duration, _span: &Span| {
                tracing::debug!("something went wrong")
            })
    )
    .service_fn(handle);

When the callbacks are called

`on_request`

The on_request callback is called when the request arrives at the middleware in Service::call just prior to passing the request to the inner service.

`on_response`

The on_response callback is called when the inner service’s response future completes with Ok(response) regardless if the response is classified as a success or a failure.

For example if you’re using ServerErrorsAsFailures as your classifier and the inner service responds with 500 Internal Server Error then the on_response callback is still called. on_failure would also be called in this case since the response was classified as a failure.

The inner Service’s response future resolves to an error.
A response is classified as a failure.
Body::poll_data returns an error.
Body::poll_trailers returns an error.
An end-of-stream is classified as a failure.

Recording fields on the span

All callbacks receive a reference to the tracing Span, corresponding to this request, produced by the closure passed to TraceLayer::make_span_with. It can be used to record field values that weren’t known when the span was created.

use http::{Request, Response, HeaderMap, StatusCode};
use hyper::Body;
use bytes::Bytes;
use tower::ServiceBuilder;
use tower_http::trace::TraceLayer;
use tracing::Span;
use std::time::Duration;

let service = ServiceBuilder::new()
    .layer(
        TraceLayer::new_for_http()
            .make_span_with(|request: &Request<Body>| {
                tracing::debug_span!(
                    "http-request",
                    status_code = tracing::field::Empty,
                )
            })
            .on_response(|response: &Response<Body>, _latency: Duration, span: &Span| {
                span.record("status_code", &tracing::field::display(response.status()));

                tracing::debug!("response generated")
            })
    )
    .service_fn(handle);

Providing classifiers

Tracing requires determining if a response is a success or failure. MakeClassifier is used to create a classifier for the incoming request. See the docs for MakeClassifier and ClassifyResponse for more details on classification.

A MakeClassifier can be provided when creating a TraceLayer:

use http::{Request, Response};
use hyper::Body;
use tower::ServiceBuilder;
use tower_http::{
    trace::TraceLayer,
    classify::{
        MakeClassifier, ClassifyResponse, ClassifiedResponse, NeverClassifyEos,
        SharedClassifier,
    },
};
use std::convert::Infallible;

// Our `MakeClassifier` that always crates `MyClassifier` classifiers.
#[derive(Copy, Clone)]
struct MyMakeClassify;

impl MakeClassifier for MyMakeClassify {
    type Classifier = MyClassifier;
    type FailureClass = &'static str;
    type ClassifyEos = NeverClassifyEos<&'static str>;

    fn make_classifier<B>(&self, req: &Request<B>) -> Self::Classifier {
        MyClassifier
    }
}

// A classifier that classifies failures as `"something went wrong..."`.
#[derive(Copy, Clone)]
struct MyClassifier;

impl ClassifyResponse for MyClassifier {
    type FailureClass = &'static str;
    type ClassifyEos = NeverClassifyEos<&'static str>;

    fn classify_response<B>(
        self,
        res: &Response<B>
    ) -> ClassifiedResponse<Self::FailureClass, Self::ClassifyEos> {
        // Classify based on the status code.
        if res.status().is_server_error() {
            ClassifiedResponse::Ready(Err("something went wrong..."))
        } else {
            ClassifiedResponse::Ready(Ok(()))
        }
    }

    fn classify_error<E>(self, error: &E) -> Self::FailureClass
    where
        E: std::fmt::Display + 'static,
    {
        "something went wrong..."
    }
}

let service = ServiceBuilder::new()
    // Create a trace layer that uses our classifier.
    .layer(TraceLayer::new(MyMakeClassify))
    .service_fn(handle);

// Since `MyClassifier` is `Clone` we can also use `SharedClassifier`
// to avoid having to define a separate `MakeClassifier`.
let service = ServiceBuilder::new()
    .layer(TraceLayer::new(SharedClassifier::new(MyClassifier)))
    .service_fn(handle);

TraceLayer comes with convenience methods for using common classifiers:

TraceLayer::new_for_http classifies based on the status code. It doesn’t consider streaming responses.
TraceLayer::new_for_grpc classifies based on the gRPC protocol and supports streaming responses.