14 releases (7 breaking)

0.8.0 Mar 14, 2021
0.7.1 Nov 14, 2020
0.7.0 Sep 13, 2020
0.3.0 Jun 12, 2020

#316 in HTTP client

Download history 35/week @ 2024-02-21 19/week @ 2024-02-28

54 downloads per month

MIT/Apache

330KB
6K SLoC

hreq

hreq is a user first async http client and server.

Early days

This library needs road testing. Bug reports and PRs are very welcome!

Principles

The principles of this library are:

  • User first API built on the http crate.
  • async (or blocking via minimal runtime).
  • Pure Rust.
use hreq::prelude::*;

fn main() -> Result<(), hreq::Error> {
    // Use plain http API request builder with
    // trait extensions for extra convenience
    // in handling query parameters and other
    // request configurations.
    let response = Request::builder()
        .uri("https://myapi.acme.com/ingest")
        .query("api_key", "secret")
        .call().block()?;

    // More convenience on the http response.
    // Like shortcuts to read or parse
    // response headers.
    let x_req_id =
        response.header_as::<usize>("x-req-id")
        .unwrap();

    // A Body type with easy ways to
    // get the content.
    let mut body = response.into_body();
    let contents = body.read_to_string().block()?;

    assert_eq!(contents, "Hello world!");

    Ok(())
}

User first

User first means that in situations where there are trade offs between ergonomics and performance, or ergonomics and correctness, extra weight will be put towards ergonomics. hreq does not attempt to win any performance or benchmark competitions at the same time as it should not be particularly slow or wasteful of system resources.

http crate

Many rust http client/servers use some variant of the http crate. It's often copied into the local source tree and extended from there.

When writing a service that uses both a web server and client crate, one often ends up with similar, but not exactly the same versions of types like http::Request and http::Response.

hreq works using extension traits only. It re-exports the http crate, but does not copy or modify it. It therefore adheres strictly to the exact API definition as set out by the http crate as well as avoids furthering the confusion of having multiple types with the same name.

Async and blocking

Rust's async story is fantastic, but not every situation requires async. hreq "fakes" being a blocking library by default having a very minimal tokio runtime (rt-core) combined with a .block() call that is placed where we expect an .await in an async situation.

All examples using .block() can be .await

It's anticipated hreq is often used in an async context, however rustdoc doesn't let us document the code that way. Everywhere the doc does .block(), you can switch that out for .await.

use hreq::prelude::*;

let res = Request::get("https://httpbin.org/get")
    .call().block(); // this can be .await in async

Why?

hreq is async through-and-through and ultimately relies on an async variant of TcpStream for it to function. Because the TCP socket is one of those things that is tightly coupled to the async event loop, TcpStream in turn needs to be provided by the runtime (tokio)

There are talks of rust providing a simple single threaded executor as part of the std lib. This only solves half of the problem since TcpStream is coupled with the runtime.

Async runtime

The async runtime is "pluggable" and comes in some different flavors.

  • TokioSingle. The default option. A minimal tokio rt-core which executes calls in one single thread. It does nothing until the current thread blocks on a future using .block().
  • TokioShared. Picks up on a shared runtime by using a Handle. This runtime cannot use the .block() extension trait since that requires having a direct connection to the tokio Runtime.
  • TokioOwned. Uses a preconfigured tokio Runtime that is "handed over" to hreq.

How to configure the options is explained in AsyncRuntime.

Tokio only

This project set out with the ambition to be runtime agnostic, specifically to also support async-std (and/or smol), however in practice that was not a viable route due to it taking too much work to maintain. Rust is likely to eventually provide a pluggable runtime mechanic, in which case this library will try to be agnostic again.

Agent, redirect and retries

All calls in hreq goes through an Agent. The agent provides three main functions:

  • Retries
  • Connection pooling
  • Cookie handling

However the simplest use of hreq creates a new agent for every call, which means connection pooling and cookie handling is only happening to a limited extent (when following redirects).

use hreq::prelude::*;

let res1 = Request::get("https://httpbin.org/get")
    .call().block();  // creates a new agent

// this call doesn't reuse any cookies or connections.
let res2 = Request::get("https://httpbin.org/get")
    .call().block();  // creates another new agent

To use connection pooling and cookies between multiple calls, we need to create an agent.

use hreq::prelude::*;
use hreq::Agent;

let mut agent = Agent::new();

let req1 = Request::get("https://httpbin.org/get")
    .with_body(()).unwrap();

let res1 = agent.send(req1).block();

let req2 = Request::get("https://httpbin.org/get")
    .with_body(()).unwrap();

// this call (tries to) reuse the connection in
// req1 since we are using the same agent.
let res2 = agent.send(req2).block();

Retries

The internet is a dangerous place and http requests fail all the time. hreq tries to be helpful and has a built in retries by default. However it will only retry when appropriate.

  • The default number of retries is 5 with a backoff going 125, 250, 500, 1000 milliseconds.
  • Only for idempotent methods: GET, HEAD, OPTIONS, TRACE, PUT and DELETE.
  • Only when the encountered error is retryable, such as BrokenPipe, ConnectionAborted, ConnectionReset, Interrupted.

To disable retries, one must use a configured agent:

use hreq::prelude::*;
use hreq::Agent;

let mut agent = Agent::new();
agent.retries(0); // disable all retries

let req = Request::get("https://httpbin.org/get")
    .with_body(()).unwrap();

let res = agent.send(req).block();

Redirects

By default hreq follows up to 5 redirects. Redirects can be turned off by using an explicit agent in the same way as for retries.

Compression

hreq supports content compression both for requests and responses. The feature is enabled by receving or setting the content-encoding header to gzip. Currently hreq only supports gzip.

Example request with gzip body:

use hreq::prelude::*;

let res = Request::post("https://my-special-server/content")
  .header("content-encoding", "gzip") // enables gzip compression
  .send("request that is compressed".to_string()).block();

The automatic compression and decompression can be turned off, see content_encode and content_decode.

Charset

Similarly to body compression hreq provides an automatic way of encoding and decoding text in request/response bodies. Rust uses utf-8 for String and assumes text bodies should be encoded as utf-8. Using the content-type we can change how hreq handles both requests and responses.

Example sending an iso-8859-1 encoded body.

use hreq::prelude::*;

// This is a &str in rust default utf-8
let content = "Und in die Bäumen hängen Löwen und Bären";

let req = Request::post("https://my-euro-server/")
    // This header converts the body to iso8859-1
    .header("content-type", "text/plain; charset=iso8859-1")
    .send(content).block();

Receiving bodies of other charset is mostly transparent to the user. It will decode the body to utf-8 if a content-type header is present in the response.

Only content types with a mime type text/* will be decoded.

The charset encoding does not need to work only with utf-8. It can transcode between different encodings as appropriate. See charset_encode_source and charset_decode_target.

Body size

Depending on how a body is provided to a request hreq may or may not be able to know the total body size. For example, when the body provided as a string hreq will set the content-size header, and when the body is a Reader, hreq will not know the content size, but it can be set by the user.

If the content size is not known for HTTP1.1, hreq is forced to use transfer-encoding: chunked. For HTTP2, this problem never arises.

JSON

By default, hreq uses the serde crate to send and receive JSON encoded bodies. Because serde is so ubiquitous in Rust, this feature is enabled by default.

use hreq::Body;
use serde_derive::Serialize;

#[derive(Serialize)]
struct MyJsonThing {
  name: String,
  age: u8,
}

let json = MyJsonThing {
  name: "Karl Kajal".to_string(),
  age: 32,
};

let body = Body::from_json(&json);

Server

hreq started as a client but now also got a simple server mechanism. It can route requests, use middleware, handle state and serve TLS.

See the server module doc for more details.

use hreq::prelude::*;

async fn start_server() {
    let mut server = Server::new();
    server.at("/hello/:name").get(hello_there);
    let (shut, addr) = server.listen(0).await.expect("Failed to listen");
    println!("Listening to: {}", addr);
    shut.shutdown().await;
}

async fn hello_there(req: http::Request<Body>) -> String {
    let name = req.path_param("name").unwrap();
    format!("Hello there {}!\n", name)
}

Capabilities

  • Async or blocking
  • Pure rust
  • HTTP/2 and HTTP/1.1
  • TLS (https)
  • Timeout for entire request and reading the response
  • Single threaded by default
  • Built as an extension to http crate.
  • Query parameter manipulation in request builder
  • Many ways to create a request body
  • Follow redirects
  • Retry on connection problems
  • HTTP/1.1 transfer-encoding chunked
  • Gzip encode/decode
  • Charset encode/decode
  • Connection pooling
  • JSON serialize/deserialize
  • Cookies

License: MIT/Apache-2.0

Dependencies

~13–29MB
~501K SLoC