twilight-http-proxy

http-proxy is a ratelimited HTTP proxy in front of the Discord API, making use of twilight.

Use

This proxy is not a "real" HTTP CONNECT TCP proxy, it instead mocks the Discord API.

Using it is as easy as changing the base of all routes from discord.com to localhost:3000, like so:

# Previously
$ curl https://discord.com/api/users/@me
# With the proxy
$ curl http://localhost:3000/api/users/@me

# Or other API versions
$ curl http://localhost:3000/users/@me
$ curl http://localhost:3000/api/users/@me
$ curl http://localhost:3000/api/v8/users/@me

The proxy actively supports routes from API v10, but will also try to request the corresponding routes on older or newer API versions if you so request in the URL.

twilight_http natively supports using twilight_http_proxy, so you can use it like this:

use twilight_http::Client;
use std::error::Error;

#[tokio::main]
async fn main() -> Result<(), Box<dyn Error + Send + Sync>> {
    let client = Client::builder()
        .proxy("localhost:3000", true)
        .ratelimiter(None)
        .build();

    Ok(())
}

This will use the running proxy, skip the ratelimiter (since the proxy does ratelimiting itself), and will request over HTTP.

If you are using a different Discord API client, make sure that you are not ratelimiting outgoing requests, because the proxy will do this instead. Very short HTTP client timeouts may also cause issues with longer ratelimits.

Multiple applications

By default, the proxy will use the token provided in the DISCORD_TOKEN enviroment variable for all requests. You can bypass this by providing a different token in the Authorization header yourself.

The proxy will keep track of ratelimits on a per-token basis, so using multiple applications is as easy as sending the header alongside your requests.

You can configure how long the proxy stores ratelimit information with these enviroment variables:

• CLIENT_DECAY_TIMEOUT (in seconds; defaults to 1 hour) sets the timeout after which ratelimiting information will be dropped due to not being used anymore
• CLIENT_CACHE_MAX_SIZE (defaults to no limit) limits the amount of ratelimiting information in the cache - if full, the least recently used ratelimiting information will be removed
• METRIC_TIMEOUT (in seconds; defaults to 5 minutes) controls how long metrics (metrics are identified by their combination of http method + route + response code + ratelimit scope) will continue to be reported past their last occurence before they are discarded. This avoids polluting your metrics with one off request metrics (9 datapoints per scrape) for long after it happened

Running via Docker

❗ The published images on Docker Hub will not work from April 14, 2023 due to Docker removing free team organizations! Use the new location described below.

Prebuilt Docker images are published on Github's container registry.

$ docker run -itd -e DISCORD_TOKEN="my token" -p 3000:80 ghcr.io/twilight-rs/http-proxy
# Or with metrics enabled
$ docker run -itd -e DISCORD_TOKEN="my token" -p 3000:80 ghcr.io/twilight-rs/http-proxy:metrics

This will set the discord token to "my token" and map the bound port to port 3000 on the host machine.

Images come in two variants: The metrics tag is a build with prometheus metrics support, the latest tag is a build without metrics support. Both support x86_64 as well as aarch64 architectures.

Running via Cargo

Build the binary:

$ cargo build --release
$ DISCORD_TOKEN="my token" PORT=3000 ./target/release/twilight-http-proxy

This will set the discord token to "my token" and bind to port 3000.

Additional configuration

HTTP2 may cause issues with high concurrency (i.e. many concurrent requests). If you encounter frequent error logs related to this, force the use of HTTP1 by setting DISABLE_HTTP2 to any value when running the proxy.

Prometheus metrics

The HTTP proxy can expose prometheus metrics when compiled with the expose-metrics feature. These metrics are then available on the /metrics endpoint. You can set the metrics key used for the histogram data by setting the METRIC_KEY environment variable.

The exported histogram includes timing percentiles, response status codes, request path and request method. Calls to the metrics endpoint itself are not included in the metrics.

Error behaviour

If processing an incoming request fails, the proxy will respond with a 5xx status code and a helpful error message in the response body. Currently, these status codes include:

• 500 if the proxy generates an invalid URI or the ratelimiter fails internally
• 501 if the client requested an unsupported API path or used an unsupported HTTP method
• 502 if the request made by the proxy fails