josecelano
commented
1 year ago The tower::builder::ServiceBuilder::timeout it only fail requests that take longer than timeout.
If the next layer takes more than timeout to respond to a request, processing is terminated and an error is returned.
That's not behavior we want. We want the server to send a 408 response when the client opens a new HTTP connection and it does not use it to make any request.
HTTP 408 status code: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/408
ActixWeb implementation
Example using ActixWeb:
- https://github.com/actix/actix-web/blob/1072d0dacf3a5a7a8e2d5fd7fe5bb47658198bf8/actix-http/tests/test_server.rs#L199
- https://github.com/actix/actix-web/blob/1072d0dacf3a5a7a8e2d5fd7fe5bb47658198bf8/actix-web/tests/test_server.rs#L849
The HTTP service builder allows you to set the "client request timeout".
/// Set client request timeout (for first request).
///
/// Defines a timeout for reading client request header. If the client does not transmit the
/// request head within this duration, the connection is terminated with a `408 Request Timeout`
/// response error.
///
/// A duration of zero disables the timeout.
///
/// By default, the client timeout is 5 seconds.
pub fn client_request_timeout(mut self, dur: Duration) -> Self {
self.client_request_timeout = dur;
self
}You can also do it with the server:
/// Sets server client timeout for first request.
///
/// Defines a timeout for reading client request head. If a client does not transmit the entire
/// set headers within this time, the request is terminated with a 408 (Request Timeout) error.
///
/// To disable timeout set value to 0.
///
/// By default client timeout is set to 5000 milliseconds.
pub fn client_request_timeout(self, dur: Duration) -> Self {
self.config.lock().unwrap().client_request_timeout = dur;
self
}Axum implementation
It seems Axum server also has a function to set the read timeout but I haven't been able to make it work:
let app = Router::new().route("/", get(entrypoint_handler)).layer(
ServiceBuilder::new()
.layer(HandleErrorLayer::new(handle_timeout_error))
.layer(TimeoutLayer::new(Duration::from_secs(5)))
.timeout(Duration::from_secs(5)),
);
let server = axum::Server::from_tcp(tcp_listener)
.expect("a new server from the previously created tcp listener.")
.http1_header_read_timeout(Duration::from_secs(5))
.serve(app.into_make_service_with_connect_info::<SocketAddr>());I've created a new repo to try different solutions: https://github.com/josecelano/axum-request-timeout/blob/main/src/app.rs
Axum uses the hyper server. And the hyper server has a method: http1_header_read_timeout
Axum Server implementation
axum-server is a hyper server implementation designed to be used with axum framework.
@programatik29 has proposed a solution for axum-server here.
Hyper
There is an open issue on the hyper repo: timeout while waiting for headers.
It seems "_The http1_header_readtimeout method only starts the timer when the first line of the http1 header is sent."
Client Header Timeout vs Client Body Timeout
I'm not sure if we should use the client_header_timeout or the client_body_timeout.
As if it's described in the this issue, the client_header_timeout "defines a timeout for reading client request header. If a client does not transmit the entire header within this time, the request is terminated with the 408 (Request Time-out) error" while the client_body_timeout "defines a timeout for reading client request body. The timeout is set only for a period between two successive read operations, not for the transmission of the whole request body. If a client does not transmit anything within this time, the request is terminated with the 408 (Request Time-out) error."
Conclusion
It seems there is no official support for this in Axum. We could
- Try to implement our solution like
axum-server. - Migrate to the
axum-server. - Keep the ActixWeb implementation until Axum supports this feature.
- Enable the Axum implementation and let the user deal with this problem for the time being. We do not even support HTTPs yet, so users must use a proxy like Nginx. They can set up the timeouts there. See links.
What do you think @da2ce7?
Links
- Nginx client_body_timeout, https://nginx.org/en/docs/http/ngx_http_core_module.html#client_body_timeout
- Nginx client_header_timeout, https://nginx.org/en/docs/http/ngx_http_core_module.html#client_header_timeout
- Nginx keepalive_timeout, https://nginx.org/en/docs/http/ngx_http_core_module.html#keepalive_timeout
programatik29
da2ce7
jmkng
As described here, we need to set a timeout for requests to avoid a denial of service attack.
How to do it: