Keep-alive connections

With HTTP/1.1, a client can reuse the same TCP connection for several requests (keep-alive). Corvis always supports it and closes idle connections after keep_alive_timeout, which defaults to 5 seconds, the same value as Node.js.

Changing the timeout

use std::time::Duration;
use corvis::ServerOptions;

let mut app = CorvisFactory::create::<AppModule>().await?;
app.set_server_options(ServerOptions::new().keep_alive_timeout(Duration::from_secs(65)));
app.listen(3000).await

.keep_alive_timeout(None) disables the limit.

Behind a load balancer

If a load balancer or proxy reuses connections to the application, Corvis's idle timeout must be greater than the load balancer's. Otherwise, Corvis may close a connection just as the load balancer sends a request over it, and the client gets a sporadic 502. For example, with a load balancer that closes at 60 seconds, use 65.

What counts as idle

Only the time between requests on the same connection. A request in progress (a slow handler or a Server-Sent Events response) does not count as idle.

The server's other timeouts are independent:

Option	Default	What it limits
headers_timeout	60 s	Time to receive the headers; protects against clients that send them drip by drip
request_timeout	300 s	Time to receive the body; responds 408
keep_alive_timeout	5 s	Inactivity between requests

See HTTP adapters for the rest of ServerOptions.