fix(net): enable QUIC keep-alive to survive backpressure stalls

crates/net/src/endpoint.rs

@@ -6,9 +6,10 @@
 
 use std::net::SocketAddr;
 use std::sync::Arc;
+use std::time::Duration;
 
 use quinn::crypto::rustls::{QuicClientConfig, QuicServerConfig};
-use quinn::{ClientConfig, Endpoint, ServerConfig};
+use quinn::{ClientConfig, Endpoint, IdleTimeout, ServerConfig, TransportConfig, VarInt};
 use rustls::DigitallySignedStruct;
 use rustls::SignatureScheme;
 use rustls::client::danger::{HandshakeSignatureValid, ServerCertVerified, ServerCertVerifier};
@@ -20,6 +21,29 @@ use crate::error::NetError;
 /// The Application-Layer Protocol Negotiation identifier for the Synvael protocol.
 pub const ALPN: &[u8] = b"synvael";
 
+/// Interval between QUIC keep-alive probes, in milliseconds.
+///
+/// Kept well below [`MAX_IDLE_TIMEOUT_MS`] so several probes elapse before the idle timeout could fire. Keep-alives are required because chunk delivery deliberately stalls the stream when the client cannot mesh fast enough: during such a flow-control stall no application data flows in either direction, and without a probe the connection would be indistinguishable from a dead peer and closed on the idle timeout.
+const KEEP_ALIVE_INTERVAL_MS: u32 = 5_000;
+
+/// Maximum time with no received packets before a connection is considered lost, in milliseconds.
+const MAX_IDLE_TIMEOUT_MS: u32 = 30_000;
+
+/// Builds the QUIC transport configuration shared by both endpoints.
+///
+/// Enables keep-alive probes and sets an explicit idle timeout; see [`KEEP_ALIVE_INTERVAL_MS`] for why probes are mandatory given the chunk stream's backpressure behaviour. All other transport parameters retain their `quinn` defaults.
+fn transport_config() -> Arc<TransportConfig> {
+    let mut transport = TransportConfig::default();
+    transport.keep_alive_interval(Some(Duration::from_millis(u64::from(
+        KEEP_ALIVE_INTERVAL_MS,
+    ))));
+    // `VarInt::from_u32` is infallible, so no fallible `IdleTimeout::try_from(Duration)` conversion is needed.
+    transport.max_idle_timeout(Some(IdleTimeout::from(VarInt::from_u32(
+        MAX_IDLE_TIMEOUT_MS,
+    ))));
+    Arc::new(transport)
+}
+
 /// Installs the process-wide default `rustls` `CryptoProvider` if one is not already installed.
 fn ensure_crypto_provider() {
     if rustls::crypto::ring::default_provider()
@@ -49,7 +73,8 @@ pub fn server_endpoint(bind: SocketAddr) -> Result<Endpoint, NetError> {
     tls_config.alpn_protocols = vec![ALPN.to_vec()];
 
     let quic_config = QuicServerConfig::try_from(tls_config)?;
-    let server_config = ServerConfig::with_crypto(Arc::new(quic_config));
+    let mut server_config = ServerConfig::with_crypto(Arc::new(quic_config));
+    server_config.transport_config(transport_config());
 
     Ok(Endpoint::server(server_config, bind)?)
 }
@@ -69,7 +94,8 @@ pub fn client_endpoint() -> Result<Endpoint, NetError> {
     tls_config.alpn_protocols = vec![ALPN.to_vec()];
 
     let quic_config = QuicClientConfig::try_from(tls_config)?;
-    let client_config = ClientConfig::new(Arc::new(quic_config));
+    let mut client_config = ClientConfig::new(Arc::new(quic_config));
+    client_config.transport_config(transport_config());
 
     let mut endpoint = Endpoint::client("0.0.0.0:0".parse().map_err(std::io::Error::other)?)?;
     endpoint.set_default_client_config(client_config);