Merge pull request 'Fix ping timeout and interval' (#4) from fix-ping into main

Reviewed-on: https://git.deuxfleurs.fr/lx/netapp/pulls/4

1 files changed, 35 insertions, 42 deletions

diff --git a/src/peering/fullmesh.rs b/src/peering/fullmesh.rs
index 012c5a0..208cfe4 100644
--- a/src/peering/fullmesh.rs
+++ b/src/peering/fullmesh.rs
@@ -22,10 +22,10 @@ use crate::NodeID;
 
 const CONN_RETRY_INTERVAL: Duration = Duration::from_secs(30);
 const CONN_MAX_RETRIES: usize = 10;
-const PING_INTERVAL: Duration = Duration::from_secs(10);
+const PING_INTERVAL: Duration = Duration::from_secs(15);
 const LOOP_DELAY: Duration = Duration::from_secs(1);
-const PING_TIMEOUT: Duration = Duration::from_secs(5);
-const FAILED_PING_THRESHOLD: usize = 3;
+const PING_TIMEOUT: Duration = Duration::from_secs(10);
+const FAILED_PING_THRESHOLD: usize = 4;
 
 // -- Protocol messages --
 
@@ -60,11 +60,26 @@ struct PeerInfoInternal {
 	all_addrs: Vec<SocketAddr>,
 
 	state: PeerConnState,
+	last_send_ping: Option<Instant>,
 	last_seen: Option<Instant>,
 	ping: VecDeque<Duration>,
 	failed_pings: usize,
 }
 
+impl PeerInfoInternal {
+	fn new(addr: SocketAddr, state: PeerConnState) -> Self {
+		Self {
+			addr,
+			all_addrs: vec![addr],
+			state,
+			last_send_ping: None,
+			last_seen: None,
+			ping: VecDeque::new(),
+			failed_pings: 0,
+		}
+	}
+}
+
 #[derive(Copy, Clone, Debug)]
 pub struct PeerInfo {
 	/// The node's identifier (its public key)
@@ -184,14 +199,7 @@ impl FullMeshPeeringStrategy {
 			if id != netapp.id {
 				known_hosts.list.insert(
 					id,
-					PeerInfoInternal {
-						addr,
-						all_addrs: vec![addr],
-						state: PeerConnState::Waiting(0, Instant::now()),
-						last_seen: None,
-						ping: VecDeque::new(),
-						failed_pings: 0,
-					},
+					PeerInfoInternal::new(addr, PeerConnState::Waiting(0, Instant::now())),
 				);
 			}
 		}
@@ -199,14 +207,7 @@ impl FullMeshPeeringStrategy {
 		if let Some(addr) = our_addr {
 			known_hosts.list.insert(
 				netapp.id,
-				PeerInfoInternal {
-					addr,
-					all_addrs: vec![addr],
-					state: PeerConnState::Ourself,
-					last_seen: None,
-					ping: VecDeque::new(),
-					failed_pings: 0,
-				},
+				PeerInfoInternal::new(addr, PeerConnState::Ourself),
 			);
 		}
 
@@ -254,7 +255,7 @@ impl FullMeshPeeringStrategy {
 					trace!("{}, {:?}", hex::encode(&id[..8]), info);
 					match info.state {
 						PeerConnState::Connected => {
-							let must_ping = match info.last_seen {
+							let must_ping = match info.last_send_ping {
 								None => true,
 								Some(t) => Instant::now() - t > PING_INTERVAL,
 							};
@@ -274,9 +275,16 @@ impl FullMeshPeeringStrategy {
 			};
 
 			// 2. Dispatch ping to hosts
-			trace!("to_ping: {} peers", to_retry.len());
-			for id in to_ping {
-				tokio::spawn(self.clone().ping(id));
+			trace!("to_ping: {} peers", to_ping.len());
+			if !to_ping.is_empty() {
+				let mut known_hosts = self.known_hosts.write().unwrap();
+				for id in to_ping.iter() {
+					known_hosts.list.get_mut(id).unwrap().last_send_ping = Some(Instant::now());
+				}
+				drop(known_hosts);
+				for id in to_ping {
+					tokio::spawn(self.clone().ping(id));
+				}
 			}
 
 			// 3. Try reconnects
@@ -534,17 +542,9 @@ impl FullMeshPeeringStrategy {
 					host.all_addrs.push(addr);
 				}
 			} else {
-				known_hosts.list.insert(
-					id,
-					PeerInfoInternal {
-						state: PeerConnState::Connected,
-						addr,
-						all_addrs: vec![addr],
-						last_seen: None,
-						ping: VecDeque::new(),
-						failed_pings: 0,
-					},
-				);
+				known_hosts
+					.list
+					.insert(id, PeerInfoInternal::new(addr, PeerConnState::Connected));
 			}
 		}
 		known_hosts.update_hash();
@@ -569,14 +569,7 @@ impl FullMeshPeeringStrategy {
 		} else {
 			PeerConnState::Waiting(0, Instant::now())
 		};
-		PeerInfoInternal {
-			addr,
-			all_addrs: vec![addr],
-			state,
-			last_seen: None,
-			ping: VecDeque::new(),
-			failed_pings: 0,
-		}
+		PeerInfoInternal::new(addr, state)
 	}
 }