refresh peers after stall

2 files changed, 19 insertions, 1 deletions

diff --git a/src/swarm/peer.rs b/src/swarm/peer.rs
index e8248bf..1240966 100644
--- a/src/swarm/peer.rs
+++ b/src/swarm/peer.rs
@@ -221,6 +221,19 @@ pub(crate) async fn remember_peer(state: &AppState, peer: &str) {
     }
 }
 
+/// Starts a new handshake with `peer` even though one is complete: its reply resets
+/// sync on both sides and brings its current peer list (`swarm::source`, after giving
+/// it up as the catch-up source).
+pub(crate) async fn rehandshake(state: &AppState, peer: &str) {
+    if let Some(peer_state) = state.get_peer_state(peer).await {
+        let mut peer_state = peer_state.lock().await;
+        if peer_state.hello_state == HelloState::Established {
+            peer_state.hello_state = HelloState::NotSent;
+        }
+    }
+    ensure_handshake(state, peer).await;
+}
+
 pub(crate) async fn send_hello_once(state: &AppState, peer: &str) -> Result<()> {
     let Some(peer_state_mutex) = state.get_peer_state(peer).await else {
         trace!(peer = %peer, "Peer was forgotten; not sending HELLO");
diff --git a/src/swarm/source.rs b/src/swarm/source.rs
index 9eac445..b1701b3 100644
--- a/src/swarm/source.rs
+++ b/src/swarm/source.rs
@@ -23,7 +23,10 @@
 //! `Timing::delivery_limit`, after which Mycelium has delivered or dropped anything it
 //! sent before. It isn't chosen again for another `delivery_limit`, so another peer
 //! takes over rather than the same one stalling again; if it is the only candidate,
-//! catching up waits that long before trying it again. Meanwhile, every minute without a sync message from it is logged as a
+//! catching up waits that long before trying it again. It also gets a fresh HELLO: a
+//! joining node learns of other peers only from HELLO_REPLYs, and the one it got from
+//! this peer may have listed none (a peer lists only its neighbours, which a new member
+//! isn't until its registration has spread). Meanwhile, every minute without a sync message from it is logged as a
 //! warning.
 
 use std::sync::atomic::Ordering;
@@ -203,6 +206,8 @@ pub(crate) fn reply_timed_out(state: &AppState, peer: &str) {
             "Catch-up source doesn't answer our sync messages; catching up from another peer"
         );
         resync_everyone(state);
+        let (state, peer) = (state.clone(), peer.to_owned());
+        tokio::spawn(async move { crate::swarm::peer::rehandshake(&state, &peer).await });
     }
 }