fix(relay): restore mesh echo and evict stale peers

Co-authored-by: npub1t2tgm7d8f995uqvmnm8h88sg3wnpp9a5xysjf6dg3tjmgt3ltulqdp8ehr <5a968df9a7494b4e019b9ecf739e088ba61097b4312124e9a88ae5b42e3f5f3e@sprout-oss.stage.blox.sqprod.co>
Signed-off-by: npub1t2tgm7d8f995uqvmnm8h88sg3wnpp9a5xysjf6dg3tjmgt3ltulqdp8ehr <5a968df9a7494b4e019b9ecf739e088ba61097b4312124e9a88ae5b42e3f5f3e@sprout-oss.stage.blox.sqprod.co>

crates/buzz-relay-mesh/src/membership.rs

@@ -7,7 +7,7 @@
 use std::collections::HashMap;
 use std::sync::atomic::{AtomicBool, AtomicU64, Ordering};
 use std::sync::{Arc, RwLock};
-use std::time::SystemTime;
+use std::time::{Duration, SystemTime};
 
 use crate::gossip::{system_time_from_millis, GossipRecord, PhiAccrual};
 use crate::registry::ReadyRecord;
@@ -15,6 +15,12 @@ use crate::status::{ConnectionState, MeshCounters, MeshPeerCounters, MeshPeerSta
 use crate::{PeerInfo, RelayMeshMembership, RuntimeId};
 
 pub const DEFAULT_PHI_SUSPECT_THRESHOLD: f64 = 8.0;
+/// Gossip records older than this are no longer useful as routing hints.
+///
+/// This is deliberately much longer than the ready-record TTL (45 seconds at
+/// the default refresh) and the gossip interval (2 seconds), so brief registry
+/// or network stalls do not churn healthy peers.
+pub const DEFAULT_STALE_PEER_TIMEOUT: Duration = Duration::from_secs(5 * 60);
 
 #[derive(Clone, Debug)]
 struct PeerState {
@@ -40,6 +46,7 @@ pub struct MeshMembership {
     /// unanchored state is fail-closed, not accept-any.
     expected_relay_pubkey: Option<String>,
     phi_suspect_threshold: f64,
+    stale_peer_timeout: Duration,
 }
 
 impl MeshMembership {
@@ -53,6 +60,7 @@ impl MeshMembership {
             foreign_relay_rejections: Arc::new(AtomicU64::new(0)),
             expected_relay_pubkey: None,
             phi_suspect_threshold: DEFAULT_PHI_SUSPECT_THRESHOLD,
+            stale_peer_timeout: DEFAULT_STALE_PEER_TIMEOUT,
         }
     }
 
@@ -68,6 +76,14 @@ impl MeshMembership {
         self
     }
 
+    /// Override the maximum heartbeat age retained in membership.
+    /// Primarily useful for deterministic tests; production uses
+    /// [`DEFAULT_STALE_PEER_TIMEOUT`].
+    pub fn with_stale_peer_timeout(mut self, timeout: Duration) -> Self {
+        self.stale_peer_timeout = timeout;
+        self
+    }
+
     pub fn local_record(&self) -> GossipRecord {
         self.local_record
             .read()
@@ -123,12 +139,22 @@ impl MeshMembership {
         }
 
         let heartbeat = system_time_from_millis(record.heartbeat_millis);
+        if heartbeat
+            .elapsed()
+            .is_ok_and(|age| age >= self.stale_peer_timeout)
+        {
+            tracing::debug!(
+                runtime_id = %record.runtime_id,
+                heartbeat_millis = record.heartbeat_millis,
+                "mesh membership ignored stale gossip record"
+            );
+            return false;
+        }
         let mut peers = self.peers.write().expect("membership lock poisoned");
         match peers.get_mut(&record.runtime_id) {
             Some(peer) if record.version <= peer.record.version => false,
             Some(peer) => {
                 peer.record = record;
-                peer.connection_state = ConnectionState::Connected;
                 peer.phi.observe(heartbeat);
                 true
             }
@@ -144,7 +170,7 @@ impl MeshMembership {
                         },
                         record,
                         phi,
-                        connection_state: ConnectionState::Connected,
+                        connection_state: ConnectionState::Disconnected,
                     },
                 );
                 true
@@ -152,6 +178,27 @@ impl MeshMembership {
         }
     }
 
+    /// Remove peers whose last gossiped heartbeat is too old to be useful.
+    /// Returns their runtime ids so the runtime can tear down lingering
+    /// connection state at the same reconcile boundary.
+    pub fn evict_stale_peers(&self) -> Vec<RuntimeId> {
+        let now = SystemTime::now();
+        let mut peers = self.peers.write().expect("membership lock poisoned");
+        let stale: Vec<_> = peers
+            .iter()
+            .filter_map(|(runtime_id, peer)| {
+                let heartbeat = system_time_from_millis(peer.record.heartbeat_millis);
+                now.duration_since(heartbeat)
+                    .is_ok_and(|age| age >= self.stale_peer_timeout)
+                    .then_some(*runtime_id)
+            })
+            .collect();
+        for runtime_id in &stale {
+            peers.remove(runtime_id);
+        }
+        stale
+    }
+
     pub fn mark_connection_state(&self, runtime_id: RuntimeId, state: ConnectionState) {
         if let Some(peer) = self
             .peers
@@ -389,7 +436,7 @@ impl RelayMeshMembership for MeshMembership {
 
 #[cfg(test)]
 mod tests {
-    use std::time::{Duration, UNIX_EPOCH};
+    use std::time::Duration;
 
     use super::*;
 
@@ -406,10 +453,8 @@ mod tests {
             draining: false,
             capabilities: vec!["reliable-stream".to_string()],
             version,
-            heartbeat_millis: (UNIX_EPOCH + Duration::from_secs(heartbeat_secs))
-                .duration_since(UNIX_EPOCH)
-                .unwrap()
-                .as_millis() as u64,
+            heartbeat_millis: crate::gossip::now_millis()
+                .saturating_add(Duration::from_secs(heartbeat_secs).as_millis() as u64),
         }
     }
 
@@ -498,4 +543,53 @@ mod tests {
         assert!(membership.local_record().draining);
         assert_eq!(membership.local_record().version, 2);
     }
+
+    #[test]
+    fn stale_peers_are_evicted_and_cannot_be_resurrected() {
+        let membership =
+            MeshMembership::new(record(1, 1, 0)).with_stale_peer_timeout(Duration::from_secs(5));
+        let mut stale = record(2, 1, 0);
+        stale.heartbeat_millis = crate::gossip::now_millis() - 10_000;
+
+        assert!(!membership.apply_gossip_record(stale.clone()));
+        assert!(membership.peers().is_empty());
+
+        let mut live = record(2, 2, 0);
+        assert!(membership.apply_gossip_record(live.clone()));
+        assert_eq!(membership.status().peer_count, 1);
+
+        live.heartbeat_millis = crate::gossip::now_millis() - 10_000;
+        // Simulate a previously admitted record aging out in place.
+        membership
+            .peers
+            .write()
+            .expect("membership lock poisoned")
+            .get_mut(&rid(2))
+            .expect("peer present")
+            .record
+            .heartbeat_millis = live.heartbeat_millis;
+        assert_eq!(membership.evict_stale_peers(), vec![rid(2)]);
+        assert!(membership.peers().is_empty());
+        assert!(!membership.apply_gossip_record(live));
+    }
+
+    #[test]
+    fn gossip_does_not_claim_transport_connection() {
+        let membership = MeshMembership::new(record(1, 1, 0));
+        assert!(membership.apply_gossip_record(record(2, 1, 0)));
+        assert_eq!(
+            membership.status().peers[0].connection_state,
+            ConnectionState::Disconnected
+        );
+        membership.mark_connection_state(rid(2), ConnectionState::Connected);
+        assert_eq!(
+            membership.status().peers[0].connection_state,
+            ConnectionState::Connected
+        );
+        assert!(membership.apply_gossip_record(record(2, 2, 1)));
+        assert_eq!(
+            membership.status().peers[0].connection_state,
+            ConnectionState::Connected
+        );
+    }
 }

crates/buzz-relay-mesh/src/runtime.rs

@@ -327,6 +327,11 @@ async fn reconcile_loop(inner: Arc<Inner>) {
 }
 
 async fn reconcile_once(inner: &Arc<Inner>) {
+    for runtime_id in inner.membership.evict_stale_peers() {
+        remove_peer(inner, runtime_id);
+        tracing::info!(peer = %runtime_id, "mesh: evicted stale peer");
+    }
+
     if let Some(registry) = &inner.registry {
         match registry.scan_ready().await {
             Ok(records) => inner.membership.apply_ready_records(records),

crates/buzz-relay/src/mesh_boot.rs

@@ -315,25 +315,39 @@ pub(crate) async fn run_demo_echo(
     tracing::info!(%session_id, %peer, "mesh demo echo: session open");
     let mut drain_tick = tokio::time::interval(std::time::Duration::from_millis(100));
     loop {
-        let frame = tokio::select! {
-            _ = drain_tick.tick() => {
-                if shutting_down.load(Ordering::Relaxed) {
-                    if let Some(community_id) = stream.community_id() {
-                        if let Err(e) = stream.send_goodbye(community_id, GoodbyeReason::Draining).await {
-                            tracing::warn!(%session_id, "mesh demo echo: draining goodbye failed: {e}");
-                        } else {
-                            tracing::info!(%session_id, "mesh demo echo: sent draining goodbye");
+        // recv_validated reads the frame before asynchronously checking its
+        // Redis fence. Keep that future alive across drain polls: cancelling
+        // and recreating it after a tick would discard an already-read frame.
+        let frame = {
+            let recv = stream.recv_validated(&directory);
+            tokio::pin!(recv);
+            loop {
+                tokio::select! {
+                    frame = &mut recv => break frame,
+                    _ = drain_tick.tick() => {
+                        if shutting_down.load(Ordering::Relaxed) {
+                            break Ok(None);
                         }
-                    } else {
-                        let _ = stream.finish();
-                        tracing::info!(%session_id, "mesh demo echo: drain before community latch — closing");
                     }
-                    return;
                 }
-                continue;
             }
-            frame = stream.recv_validated(&directory) => frame,
         };
+        if shutting_down.load(Ordering::Relaxed) {
+            if let Some(community_id) = stream.community_id() {
+                if let Err(e) = stream
+                    .send_goodbye(community_id, GoodbyeReason::Draining)
+                    .await
+                {
+                    tracing::warn!(%session_id, "mesh demo echo: draining goodbye failed: {e}");
+                } else {
+                    tracing::info!(%session_id, "mesh demo echo: sent draining goodbye");
+                }
+            } else {
+                let _ = stream.finish();
+                tracing::info!(%session_id, "mesh demo echo: drain before community latch — closing");
+            }
+            return;
+        }
         match frame {
             Ok(Some(ReliableFrame::Data(payload))) => {
                 // recv_validated latched the community from the frame it just