1 files changed, 86 insertions, 12 deletions

diff --git a/src/swarm/peer/keepalive.rs b/src/swarm/peer/keepalive.rs
index 810c377..460b71e 100644
--- a/src/swarm/peer/keepalive.rs
+++ b/src/swarm/peer/keepalive.rs
@@ -1,12 +1,19 @@
-//! Keepalives (STATUS / STATUS_REPLY) and the self-healing re-handshake when they show our
-//! documents differ while sync is idle.
+//! Keepalives and the self-healing re-handshake.
+//!
+//! A quiet pair's keepalive (`PeerState::take_keepalive_due`) is a PING, answered with a
+//! PONG by the peer's paths layer (`transport::paths`), over the same path: it tells both
+//! sides the other is alive, and gives the round trip (`swarm::rtt`). Every `Timing::status_refresh` it is a
+//! STATUS instead, answered with a STATUS_REPLY: each side's state (heads, readiness,
+//! epoch), which also go out as they change (relay announcements, the epoch after a
+//! switch, readiness: `announce`). When a STATUS shows our documents differ while sync
+//! is idle, the pair re-handshakes (`heal_if_diverged`).
 
 use std::sync::atomic::Ordering;
 use std::time::Duration;
 
 use anyhow::Result;
 use tokio::time::Instant;
-use tracing::warn;
+use tracing::{debug, info, warn};
 
 use super::handshake::ensure_handshake;
 use crate::state::{AppState, Liveness, Standing};
@@ -14,33 +21,100 @@ use crate::swarm::sync::SyncEvent;
 use crate::swarm::topology;
 use crate::transport::{self, ParsedMessage};
 
-/// Sends `peer` a STATUS with our status, noting the time for its round trip.
+/// Sends `peer` a STATUS with our status.
 pub(crate) async fn send_status(state: &AppState, peer: &str) -> Result<()> {
     let status = local_status(state, peer).await;
-    state.lock_rtt().status_sent(peer, Instant::now());
+    if let Some(peer_state) = state.get_peer_state(peer).await {
+        peer_state.lock().await.last_status_at = Some(Instant::now());
+    }
     state
         .transport
         .send_message(peer, &ParsedMessage::Status(status))
         .await
 }
 
+/// Sends `peer` a PING, noting it for its round trip.
+pub(crate) async fn send_ping(state: &AppState, peer: &str) -> Result<()> {
+    let nonce = crate::transport::paths::random_nonce();
+    state.lock_rtt().ping_sent(peer, nonce, Instant::now());
+    state
+        .transport
+        .send_message(peer, &ParsedMessage::Ping(nonce))
+        .await
+}
+
+/// A PING or PONG from `peer`: it is alive, and a PONG may answer our PING (its round
+/// trip). A transport without the paths layer (`Transport::answers_pings`: in tests)
+/// leaves answering a PING to us. Only peers we know already: a PING makes none.
+pub(super) async fn note_alive(state: &AppState, peer: &str, message: &ParsedMessage) {
+    let now = Instant::now();
+    match message {
+        ParsedMessage::Ping(nonce) if !state.transport.answers_pings() => {
+            let pong = ParsedMessage::Pong(*nonce);
+            if let Err(error) = state.transport.send_message(peer, &pong).await {
+                debug!(peer = %peer, error = %format!("{error:#}"), "PONG failed");
+            }
+        }
+        ParsedMessage::Pong(nonce) => state.lock_rtt().pong_received(peer, *nonce, now),
+        _ => {}
+    }
+    let Some(peer_state) = state.get_peer_state(peer).await else {
+        return;
+    };
+    let reactivated = {
+        let mut peer_state = peer_state.lock().await;
+        if peer_state.unreachable_since.is_some() {
+            // Reachable again (`mark_seen` clears it).
+            state.topology_changed.notify_one();
+        }
+        peer_state.mark_seen(now)
+    };
+    crate::swarm::source::heard_from(state, peer);
+    if reactivated {
+        info!(peer = %peer, "Dormant peer is active again; re-handshaking");
+        ensure_handshake(state, peer).await;
+    }
+}
+
 /// How long shutting down waits for the last STATUSes (`farewell`).
 const FAREWELL_TIMEOUT: Duration = Duration::from_secs(2);
 
-/// At shutdown: one last STATUS to each neighbour, saying we aren't ready. A neighbour
-/// catching up from us picks another source now, instead of once we go dormant. Bounded
-/// in time: our transport may be gone already.
-pub(crate) async fn farewell(state: &AppState) {
-    state.stopping.store(true, Ordering::Release);
+/// A STATUS to each neighbour, at most `within` (all sent: none).
+async fn status_to_neighbours(state: &AppState, within: Option<Duration>) {
     let mut statuses = tokio::task::JoinSet::new();
     for (peer, _) in state.peers_snapshot().await {
         if topology::is_neighbour(state, &peer) {
             let state = state.clone();
-            statuses.spawn(async move { send_status(&state, &peer).await });
+            statuses.spawn(async move {
+                if let Err(error) = send_status(&state, &peer).await {
+                    debug!(peer = %peer, error = %format!("{error:#}"), "STATUS failed");
+                }
+            });
         }
     }
     let all_sent = async { while statuses.join_next().await.is_some() {} };
-    let _ = tokio::time::timeout(FAREWELL_TIMEOUT, all_sent).await;
+    match within {
+        Some(within) => {
+            let _ = tokio::time::timeout(within, all_sent).await;
+        }
+        None => all_sent.await,
+    }
+}
+
+/// Our readiness changed (we became ready, or ready again after a rejoin): a STATUS
+/// to each neighbour now, rather than with the next refresh. A peer opens sync towards
+/// us only once it knows we are ready.
+pub(crate) fn announce(state: &AppState) {
+    let state = state.clone();
+    tokio::spawn(async move { status_to_neighbours(&state, None).await });
+}
+
+/// At shutdown: one last STATUS to each neighbour, saying we aren't ready. A neighbour
+/// catching up from us picks another source now, instead of once we go dormant. Bounded
+/// in time: our transport may be gone already.
+pub(crate) async fn farewell(state: &AppState) {
+    state.stopping.store(true, Ordering::Release);
+    status_to_neighbours(state, Some(FAREWELL_TIMEOUT)).await;
 }
 
 /// Our status for `peer`.