From 94a01c20753b1973ab2c016c3c84a21c59478169 Mon Sep 17 00:00:00 2001
From: Quinn Wilton <quinn@quinnwilton.com>
Date: Tue, 23 Jan 2024 14:29:37 -0800
Subject: [PATCH] feat: redial `node_addresses` at an interval on connection
 close

---
 homestar-runtime/config/defaults.toml         |  1 +
 homestar-runtime/src/event_handler.rs         |  2 ++
 homestar-runtime/src/event_handler/cache.rs   |  6 +++++
 homestar-runtime/src/event_handler/event.rs   |  8 ++++++
 .../src/event_handler/swarm_event.rs          | 26 +++++++++++++++++++
 .../src/settings/libp2p_config.rs             |  4 +++
 6 files changed, 47 insertions(+)

diff --git a/homestar-runtime/config/defaults.toml b/homestar-runtime/config/defaults.toml
index 48181388..118cc0c7 100644
--- a/homestar-runtime/config/defaults.toml
+++ b/homestar-runtime/config/defaults.toml
@@ -26,6 +26,7 @@ announce_addresses = []
 transport_connection_timeout = 60
 max_connected_peers = 32
 max_announce_addresses = 10
+dial_interval = 60
 
 [node.network.libp2p.mdns]
 enable = true
diff --git a/homestar-runtime/src/event_handler.rs b/homestar-runtime/src/event_handler.rs
index 14460d51..2e3c7c9b 100644
--- a/homestar-runtime/src/event_handler.rs
+++ b/homestar-runtime/src/event_handler.rs
@@ -148,6 +148,7 @@ struct Rendezvous {
 
 // Connected peers configuration and state
 struct Connections {
+    dial_interval: Duration,
     peers: FnvHashMap<PeerId, ConnectedPoint>,
     max_peers: u32,
 }
@@ -190,6 +191,7 @@ where
             query_senders: FnvHashMap::default(),
             request_response_senders: FnvHashMap::default(),
             connections: Connections {
+                dial_interval: settings.libp2p.dial_interval,
                 peers: FnvHashMap::default(),
                 max_peers: settings.libp2p.max_connected_peers,
             },
diff --git a/homestar-runtime/src/event_handler/cache.rs b/homestar-runtime/src/event_handler/cache.rs
index a4fedb8b..10e5b89a 100644
--- a/homestar-runtime/src/event_handler/cache.rs
+++ b/homestar-runtime/src/event_handler/cache.rs
@@ -51,6 +51,7 @@ pub(crate) enum CacheData {
 pub(crate) enum DispatchEvent {
     RegisterPeer,
     DiscoverPeers,
+    DialPeer,
 }
 
 /// Setup a cache with an eviction listener.
@@ -78,6 +79,11 @@ pub(crate) fn setup_cache(
                         let _ = tx.send(Event::DiscoverPeers(rendezvous_node.to_owned()));
                     };
                 }
+                DispatchEvent::DialPeer => {
+                    if let Some(CacheData::Peer(node)) = val.data.get("node") {
+                        let _ = tx.send(Event::DialPeer(node.to_owned()));
+                    };
+                }
             }
         }
     };
diff --git a/homestar-runtime/src/event_handler/event.rs b/homestar-runtime/src/event_handler/event.rs
index a239b05e..ce095e7e 100644
--- a/homestar-runtime/src/event_handler/event.rs
+++ b/homestar-runtime/src/event_handler/event.rs
@@ -132,6 +132,8 @@ pub(crate) enum Event {
     DiscoverPeers(PeerId),
     /// Dynamically get listeners for the swarm.
     GetNodeInfo(AsyncChannelSender<DynamicNodeInfo>),
+    /// Dial a peer.
+    DialPeer(PeerId),
 }
 
 #[allow(unreachable_patterns)]
@@ -293,6 +295,12 @@ impl Event {
                     );
                 }
             }
+            Event::DialPeer(peer_id) => {
+                event_handler
+                    .swarm
+                    .dial(peer_id)
+                    .map_err(anyhow::Error::new)?;
+            }
             _ => {}
         }
         Ok(())
diff --git a/homestar-runtime/src/event_handler/swarm_event.rs b/homestar-runtime/src/event_handler/swarm_event.rs
index b8a3d0e1..3dc75500 100644
--- a/homestar-runtime/src/event_handler/swarm_event.rs
+++ b/homestar-runtime/src/event_handler/swarm_event.rs
@@ -1157,6 +1157,32 @@ async fn handle_swarm_event<DB: Database>(
                     peer_id = peer_id.to_string(),
                     "removed peer from kademlia table"
                 );
+            } else {
+                debug!(
+                    subject = "libp2p.conn.closed",
+                    category = "handle_swarm_event",
+                    peer_id = peer_id.to_string(),
+                    "redialing trusted peer in {interval:?}",
+                    interval = event_handler.connections.dial_interval
+                );
+
+                // Dial peers again at dial interval
+                event_handler
+                    .cache
+                    .insert(
+                        format!("{}-dial", peer_id),
+                        CacheValue::new(
+                            event_handler.connections.dial_interval,
+                            HashMap::from([
+                                (
+                                    "on_expiration".to_string(),
+                                    CacheData::OnExpiration(cache::DispatchEvent::DialPeer),
+                                ),
+                                ("node".to_string(), CacheData::Peer(peer_id)),
+                            ]),
+                        ),
+                    )
+                    .await;
             }
 
             #[cfg(feature = "websocket-notify")]
diff --git a/homestar-runtime/src/settings/libp2p_config.rs b/homestar-runtime/src/settings/libp2p_config.rs
index 733e5465..d3ced760 100644
--- a/homestar-runtime/src/settings/libp2p_config.rs
+++ b/homestar-runtime/src/settings/libp2p_config.rs
@@ -39,6 +39,9 @@ pub(crate) struct Libp2p {
     /// Transport connection timeout.
     #[serde_as(as = "DurationSeconds<u64>")]
     pub(crate) transport_connection_timeout: Duration,
+    /// Dial interval.
+    #[serde_as(as = "DurationSeconds<u64>")]
+    pub(crate) dial_interval: Duration,
 }
 
 /// DHT settings.
@@ -138,6 +141,7 @@ impl Default for Libp2p {
             pubsub: Pubsub::default(),
             rendezvous: Rendezvous::default(),
             transport_connection_timeout: Duration::new(60, 0),
+            dial_interval: Duration::new(60, 0),
         }
     }
 }