Spawn an async test per P2P message to try and resolve latency issues

coordinator/src/main.rs

@@ -330,6 +330,13 @@ pub async fn handle_p2p<D: Db, P: P2p>(
 ) {
   loop {
     let mut msg = p2p.receive().await;
+    // Spawn a dedicated task to handle this message, ensuring any singularly latent message
+    // doesn't hold everything up
+    // TODO2: Move to one task per tributary
+    tokio::spawn({
+      let p2p = p2p.clone();
+      let tributaries = tributaries.clone();
+      async move {
         match msg.kind {
           P2pMessageKind::KeepAlive => {}
 
@@ -337,7 +344,7 @@ pub async fn handle_p2p<D: Db, P: P2p>(
             let tributaries = tributaries.read().await;
             let Some(tributary) = tributaries.get(&genesis) else {
               log::debug!("received p2p message for unknown network");
-          continue;
+              return;
             };
 
             log::trace!("handling message for tributary {:?}", tributary.spec.set());
@@ -350,13 +357,13 @@ pub async fn handle_p2p<D: Db, P: P2p>(
           P2pMessageKind::Heartbeat(genesis) => {
             if msg.msg.len() != 40 {
               log::error!("validator sent invalid heartbeat");
-          continue;
+              return;
             }
 
             let tributaries = tributaries.read().await;
             let Some(tributary) = tributaries.get(&genesis) else {
               log::debug!("received heartbeat message for unknown network");
-          continue;
+              return;
             };
             let tributary_read = tributary.tributary.read().await;
 
@@ -391,7 +398,7 @@ pub async fn handle_p2p<D: Db, P: P2p>(
             }
             if !selected {
               log::debug!("received heartbeat and not selected to respond");
-          continue;
+              return;
             }
 
             log::debug!("received heartbeat and selected to respond");
@@ -414,31 +421,12 @@ pub async fn handle_p2p<D: Db, P: P2p>(
             let mut msg_ref: &[u8] = msg.msg.as_ref();
             let Ok(block) = Block::<Transaction>::read(&mut msg_ref) else {
               log::error!("received block message with an invalidly serialized block");
-          continue;
+              return;
             };
             // Get just the commit
             msg.msg.drain(.. (msg.msg.len() - msg_ref.len()));
             msg.msg.drain((msg.msg.len() - 8) ..);
 
-        // Spawn a dedicated task to add this block, as it may take a notable amount of time
-        // While we could use a long-lived task to add each block, that task would only add one
-        // block at a time *across all tributaries*
-        // We either need:
-        // 1) One task per tributary
-        // 2) Background tasks
-        // 3) For sync_block to return instead of waiting for provided transactions which are
-        //    missing
-        // sync_block waiting is preferable since we know the block is valid by its commit, meaning
-        // we are the node behind
-        // As for 1/2, 1 may be preferable since this message may frequently occur
-        // This is suitably performant, as tokio HTTP servers will even spawn a new task per
-        // connection
-        // In order to reduce congestion though, we should at least check if we take value from
-        // this message before running spawn
-        // TODO2
-        tokio::spawn({
-          let tributaries = tributaries.clone();
-          async move {
             let tributaries = tributaries.read().await;
             let Some(tributary) = tributaries.get(&genesis) else {
               log::debug!("received block message for unknown network");
@@ -448,10 +436,10 @@ pub async fn handle_p2p<D: Db, P: P2p>(
             let res = tributary.tributary.read().await.sync_block(block, msg.msg).await;
             log::debug!("received block from {:?}, sync_block returned {}", msg.sender, res);
           }
+        }
+      }
     });
   }
-    }
-  }
 }
 
 pub async fn publish_transaction<D: Db, P: P2p>(

coordinator/tributary/src/tendermint/mod.rs

@@ -399,6 +399,8 @@ impl<D: Db, T: TransactionTrait, P: P2p> Network for TendermintNetwork<D, T, P>
             hex::encode(hash),
             hex::encode(self.genesis)
           );
+          // TODO: Use a notification system for when we have a new provided, in order to minimize
+          // latency
           sleep(Duration::from_secs(Self::block_time().into())).await;
         }
         _ => return invalid_block(),

coordinator/tributary/tendermint/src/lib.rs

@@ -189,8 +189,7 @@ impl<N: Network + 'static> TendermintMachine<N> {
       // Push it on to the queue. This is done so we only handle one message at a time, and so we
       // can handle our own message before broadcasting it. That way, we fail before before
       // becoming malicious
-      // push_front to prioritize our own messages
+      self.queue.push_back(msg);
-      self.queue.push_front(msg);
     }
   }
 
@@ -220,6 +219,12 @@ impl<N: Network + 'static> TendermintMachine<N> {
     // Sleep until this round ends
     let round_end = self.block.end_time[&end_round];
     let time_until_round_end = round_end.instant().saturating_duration_since(Instant::now());
+    if time_until_round_end == Duration::ZERO {
+      log::trace!(
+        "resetting when prior round ended {}ms ago",
+        Instant::now().saturating_duration_since(round_end.instant()).as_millis(),
+      );
+    }
     log::trace!("sleeping until round ends in {}ms", time_until_round_end.as_millis());
     sleep(time_until_round_end).await;
 
@@ -575,6 +580,13 @@ impl<N: Network + 'static> TendermintMachine<N> {
       Err(TendermintError::Temporal)?;
     }
 
+    if (msg.block == self.block.number) &&
+      (msg.round == self.block.round().number) &&
+      (msg.data.step() == Step::Propose)
+    {
+      log::trace!("received Propose for block {}, round {}", msg.block.0, msg.round.0);
+    }
+
     // If this is a precommit, verify its signature
     self.verify_precommit_signature(signed)?;