[fix] #4083: Introduce predictable ordering after too many failed vie…

…w changes

Signed-off-by: Marin Veršić <[email protected]>

core/src/sumeragi/main_loop.rs

@@ -606,6 +606,7 @@ fn process_message_independent(
                 if cache_full || (deadline_reached && cache_non_empty) {
                     let transactions = sumeragi.transaction_cache.clone();
                     info!(%addr, txns=%transactions.len(), "Creating block...");
+                    let create_block_start_time = Instant::now();
 
                     // TODO: properly process triggers!
                     let mut new_wsv = sumeragi.wsv.clone();
@@ -625,10 +626,15 @@ fn process_message_independent(
                         }
                     };
 
+                    let created_in = create_block_start_time.elapsed();
                     if let Some(current_topology) = current_topology.is_consensus_required() {
-                        info!(%addr, block_payload_hash=%new_block.payload().hash(), "Block created");
-                        *voting_block = Some(VotingBlock::new(new_block.clone(), new_wsv));
+                        info!(%addr, created_in_ms=%created_in.as_millis(), block_payload_hash=%new_block.payload().hash(), "Block created");
+
+                        if created_in > sumeragi.pipeline_time() / 2 {
+                            warn!("Creating block takes too much time. This might prevent consensus from operating. Consider increasing `commit_time` or decreasing `max_transactions_in_block`");
+                        }
 
+                        *voting_block = Some(VotingBlock::new(new_block.clone(), new_wsv));
                         let msg = MessagePacket::new(
                             view_change_proof_chain.clone(),
                             Some(BlockCreated::from(new_block).into()),

core/src/sumeragi/network_topology.rs

@@ -194,6 +194,26 @@ impl Topology {
         // Rotate all once for every view_change
         topology.rotate_all_n(view_change_index);
 
+        {
+            // FIXME: This is a hack to prevent consensus from running amock due to
+            // a bug in the implementation by reverting to predictable ordering
+
+            let view_change_limit: usize = view_change_index
+                .saturating_sub(10)
+                .try_into()
+                .expect("u64 must fit into usize");
+
+            if view_change_limit > 1 {
+                iroha_logger::error!("Restarting consensus(internal bug). Report to developers");
+                let mut peers: Vec<_> = topology.ordered_peers.iter().cloned().collect();
+
+                peers.sort();
+                let peers_count = peers.len();
+                peers.rotate_right(view_change_limit % peers_count);
+                topology = Topology::new(peers.into_iter().collect());
+            }
+        }
+
         topology
     }