add metrics to gossip

kbirdg · May 27, 2022 · 2183b44 · 2183b44
1 parent bdf67e4
commit 2183b44
Show file tree

Hide file tree

Showing 2 changed files with 33 additions and 1 deletion.
diff --git a/crates/sui-core/src/authority.rs b/crates/sui-core/src/authority.rs
@@ -101,6 +101,11 @@ pub struct AuthorityMetrics {
     num_input_objs: Histogram,
     num_shared_objects: Histogram,
     batch_size: Histogram,
+
+    pub gossip_queued_count: IntCounter,
+    pub gossip_sync_count: IntCounter,
+    pub gossip_task_success_count: IntCounter,
+    pub gossip_task_error_count: IntCounter,
 }
 
 // Override default Prom buckets for positive numbers in 0-50k range
@@ -162,6 +167,26 @@ impl AuthorityMetrics {
                 POSITIVE_INT_BUCKETS.to_vec()
             )
             .unwrap(),
+            gossip_queued_count: register_int_counter!(
+                "gossip_queued_count",
+                "Number of digests queued from gossip peers",
+            )
+            .unwrap(),
+            gossip_sync_count: register_int_counter!(
+                "gossip_sync_count",
+                "Number of certificates downloaded from gossip peers"
+            )
+            .unwrap(),
+            gossip_task_success_count: register_int_counter!(
+                "gossip_task_success_count",
+                "Number of gossip tasks that completed successfully"
+            )
+            .unwrap(),
+            gossip_task_error_count: register_int_counter!(
+                "gossip_task_error_count",
+                "Number of gossip tasks that completed with errors"
+            )
+            .unwrap(),
         }
     }
 }

diff --git a/crates/sui-core/src/authority_active/gossip/mod.rs b/crates/sui-core/src/authority_active/gossip/mod.rs
@@ -123,9 +123,15 @@ where
         let (finished_name, _result) = gossip_tasks.select_next_some().await;
         if let Err(err) = _result {
             active_authority.set_failure_backoff(finished_name).await;
+            active_authority.state.metrics.gossip_task_error_count.inc();
             error!("Peer {:?} returned error: {:?}", finished_name, err);
         } else {
             active_authority.set_success_backoff(finished_name).await;
+            active_authority
+                .state
+                .metrics
+                .gossip_task_success_count
+                .inc();
             debug!("End gossip from peer {:?}", finished_name);
         }
         peer_names.remove(&finished_name);
@@ -194,7 +200,6 @@ where
         let result =
             tokio::task::spawn(async move { self.peer_gossip_for_duration(duration).await }).await;
 
-        // todo: log e
         match result {
             Err(_e) => (
                 peer_name,
@@ -236,6 +241,7 @@ where
                                     tokio::time::sleep(Duration::from_millis(EACH_ITEM_DELAY_MS)).await;
                                     digest
                                 });
+                                self.state.metrics.gossip_queued_count.inc();
 
                             }
                             self.max_seq = Some(seq + 1);
@@ -283,6 +289,7 @@ where
                     },
                 )
                 .await?;
+            self.state.metrics.gossip_sync_count.inc();
             Ok(())
         } else {
             // The authority did not return the certificate, despite returning info