Add some metrics for random beacon (MystenLabs#15450)

narwhal/.clippy.toml

@@ -3,7 +3,7 @@ cognitive-complexity-threshold = 100
 # types are used for safety encoding
 type-complexity-threshold = 10000
 # big constructors
-too-many-arguments-threshold =  15
+too-many-arguments-threshold =  20
 
 disallowed-methods = [
     {path = "anyhow", reason = "we prefer to use eyre"},

narwhal/primary/src/metrics.rs

@@ -360,6 +360,10 @@ pub struct PrimaryMetrics {
     pub fetched_certificates_verified_directly: IntCounter,
     // Total number of fetched certificates verified indirectly.
     pub fetched_certificates_verified_indirectly: IntCounter,
+    /// The number of shares held by this node after the random beacon DKG protocol completed.
+    pub state_handler_random_beacon_dkg_num_shares: IntGauge,
+    /// The randomness round that currently in progress.
+    pub state_handler_current_randomness_round: IntGauge,
 }
 
 impl PrimaryMetrics {
@@ -555,6 +559,16 @@ impl PrimaryMetrics {
                 "Total number of fetched certificates verified indirectly.",
                 registry
             ).unwrap(),
+            state_handler_random_beacon_dkg_num_shares: register_int_gauge_with_registry!(
+                "state_handler_random_beacon_dkg_num_shares",
+                "The number of shares held by this node after the random beacon DKG protocol completed.",
+                registry
+            ).unwrap(),
+            state_handler_current_randomness_round: register_int_gauge_with_registry!(
+                "state_handler_current_randomness_round",
+                "The randomness round that currently in progress.",
+                registry
+            ).unwrap(),
         }
     }
 }

narwhal/primary/src/primary.rs

@@ -515,7 +515,7 @@ impl Primary {
             tx_headers,
             tx_narwhal_round_updates,
             rx_committed_own_headers,
-            node_metrics,
+            node_metrics.clone(),
             leader_schedule.clone(),
         );
 
@@ -544,6 +544,7 @@ impl Primary {
             leader_schedule,
             network,
             randomness_store,
+            node_metrics,
         );
         handles.push(state_handler_handle);
 

narwhal/primary/src/state_handler.rs

@@ -2,6 +2,7 @@
 // Copyright (c) Mysten Labs, Inc.
 // SPDX-License-Identifier: Apache-2.0
 use crate::consensus::LeaderSchedule;
+use crate::metrics::PrimaryMetrics;
 use config::{AuthorityIdentifier, ChainIdentifier, Committee};
 use crypto::{RandomnessPartialSignature, RandomnessPrivateKey};
 use fastcrypto::encoding::{Encoding, Hex};
@@ -82,6 +83,7 @@ pub struct StateHandler {
 //    randomness round is incremented.
 struct RandomnessState {
     store: RandomnessStore,
+    metrics: Arc<PrimaryMetrics>,
 
     // A channel to send system messages to the proposer.
     tx_system_messages: Sender<SystemMessage>,
@@ -124,6 +126,7 @@ impl RandomnessState {
         vss_key_output: Arc<OnceLock<PublicVssKey>>,
         tx_system_messages: Sender<SystemMessage>,
         store: RandomnessStore,
+        metrics: Arc<PrimaryMetrics>,
     ) -> Option<Self> {
         if !protocol_config.random_beacon() {
             info!("random beacon: disabled");
@@ -184,9 +187,20 @@ impl RandomnessState {
             "random beacon: state initialized with authority_id={authority_id}, total_weight={total_weight}, t={t}, num_nodes={num_nodes}, oracle initial_prefix={prefix_str:?}",
         );
 
+        // Load existing data from store.
         let dkg_output = store.dkg_output();
+        if let Some(dkg_output) = &dkg_output {
+            metrics
+                .state_handler_random_beacon_dkg_num_shares
+                .set(dkg_output.shares.as_ref().map_or(0, |shares| shares.len()) as i64);
+        }
+        metrics
+            .state_handler_current_randomness_round
+            .set(store.randomness_round().0 as i64);
+
         Some(Self {
             store,
+            metrics,
             tx_system_messages,
             party,
             has_sent_confirmation: false,
@@ -205,6 +219,9 @@ impl RandomnessState {
     }
 
     fn set_dkg_output(&mut self, output: dkg::Output<PkG, EncG>) {
+        self.metrics
+            .state_handler_random_beacon_dkg_num_shares
+            .set(output.shares.as_ref().map_or(0, |shares| shares.len()) as i64);
         self.store.set_dkg_output(&output);
         self.dkg_output = Some(output);
     }
@@ -320,6 +337,9 @@ impl RandomnessState {
             return;
         }
         debug!("random beacon: updating local randomness round to {round:?}");
+        self.metrics
+            .state_handler_current_randomness_round
+            .set(round.0 as i64);
         self.store.set_randomness_round(round);
         self.partial_sigs.retain(|&(r, _), _| r >= round);
         self.send_partial_signatures().await;
@@ -525,6 +545,7 @@ impl StateHandler {
         leader_schedule: LeaderSchedule,
         network: anemo::Network,
         randomness_store: RandomnessStore,
+        metrics: Arc<PrimaryMetrics>,
     ) -> JoinHandle<()> {
         let state_handler = Self {
             authority_id,
@@ -544,6 +565,7 @@ impl StateHandler {
                 vss_key_output,
                 tx_system_messages,
                 randomness_store,
+                metrics,
             ),
             network,
         };

narwhal/primary/src/tests/state_handler_tests.rs

@@ -8,6 +8,7 @@ use fastcrypto::{
     traits::{KeyPair, ToFromBytes},
 };
 
+use prometheus::Registry;
 use test_utils::CommitteeFixture;
 
 #[tokio::test]
@@ -30,6 +31,7 @@ async fn dkg() {
     for authority in fixture.authorities() {
         let network = test_utils::test_network(authority.network_keypair(), authority.address());
         let randomness_store = RandomnessStore::new_for_tests();
+        let metrics = Arc::new(PrimaryMetrics::new(&Registry::new()));
         let vss_key_output = Arc::new(OnceLock::new());
 
         let (tx_system_messages, rx_system_messages) = test_utils::test_channel!(1);
@@ -57,6 +59,7 @@ async fn dkg() {
             vss_key_output.clone(),
             tx_system_messages,
             randomness_store,
+            metrics,
         )
         .unwrap();
 
@@ -139,6 +142,7 @@ async fn resume_from_store() {
         let randomness_store = RandomnessStore::new_for_tests();
         randomness_stores.push(randomness_store.clone());
         let vss_key_output = Arc::new(OnceLock::new());
+        let metrics = Arc::new(PrimaryMetrics::new(&Registry::new()));
 
         let (tx_system_messages, rx_system_messages) = test_utils::test_channel!(2);
 
@@ -165,6 +169,7 @@ async fn resume_from_store() {
             vss_key_output.clone(),
             tx_system_messages,
             randomness_store,
+            metrics,
         )
         .unwrap();
 
@@ -218,6 +223,7 @@ async fn resume_from_store() {
     for (i, authority) in fixture.authorities().enumerate() {
         let network = test_utils::test_network(authority.network_keypair(), authority.address());
         let (tx_system_messages, rx_system_messages) = test_utils::test_channel!(2);
+        let metrics = Arc::new(PrimaryMetrics::new(&Registry::new()));
 
         let randomness_private_key = RandomnessPrivateKey::from(
             fastcrypto::groups::bls12381::Scalar::from_byte_array(
@@ -242,6 +248,7 @@ async fn resume_from_store() {
             vss_key_outputs[i].clone(),
             tx_system_messages,
             randomness_stores[i].clone(),
+            metrics,
         )
         .unwrap();
         // called by StateHandler::run on restart