feat: add some more metrics from leanMetrics

crates/blockchain/src/lib.rs

@@ -14,6 +14,7 @@ use tokio::sync::mpsc;
 use tracing::{error, info, warn};
 
 pub mod key_manager;
+mod metrics;
 pub mod store;
 
 /// Messages sent from the blockchain to the P2P layer for publishing.
@@ -90,6 +91,9 @@ impl BlockChainServer {
         let slot = time_since_genesis / SECONDS_PER_SLOT;
         let interval = time_since_genesis % SECONDS_PER_SLOT;
 
+        // Update current slot metric
+        metrics::update_current_slot(slot);
+
         // Produce attestations at interval 1
         if interval == 1 {
             self.produce_attestations(slot);
@@ -99,17 +103,14 @@ impl BlockChainServer {
         let has_proposal = false;
 
         self.store.on_tick(timestamp, has_proposal);
+
+        // Update safe target slot metric (updated by store.on_tick at interval 2)
+        metrics::update_safe_target_slot(self.store.safe_target_slot());
     }
 
     fn produce_attestations(&mut self, slot: u64) {
         // Get the head state to determine number of validators
-        let head_state = match self.store.head_state() {
-            Some(state) => state,
-            None => {
-                warn!(%slot, "Cannot produce attestations: no head state");
-                return;
-            }
-        };
+        let head_state = self.store.head_state();
 
         let num_validators = head_state.validators.len() as u64;
 
@@ -149,14 +150,16 @@ impl BlockChainServer {
             };
 
             // Publish to gossip network
-            if let Err(err) = self
+            let Ok(_) = self
                 .p2p_tx
                 .send(OutboundGossip::PublishAttestation(signed_attestation))
-            {
-                error!(%slot, %validator_id, %err, "Failed to publish attestation");
-            } else {
-                info!(%slot, %validator_id, "Published attestation");
-            }
+                .inspect_err(
+                    |err| error!(%slot, %validator_id, %err, "Failed to publish attestation"),
+                )
+            else {
+                continue;
+            };
+            info!(%slot, %validator_id, "Published attestation");
         }
     }
 
@@ -166,7 +169,10 @@ impl BlockChainServer {
             warn!(%slot, %err, "Failed to process block");
             return;
         }
-        update_head_slot(slot);
+        metrics::update_head_slot(slot);
+        metrics::update_latest_justified_slot(self.store.latest_justified().slot);
+        metrics::update_latest_finalized_slot(self.store.latest_finalized().slot);
+        metrics::update_validators_count(self.store.head_state().validators.len() as u64);
     }
 
     fn on_gossip_attestation(&mut self, attestation: SignedAttestation) {
@@ -228,12 +234,3 @@ impl GenServer for BlockChainServer {
         CastResponse::NoReply
     }
 }
-
-fn update_head_slot(slot: u64) {
-    static LEAN_HEAD_SLOT: std::sync::LazyLock<prometheus::IntGauge> =
-        std::sync::LazyLock::new(|| {
-            prometheus::register_int_gauge!("lean_head_slot", "Latest slot of the lean chain")
-                .unwrap()
-        });
-    LEAN_HEAD_SLOT.set(slot.try_into().unwrap());
-}

crates/blockchain/src/metrics.rs

@@ -0,0 +1,57 @@
+//! Prometheus metrics for the blockchain module.
+
+pub fn update_head_slot(slot: u64) {
+    static LEAN_HEAD_SLOT: std::sync::LazyLock<prometheus::IntGauge> =
+        std::sync::LazyLock::new(|| {
+            prometheus::register_int_gauge!("lean_head_slot", "Latest slot of the lean chain")
+                .unwrap()
+        });
+    LEAN_HEAD_SLOT.set(slot.try_into().unwrap());
+}
+
+pub fn update_latest_justified_slot(slot: u64) {
+    static LEAN_LATEST_JUSTIFIED_SLOT: std::sync::LazyLock<prometheus::IntGauge> =
+        std::sync::LazyLock::new(|| {
+            prometheus::register_int_gauge!("lean_latest_justified_slot", "Latest justified slot")
+                .unwrap()
+        });
+    LEAN_LATEST_JUSTIFIED_SLOT.set(slot.try_into().unwrap());
+}
+
+pub fn update_latest_finalized_slot(slot: u64) {
+    static LEAN_LATEST_FINALIZED_SLOT: std::sync::LazyLock<prometheus::IntGauge> =
+        std::sync::LazyLock::new(|| {
+            prometheus::register_int_gauge!("lean_latest_finalized_slot", "Latest finalized slot")
+                .unwrap()
+        });
+    LEAN_LATEST_FINALIZED_SLOT.set(slot.try_into().unwrap());
+}
+
+pub fn update_current_slot(slot: u64) {
+    static LEAN_CURRENT_SLOT: std::sync::LazyLock<prometheus::IntGauge> =
+        std::sync::LazyLock::new(|| {
+            prometheus::register_int_gauge!("lean_current_slot", "Current slot of the lean chain")
+                .unwrap()
+        });
+    LEAN_CURRENT_SLOT.set(slot.try_into().unwrap());
+}
+
+pub fn update_validators_count(count: u64) {
+    static LEAN_VALIDATORS_COUNT: std::sync::LazyLock<prometheus::IntGauge> =
+        std::sync::LazyLock::new(|| {
+            prometheus::register_int_gauge!(
+                "lean_validators_count",
+                "Number of validators managed by a node"
+            )
+            .unwrap()
+        });
+    LEAN_VALIDATORS_COUNT.set(count.try_into().unwrap());
+}
+
+pub fn update_safe_target_slot(slot: u64) {
+    static LEAN_SAFE_TARGET_SLOT: std::sync::LazyLock<prometheus::IntGauge> =
+        std::sync::LazyLock::new(|| {
+            prometheus::register_int_gauge!("lean_safe_target_slot", "Safe target slot").unwrap()
+        });
+    LEAN_SAFE_TARGET_SLOT.set(slot.try_into().unwrap());
+}

crates/blockchain/src/store.rs

@@ -676,8 +676,15 @@ impl Store {
     }
 
     /// Returns a reference to the head state if it exists.
-    pub fn head_state(&self) -> Option<&State> {
-        self.states.get(&self.head)
+    pub fn head_state(&self) -> &State {
+        self.states
+            .get(&self.head)
+            .expect("head state is always available")
+    }
+
+    /// Returns the slot of the current safe target block.
+    pub fn safe_target_slot(&self) -> u64 {
+        self.blocks[&self.safe_target].slot
     }
 }
 

docs/metrics.md

@@ -0,0 +1,68 @@
+# Metrics
+
+We collect various metrics and serve them via a Prometheus-compatible HTTP endpoint at `http://<metrics_address>:<metrics_port>/metrics` (default: `http://127.0.0.1:5054/metrics`).
+
+The exposed metrics follow [the leanMetrics specification](https://github.com/leanEthereum/leanMetrics/blob/3b32b300cca5ed7a7a2b3f142273fae9dbc171bf/metrics.md), with some metrics not yet implemented. We have a full list of implemented metrics below, with a checkbox indicating whether each metric is currently supported or not.
+
+## Node Info Metrics
+
+| Name   | Type  | Usage | Sample collection event | Labels | Supported     |
+|--------|-------|-------|-------------------------|--------|---------------|
+| `lean_node_info` | Gauge | Node information (always 1) | On node start | name, version | □ |
+| `lean_node_start_time_seconds` | Gauge | Start timestamp | On node start | | □ |
+
+
+## PQ Signature Metrics
+
+| Name   | Type  | Usage | Sample collection event | Labels | Buckets | Supported |
+|--------|-------|-------|-------------------------|--------|---------|-----------|
+| `lean_pq_sig_attestation_signing_time_seconds` | Histogram | Time taken to sign an attestation | On each attestation signing | | 0.005, 0.01, 0.025, 0.05, 0.1, 1 | □ |
+| `lean_pq_sig_attestation_verification_time_seconds` | Histogram | Time taken to verify an attestation signature | On each `signature.verify()` on an attestation | | 0.005, 0.01, 0.025, 0.05, 0.1, 1 | □ |
+| `lean_pq_sig_aggregated_signatures_total` | Counter | Total number of aggregated signatures | On `build_attestation_signatures()` | | 0.005, 0.01, 0.025, 0.05, 0.1, 1 | □ |
+| `lean_pq_sig_attestations_in_aggregated_signatures_total` | Counter | Total number of attestations included into aggregated signatures | On `build_attestation_signatures()` | | 0.005, 0.01, 0.025, 0.05, 0.1, 1 | □ |
+| `lean_pq_sig_attestation_signatures_building_time_seconds` | Histogram | Time taken to verify an aggregated attestation signature | On `build_attestation_signatures()` | | 0.005, 0.01, 0.025, 0.05, 0.1, 1 | □ |
+| `lean_pq_sig_aggregated_signatures_verification_time_seconds` | Histogram | Time taken to verify an aggregated attestation signature | On validate aggregated signature | | 0.005, 0.01, 0.025, 0.05, 0.1, 1 | □ |
+| `lean_pq_sig_aggregated_signatures_valid_total`| Counter | Total number of valid aggregated signatures | On validate aggregated signature | | | □ |
+| `lean_pq_sig_aggregated_signatures_invalid_total`| Counter | Total number of invalid aggregated signatures | On validate aggregated signature | | | □ |
+
+## Fork-Choice Metrics
+
+| Name   | Type  | Usage | Sample collection event | Labels | Buckets | Supported |
+|--------|-------|-------|-------------------------|--------|---------|-----------|
+| `lean_head_slot` | Gauge | Latest slot of the lean chain | On get fork choice head | | | ✅ |
+| `lean_current_slot` | Gauge | Current slot of the lean chain | On scrape | | | ✅ |
+| `lean_safe_target_slot` | Gauge | Safe target slot | On safe target update | | | ✅ |
+|`lean_fork_choice_block_processing_time_seconds`| Histogram | Time taken to process block | On fork choice process block | | 0.005, 0.01, 0.025, 0.05, 0.1, 1 | □ |
+|`lean_attestations_valid_total`| Counter | Total number of valid attestations | On validate attestation | source=block,gossip | | □ |
+|`lean_attestations_invalid_total`| Counter | Total number of invalid attestations | On validate attestation | source=block,gossip | | □ |
+|`lean_attestation_validation_time_seconds`| Histogram | Time taken to validate attestation | On validate attestation | | 0.005, 0.01, 0.025, 0.05, 0.1, 1 | □ |
+| `lean_fork_choice_reorgs_total` | Counter | Total number of fork choice reorgs | On fork choice reorg | | | □ |
+| `lean_fork_choice_reorg_depth` | Histogram | Depth of fork choice reorgs (in blocks) | On fork choice reorg | | 1, 2, 3, 5, 7, 10, 20, 30, 50, 100 | □ |
+
+## State Transition Metrics
+
+| Name   | Type  | Usage | Sample collection event | Labels | Buckets | Supported |
+|--------|-------|-------|-------------------------|--------|---------|-----------|
+| `lean_latest_justified_slot` | Gauge | Latest justified slot | On state transition | | | ✅ |
+| `lean_latest_finalized_slot` | Gauge | Latest finalized slot | On state transition | | | ✅ |
+| `lean_finalizations_total` | Counter | Total number of finalization attempts | On finalization attempt | result=success,error | | □ |
+|`lean_state_transition_time_seconds`| Histogram | Time to process state transition | On state transition | | 0.25, 0.5, 0.75, 1, 1.25, 1.5, 2, 2.5, 3, 4 | □ |
+|`lean_state_transition_slots_processed_total`| Counter | Total number of processed slots | On state transition process slots | | | □ |
+|`lean_state_transition_slots_processing_time_seconds`| Histogram | Time taken to process slots | On state transition process slots | | 0.005, 0.01, 0.025, 0.05, 0.1, 1 | □ |
+|`lean_state_transition_block_processing_time_seconds`| Histogram | Time taken to process block | On state transition process block | | 0.005, 0.01, 0.025, 0.05, 0.1, 1 | □ |
+|`lean_state_transition_attestations_processed_total`| Counter | Total number of processed attestations | On state transition process attestations | | | □ |
+|`lean_state_transition_attestations_processing_time_seconds`| Histogram | Time taken to process attestations | On state transition process attestations | | 0.005, 0.01, 0.025, 0.05, 0.1, 1 | □ |
+
+## Validator Metrics
+
+| Name   | Type  | Usage | Sample collection event | Labels | Supported |
+|--------|-------|-------|-------------------------|--------|-----------|
+|`lean_validators_count`| Gauge | Number of validators managed by a node | On scrape |  | ✅ |
+
+## Network Metrics
+
+| Name   | Type  | Usage | Sample collection event | Labels | Supported |
+|--------|-------|-------|-------------------------|--------|-----------|
+|`lean_connected_peers`| Gauge | Number of connected peers | On scrape | client=lantern,qlean,ream,zeam | □ |
+|`lean_peer_connection_events_total`| Counter | Total number of peer connection events | On peer connection | direction=inbound,outbound<br>result=success,timeout,error | □ |
+|`lean_peer_disconnection_events_total`| Counter | Total number of peer disconnection events | On peer disconnection | direction=inbound,outbound<br>reason=timeout,remote_close,local_close,error | □ |