[State Sync] Small config tweaks and metrics improvements.

Author: JoshLind

config/src/config/network_config.rs

@@ -39,7 +39,7 @@ pub const PING_FAILURES_TOLERATED: u64 = 5;
 pub const CONNECTIVITY_CHECK_INTERVAL_MS: u64 = 5000;
 pub const MAX_CONCURRENT_NETWORK_REQS: usize = 100;
 pub const MAX_CONNECTION_DELAY_MS: u64 = 60_000; /* 1 minute */
-pub const MAX_FULLNODE_OUTBOUND_CONNECTIONS: usize = 5;
+pub const MAX_FULLNODE_OUTBOUND_CONNECTIONS: usize = 4;
 pub const MAX_INBOUND_CONNECTIONS: usize = 100;
 pub const MAX_FRAME_SIZE: usize = 16 * 1024 * 1024; /* 16 MiB */
 pub const CONNECTION_BACKOFF_BASE: u64 = 2;

config/src/config/state_sync_config.rs

@@ -186,7 +186,7 @@ impl Default for DataStreamingServiceConfig {
     fn default() -> Self {
         Self {
             global_summary_refresh_interval_ms: 50,
-            max_concurrent_requests: 3,
+            max_concurrent_requests: 2,
             max_data_stream_channel_sizes: 1000,
             max_request_retry: 3,
             max_notification_id_mappings: 2000,

state-sync/aptos-data-client/src/aptosnet/metrics.rs

@@ -64,6 +64,16 @@ pub static IN_FLIGHT_POLLS: Lazy<IntGaugeVec> = Lazy::new(|| {
     .unwrap()
 });
 
+/// Gauge for tracking the number of connected peers (priority and regular)
+pub static CONNECTED_PEERS: Lazy<IntGaugeVec> = Lazy::new(|| {
+    register_int_gauge_vec!(
+        "aptos_data_client_connected_peers",
+        "Gauge related to the number of connected peers",
+        &["peer_type"]
+    )
+    .unwrap()
+});
+
 /// Gauge for the highest advertised data
 pub static HIGHEST_ADVERTISED_DATA: Lazy<IntGaugeVec> = Lazy::new(|| {
     register_int_gauge_vec!(
@@ -140,8 +150,8 @@ pub fn increment_request_counter(
 }
 
 /// Sets the gauge with the specific label and value
-pub fn set_gauge(counter: &Lazy<IntGaugeVec>, label: String, value: u64) {
-    counter.with_label_values(&[&label]).set(value as i64);
+pub fn set_gauge(counter: &Lazy<IntGaugeVec>, label: &str, value: u64) {
+    counter.with_label_values(&[label]).set(value as i64);
 }
 
 /// Starts the timer for the provided histogram and label values.

state-sync/aptos-data-client/src/aptosnet/mod.rs

@@ -281,14 +281,7 @@ impl AptosNetDataClient {
         // Log the peers, periodically.
         sample!(
             SampleRate::Duration(Duration::from_secs(PEER_LOG_FREQ_SECS)),
-            info!(
-                (LogSchema::new(LogEntry::PeerStates)
-                    .event(LogEvent::PriorityAndRegularPeers)
-                    .message(&format!(
-                        "Number of priority peers: {:?}. Number of regular peers: {:?}",
-                        priority_peers.len(), regular_peers.len(),
-                    )))
-            );
+            update_connected_peer_metrics(priority_peers.len(), regular_peers.len());
         );
 
         Ok((priority_peers, regular_peers))
@@ -701,12 +694,37 @@ fn update_in_flight_metrics(label: &str, num_in_flight_polls: u64) {
         SampleRate::Frequency(IN_FLIGHT_METRICS_SAMPLE_FREQ),
         set_gauge(
             &metrics::IN_FLIGHT_POLLS,
-            label.into(),
+            label,
             num_in_flight_polls,
         );
     );
 }
 
+/// Updates the metrics for the number of connected peers (priority and regular)
+fn update_connected_peer_metrics(num_priority_peers: usize, num_regular_peers: usize) {
+    // Log the number of connected peers
+    info!(
+        (LogSchema::new(LogEntry::PeerStates)
+            .event(LogEvent::PriorityAndRegularPeers)
+            .message(&format!(
+                "Number of priority peers: {:?}. Number of regular peers: {:?}",
+                num_priority_peers, num_regular_peers,
+            )))
+    );
+
+    // Update the connected peer metrics
+    set_gauge(
+        &metrics::CONNECTED_PEERS,
+        PRIORITIZED_PEER,
+        num_priority_peers as u64,
+    );
+    set_gauge(
+        &metrics::CONNECTED_PEERS,
+        REGULAR_PEER,
+        num_regular_peers as u64,
+    );
+}
+
 /// Spawns a dedicated poller for the given peer.
 pub(crate) fn poll_peer(
     data_client: AptosNetDataClient,
@@ -793,9 +811,9 @@ fn update_advertised_data_metrics(global_data_summary: GlobalDataSummary) {
             DataType::TransactionOutputs => optimal_chunk_sizes.transaction_output_chunk_size,
             DataType::Transactions => optimal_chunk_sizes.transaction_chunk_size,
         };
-        metrics::set_gauge(
+        set_gauge(
             &metrics::OPTIMAL_CHUNK_SIZES,
-            data_type.as_str().into(),
+            data_type.as_str(),
             optimal_chunk_size,
         );
     }
@@ -807,9 +825,9 @@ fn update_advertised_data_metrics(global_data_summary: GlobalDataSummary) {
         .map(|ledger_info| ledger_info.ledger_info().version());
     if let Some(highest_advertised_version) = highest_advertised_version {
         for data_type in DataType::get_all_types() {
-            metrics::set_gauge(
+            set_gauge(
                 &metrics::HIGHEST_ADVERTISED_DATA,
-                data_type.as_str().into(),
+                data_type.as_str(),
                 highest_advertised_version,
             );
         }
@@ -824,9 +842,9 @@ fn update_advertised_data_metrics(global_data_summary: GlobalDataSummary) {
             DataType::Transactions => advertised_data.lowest_transaction_version(),
         };
         if let Some(lowest_advertised_version) = lowest_advertised_version {
-            metrics::set_gauge(
+            set_gauge(
                 &metrics::LOWEST_ADVERTISED_DATA,
-                data_type.as_str().into(),
+                data_type.as_str(),
                 lowest_advertised_version,
             );
         }