merge queue: embarking unstable (675a231) and [#5177 + #5192 + #4870 + #5029] together

beacon_node/http_api/src/lib.rs

@@ -67,7 +67,7 @@ use std::path::PathBuf;
 use std::pin::Pin;
 use std::sync::Arc;
 use sysinfo::{System, SystemExt};
-use system_health::observe_system_health_bn;
+use system_health::{observe_nat, observe_system_health_bn};
 use task_spawner::{Priority, TaskSpawner};
 use tokio::sync::{
     mpsc::{Sender, UnboundedSender},
@@ -4071,13 +4071,7 @@ pub fn serve<T: BeaconChainTypes>(
         .and(warp::path::end())
         .then(|task_spawner: TaskSpawner<T::EthSpec>| {
             task_spawner.blocking_json_task(Priority::P1, move || {
-                Ok(api_types::GenericResponse::from(
-                    lighthouse_network::metrics::NAT_OPEN
-                        .as_ref()
-                        .map(|v| v.get())
-                        .unwrap_or(0)
-                        != 0,
-                ))
+                Ok(api_types::GenericResponse::from(observe_nat()))
             })
         });
 

beacon_node/lighthouse_network/src/config.rs

@@ -350,7 +350,7 @@ impl Default for Config {
             enr_udp6_port: None,
             enr_quic6_port: None,
             enr_tcp6_port: None,
-            target_peers: 50,
+            target_peers: 100,
             gs_config,
             discv5_config,
             boot_nodes_enr: vec![],

beacon_node/lighthouse_network/src/discovery/mod.rs

@@ -59,7 +59,7 @@ const MAX_DISCOVERY_RETRY: usize = 3;
 /// Note: we always allow a single FindPeers query, so we would be
 /// running a maximum of `MAX_CONCURRENT_SUBNET_QUERIES + 1`
 /// discovery queries at a time.
-const MAX_CONCURRENT_SUBNET_QUERIES: usize = 2;
+const MAX_CONCURRENT_SUBNET_QUERIES: usize = 4;
 /// The max number of subnets to search for in a single subnet discovery query.
 const MAX_SUBNETS_IN_QUERY: usize = 3;
 /// The number of closest peers to search for when doing a regular peer search.
@@ -1003,11 +1003,13 @@ impl<TSpec: EthSpec> NetworkBehaviour for Discovery<TSpec> {
                         }
                         discv5::Event::SocketUpdated(socket_addr) => {
                             info!(self.log, "Address updated"; "ip" => %socket_addr.ip(), "udp_port" => %socket_addr.port());
-                            metrics::inc_counter(&metrics::ADDRESS_UPDATE_COUNT);
-                            metrics::check_nat();
+                            // We have SOCKET_UPDATED messages. This occurs when discovery has a majority of
+                            // users reporting an external port and our ENR gets updated.
+                            // Which means we are able to do NAT traversal.
+                            metrics::set_gauge_vec(&metrics::NAT_OPEN, &["discv5"], 1);
+
                             // Discv5 will have updated our local ENR. We save the updated version
                             // to disk.
-
                             if (self.update_ports.tcp4 && socket_addr.is_ipv4())
                                 || (self.update_ports.tcp6 && socket_addr.is_ipv6())
                             {

beacon_node/lighthouse_network/src/metrics.rs

@@ -1,29 +1,17 @@
 pub use lighthouse_metrics::*;
 
 lazy_static! {
-    pub static ref NAT_OPEN: Result<IntCounter> = try_create_int_counter(
+    pub static ref NAT_OPEN: Result<IntGaugeVec> = try_create_int_gauge_vec(
         "nat_open",
-        "An estimate indicating if the local node is exposed to the internet."
+        "An estimate indicating if the local node is reachable from external nodes",
+        &["protocol"]
     );
     pub static ref ADDRESS_UPDATE_COUNT: Result<IntCounter> = try_create_int_counter(
         "libp2p_address_update_total",
         "Count of libp2p socked updated events (when our view of our IP address has changed)"
     );
-    pub static ref PEERS_CONNECTED: Result<IntGauge> = try_create_int_gauge(
-        "libp2p_peers",
-        "Count of libp2p peers currently connected"
-    );
-
-    pub static ref TCP_PEERS_CONNECTED: Result<IntGauge> = try_create_int_gauge(
-        "libp2p_tcp_peers",
-        "Count of libp2p peers currently connected via TCP"
-    );
-
-    pub static ref QUIC_PEERS_CONNECTED: Result<IntGauge> = try_create_int_gauge(
-        "libp2p_quic_peers",
-        "Count of libp2p peers currently connected via QUIC"
-    );
-
+    pub static ref PEERS_CONNECTED: Result<IntGaugeVec> =
+        try_create_int_gauge_vec("libp2p_peers", "Count of libp2p peers currently connected", &["direction", "transport"]);
     pub static ref PEER_CONNECT_EVENT_COUNT: Result<IntCounter> = try_create_int_counter(
         "libp2p_peer_connect_event_total",
         "Count of libp2p peer connect events (not the current number of connected peers)"
@@ -32,13 +20,10 @@ lazy_static! {
         "libp2p_peer_disconnect_event_total",
         "Count of libp2p peer disconnect events"
     );
-    pub static ref DISCOVERY_SENT_BYTES: Result<IntGauge> = try_create_int_gauge(
-        "discovery_sent_bytes",
-        "The number of bytes sent in discovery"
-    );
-    pub static ref DISCOVERY_RECV_BYTES: Result<IntGauge> = try_create_int_gauge(
-        "discovery_recv_bytes",
-        "The number of bytes received in discovery"
+    pub static ref DISCOVERY_BYTES: Result<IntGaugeVec> = try_create_int_gauge_vec(
+        "discovery_bytes",
+        "The number of bytes sent and received in discovery",
+        &["direction"]
     );
     pub static ref DISCOVERY_QUEUE: Result<IntGauge> = try_create_int_gauge(
         "discovery_queue_size",
@@ -135,17 +120,6 @@ lazy_static! {
             &["type"]
         );
 
-    /*
-     * Inbound/Outbound peers
-     */
-    /// The number of peers that dialed us.
-    pub static ref NETWORK_INBOUND_PEERS: Result<IntGauge> =
-        try_create_int_gauge("network_inbound_peers","The number of peers that are currently connected that have dialed us.");
-
-    /// The number of peers that we dialed us.
-    pub static ref NETWORK_OUTBOUND_PEERS: Result<IntGauge> =
-        try_create_int_gauge("network_outbound_peers","The number of peers that are currently connected that we dialed.");
-
     /*
      * Peer Reporting
      */
@@ -156,31 +130,11 @@ lazy_static! {
     );
 }
 
-/// Checks if we consider the NAT open.
-///
-/// Conditions for an open NAT:
-/// 1. We have 1 or more SOCKET_UPDATED messages. This occurs when discovery has a majority of
-///    users reporting an external port and our ENR gets updated.
-/// 2. We have 0 SOCKET_UPDATED messages (can be true if the port was correct on boot), then we
-///    rely on whether we have any inbound messages. If we have no socket update messages, but
-///    manage to get at least one inbound peer, we are exposed correctly.
-pub fn check_nat() {
-    // NAT is already deemed open.
-    if NAT_OPEN.as_ref().map(|v| v.get()).unwrap_or(0) != 0 {
-        return;
-    }
-    if ADDRESS_UPDATE_COUNT.as_ref().map(|v| v.get()).unwrap_or(0) != 0
-        || NETWORK_INBOUND_PEERS.as_ref().map(|v| v.get()).unwrap_or(0) != 0_i64
-    {
-        inc_counter(&NAT_OPEN);
-    }
-}
-
 pub fn scrape_discovery_metrics() {
     let metrics =
         discv5::metrics::Metrics::from(discv5::Discv5::<discv5::DefaultProtocolId>::raw_metrics());
     set_float_gauge(&DISCOVERY_REQS, metrics.unsolicited_requests_per_second);
     set_gauge(&DISCOVERY_SESSIONS, metrics.active_sessions as i64);
-    set_gauge(&DISCOVERY_SENT_BYTES, metrics.bytes_sent as i64);
-    set_gauge(&DISCOVERY_RECV_BYTES, metrics.bytes_recv as i64);
+    set_gauge_vec(&DISCOVERY_BYTES, &["inbound"], metrics.bytes_recv as i64);
+    set_gauge_vec(&DISCOVERY_BYTES, &["outbound"], metrics.bytes_sent as i64);
 }

beacon_node/lighthouse_network/src/peer_manager/mod.rs

@@ -726,29 +726,14 @@ impl<TSpec: EthSpec> PeerManager<TSpec> {
             return;
         }
 
-        let mut connected_peer_count = 0;
-        let mut inbound_connected_peers = 0;
-        let mut outbound_connected_peers = 0;
         let mut clients_per_peer = HashMap::new();
 
         for (_peer, peer_info) in self.network_globals.peers.read().connected_peers() {
-            connected_peer_count += 1;
-            if let PeerConnectionStatus::Connected { n_in, .. } = peer_info.connection_status() {
-                if *n_in > 0 {
-                    inbound_connected_peers += 1;
-                } else {
-                    outbound_connected_peers += 1;
-                }
-            }
             *clients_per_peer
                 .entry(peer_info.client().kind.to_string())
                 .or_default() += 1;
         }
 
-        metrics::set_gauge(&metrics::PEERS_CONNECTED, connected_peer_count);
-        metrics::set_gauge(&metrics::NETWORK_INBOUND_PEERS, inbound_connected_peers);
-        metrics::set_gauge(&metrics::NETWORK_OUTBOUND_PEERS, outbound_connected_peers);
-
         for client_kind in ClientKind::iter() {
             let value = clients_per_peer.get(&client_kind.to_string()).unwrap_or(&0);
             metrics::set_gauge_vec(
@@ -853,11 +838,8 @@ impl<TSpec: EthSpec> PeerManager<TSpec> {
         // start a ping and status timer for the peer
         self.status_peers.insert(*peer_id);
 
-        let connected_peers = self.network_globals.connected_peers() as i64;
-
         // increment prometheus metrics
         metrics::inc_counter(&metrics::PEER_CONNECT_EVENT_COUNT);
-        metrics::set_gauge(&metrics::PEERS_CONNECTED, connected_peers);
 
         true
     }

beacon_node/lighthouse_network/src/peer_manager/network_behaviour.rs

@@ -154,6 +154,8 @@ impl<TSpec: EthSpec> NetworkBehaviour for PeerManager<TSpec> {
                 self.on_dial_failure(peer_id);
             }
             FromSwarm::ExternalAddrConfirmed(_) => {
+                // We have an external address confirmed, means we are able to do NAT traversal.
+                metrics::set_gauge_vec(&metrics::NAT_OPEN, &["libp2p"], 1);
                 // TODO: we likely want to check this against our assumed external tcp
                 // address
             }
@@ -243,25 +245,25 @@ impl<TSpec: EthSpec> PeerManager<TSpec> {
             self.events.push(PeerManagerEvent::MetaData(peer_id));
         }
 
-        // Check NAT if metrics are enabled
-        if self.network_globals.local_enr.read().udp4().is_some() {
-            metrics::check_nat();
-        }
-
         // increment prometheus metrics
         if self.metrics_enabled {
             let remote_addr = endpoint.get_remote_address();
+            let direction = if endpoint.is_dialer() {
+                "outbound"
+            } else {
+                "inbound"
+            };
             match remote_addr.iter().find(|proto| {
                 matches!(
                     proto,
                     multiaddr::Protocol::QuicV1 | multiaddr::Protocol::Tcp(_)
                 )
             }) {
                 Some(multiaddr::Protocol::QuicV1) => {
-                    metrics::inc_gauge(&metrics::QUIC_PEERS_CONNECTED);
+                    metrics::inc_gauge_vec(&metrics::PEERS_CONNECTED, &[direction, "quic"]);
                 }
                 Some(multiaddr::Protocol::Tcp(_)) => {
-                    metrics::inc_gauge(&metrics::TCP_PEERS_CONNECTED);
+                    metrics::inc_gauge_vec(&metrics::PEERS_CONNECTED, &[direction, "tcp"]);
                 }
                 Some(_) => unreachable!(),
                 None => {
@@ -339,17 +341,23 @@ impl<TSpec: EthSpec> PeerManager<TSpec> {
         let remote_addr = endpoint.get_remote_address();
         // Update the prometheus metrics
         if self.metrics_enabled {
+            let direction = if endpoint.is_dialer() {
+                "outbound"
+            } else {
+                "inbound"
+            };
+
             match remote_addr.iter().find(|proto| {
                 matches!(
                     proto,
                     multiaddr::Protocol::QuicV1 | multiaddr::Protocol::Tcp(_)
                 )
             }) {
                 Some(multiaddr::Protocol::QuicV1) => {
-                    metrics::dec_gauge(&metrics::QUIC_PEERS_CONNECTED);
+                    metrics::dec_gauge_vec(&metrics::PEERS_CONNECTED, &[direction, "quic"]);
                 }
                 Some(multiaddr::Protocol::Tcp(_)) => {
-                    metrics::dec_gauge(&metrics::TCP_PEERS_CONNECTED);
+                    metrics::dec_gauge_vec(&metrics::PEERS_CONNECTED, &[direction, "tcp"]);
                 }
                 // If it's an unknown protocol we already logged when connection was established.
                 _ => {}

beacon_node/lighthouse_network/src/service/mod.rs

@@ -51,7 +51,7 @@ mod gossip_cache;
 pub mod gossipsub_scoring_parameters;
 pub mod utils;
 /// The number of peers we target per subnet for discovery queries.
-pub const TARGET_SUBNET_PEERS: usize = 6;
+pub const TARGET_SUBNET_PEERS: usize = 3;
 
 const MAX_IDENTIFY_ADDRESSES: usize = 10;
 

beacon_node/network/src/sync/backfill_sync/mod.rs

@@ -332,7 +332,16 @@ impl<T: BeaconChainTypes> BackFillSync<T> {
                     }
                     // If we have run out of peers in which to retry this batch, the backfill state
                     // transitions to a paused state.
-                    self.retry_batch_download(network, id)?;
+                    // We still need to reset the state for all the affected batches, so we should not
+                    // short circuit early
+                    if self.retry_batch_download(network, id).is_err() {
+                        debug!(
+                            self.log,
+                            "Batch could not be retried";
+                            "batch_id" => id,
+                            "error" => "no synced peers"
+                        );
+                    }
                 } else {
                     debug!(self.log, "Batch not found while removing peer";
                         "peer" => %peer_id, "batch" => id)

beacon_node/src/config.rs

@@ -1122,8 +1122,6 @@ pub fn set_network_config(
         config.target_peers = target_peers_str
             .parse::<usize>()
             .map_err(|_| format!("Invalid number of target peers: {}", target_peers_str))?;
-    } else {
-        config.target_peers = 80; // default value
     }
 
     if let Some(value) = cli_args.value_of("network-load") {