enable span tracing latency via prometheus

crates/sui-core/src/node_sync/node_follower.rs

@@ -309,10 +309,8 @@ where
 mod test {
     use super::*;
     use crate::{
-        authority_active::gossip::GossipMetrics,
-        authority_client::NetworkAuthorityClient,
-        node_sync::SyncStatus,
-        test_utils::{spawn_test_authorities, test_authority_aggregator},
+        authority_active::gossip::GossipMetrics, authority_client::NetworkAuthorityClient,
+        node_sync::SyncStatus, test_utils::test_authority_aggregator,
     };
     use std::sync::{Arc, Mutex};
     use sui_types::{
@@ -322,7 +320,8 @@ mod test {
         object::Object,
     };
     use test_utils::{
-        authority::test_and_configure_authority_configs, messages::make_transfer_sui_transaction,
+        authority::{spawn_test_authorities, test_and_configure_authority_configs},
+        messages::make_transfer_sui_transaction,
     };
     use tokio::{sync::broadcast, time::Instant};
 

crates/sui-core/src/test_utils.rs

@@ -23,32 +23,10 @@ use sui_types::{
     object::Object,
 };
 
-// Can't import SuiNode directly from sui_node - circular dependency
-use test_utils::authority::{start_node, SuiNode};
-
 use futures::StreamExt;
 use tokio::time::sleep;
 use tracing::info;
 
-/// Spawn all authorities in the test committee into a separate tokio task.
-pub async fn spawn_test_authorities<I>(objects: I, config: &NetworkConfig) -> Vec<SuiNode>
-where
-    I: IntoIterator<Item = Object> + Clone,
-{
-    let mut handles = Vec::new();
-    for validator in config.validator_configs() {
-        let node = start_node(validator).await;
-        let state = node.state();
-
-        for o in objects.clone() {
-            state.insert_genesis_object(o).await
-        }
-
-        handles.push(node);
-    }
-    handles
-}
-
 /// Create a test authority aggregator.
 /// (duplicated from test-utils/src/authority.rs - that function can't be used
 /// in sui-core because of type name conflicts (sui_core::safe_client::SafeClient vs

crates/sui-node/src/lib.rs

@@ -72,20 +72,11 @@ pub struct SuiNode {
 }
 
 impl SuiNode {
-    pub async fn start(config: &NodeConfig) -> Result<SuiNode> {
+    pub async fn start(config: &NodeConfig, prometheus_registry: Registry) -> Result<SuiNode> {
         // TODO: maybe have a config enum that takes care of this for us.
         let is_validator = config.consensus_config().is_some();
         let is_full_node = !is_validator;
 
-        //
-        // Start metrics server
-        //
-        info!(
-            "Starting Prometheus HTTP endpoint at {}",
-            config.metrics_address
-        );
-        let prometheus_registry = metrics::start_prometheus_server(config.metrics_address);
-
         info!(node =? config.protocol_public_key(),
             "Initializing sui-node listening on {}", config.network_address
         );

crates/sui-node/src/main.rs

@@ -7,10 +7,11 @@ use multiaddr::Multiaddr;
 use std::path::PathBuf;
 use std::time::Duration;
 use sui_config::{Config, NodeConfig};
+use sui_node::metrics;
 use sui_telemetry::send_telemetry_event;
 use tokio::task;
 use tokio::time::sleep;
-use tracing::warn;
+use tracing::{info, warn};
 
 #[derive(Parser)]
 #[clap(rename_all = "kebab-case", version)]
@@ -43,16 +44,22 @@ const PROF_DUMP: &[u8] = b"prof.dump\0";
 
 #[tokio::main]
 async fn main() -> Result<()> {
+    let args = Args::parse();
+    let mut config = NodeConfig::load(&args.config_path)?;
+
+    let prometheus_registry = metrics::start_prometheus_server(config.metrics_address);
+    info!(
+        "Started Prometheus HTTP endpoint at {}",
+        config.metrics_address
+    );
+
     // Initialize logging
     let (_guard, filter_handle) =
         telemetry_subscribers::TelemetryConfig::new(env!("CARGO_BIN_NAME"))
             .with_env()
+            .with_prom_registry(&prometheus_registry)
             .init();
 
-    let args = Args::parse();
-
-    let mut config = NodeConfig::load(&args.config_path)?;
-
     if let Some(listen_address) = args.listen_address {
         config.network_address = listen_address;
     }
@@ -124,7 +131,7 @@ async fn main() -> Result<()> {
 
     sui_node::admin::start_admin_server(config.admin_interface_port, filter_handle);
 
-    let node = sui_node::SuiNode::start(&config).await?;
+    let node = sui_node::SuiNode::start(&config, prometheus_registry).await?;
     node.wait().await?;
 
     Ok(())

crates/sui-swarm/Cargo.toml

@@ -15,6 +15,7 @@ futures = "0.3.23"
 tempfile = "3.3.0"
 tonic-health = "0.6.0"
 tap = "1.0.1"
+prometheus = "0.13.1"
 
 sui-config = { path = "../sui-config" }
 sui-node = { path = "../sui-node" }

crates/sui-swarm/src/memory/node.rs

@@ -4,6 +4,7 @@
 use anyhow::anyhow;
 use anyhow::Result;
 use futures::FutureExt;
+use prometheus::Registry;
 use std::thread;
 use sui_config::NodeConfig;
 use sui_node::SuiNode;
@@ -177,9 +178,8 @@ impl Container {
                 }
             };
             let runtime = builder.enable_all().build().unwrap();
-
             runtime.block_on(async move {
-                let _server = SuiNode::start(&config).await.unwrap();
+                let _server = SuiNode::start(&config, Registry::new()).await.unwrap();
                 // Notify that we've successfully started the node
                 let _ = startup_sender.send(());
                 // run until canceled

crates/sui/Cargo.toml

@@ -60,6 +60,7 @@ jemalloc-ctl = "^0.5"
 [dev-dependencies]
 tempfile = "3.3.0"
 futures = "0.3.23"
+prometheus = "0.13.1"
 
 typed-store = { git = "https://github.com/MystenLabs/mysten-infra", rev = "d96230a9272c322a7eefac49708aadfff1eed77e"}
 typed-store-macros = { git = "https://github.com/MystenLabs/mysten-infra", rev = "d96230a9272c322a7eefac49708aadfff1eed77e"}

crates/sui/tests/full_node_tests.rs

@@ -1,10 +1,6 @@
 // Copyright (c) 2022, Mysten Labs, Inc.
 // SPDX-License-Identifier: Apache-2.0
 
-use std::net::SocketAddr;
-use std::str::FromStr;
-use std::{collections::BTreeMap, sync::Arc};
-
 use futures::future;
 use jsonrpsee::core::client::{Client, ClientT, Subscription, SubscriptionClientT};
 use jsonrpsee::http_client::{HttpClient, HttpClientBuilder};
@@ -13,6 +9,10 @@ use jsonrpsee::ws_client::WsClientBuilder;
 use move_core_types::account_address::AccountAddress;
 use move_core_types::identifier::Identifier;
 use move_core_types::language_storage::ModuleId;
+use prometheus::Registry;
+use std::net::SocketAddr;
+use std::str::FromStr;
+use std::{collections::BTreeMap, sync::Arc};
 use sui_types::base_types::SequenceNumber;
 use sui_types::event::TransferType;
 use sui_types::object::Owner;
@@ -115,7 +115,7 @@ async fn test_full_node_follows_txes() -> Result<(), anyhow::Error> {
     let (swarm, mut context, _) = setup_network_and_wallet().await?;
 
     let config = swarm.config().generate_fullnode_config();
-    let node = SuiNode::start(&config).await?;
+    let node = SuiNode::start(&config, Registry::new()).await?;
 
     let (transferred_object, _, receiver, digest) = transfer_coin(&mut context).await?;
     wait_for_tx(digest, node.state().clone()).await;
@@ -145,7 +145,7 @@ async fn test_full_node_shared_objects() -> Result<(), anyhow::Error> {
     let (swarm, context, _) = setup_network_and_wallet().await?;
 
     let config = swarm.config().generate_fullnode_config();
-    let node = SuiNode::start(&config).await?;
+    let node = SuiNode::start(&config, Registry::new()).await?;
 
     let sender = context.keystore.addresses().get(0).cloned().unwrap();
 
@@ -166,7 +166,7 @@ async fn test_full_node_move_function_index() -> Result<(), anyhow::Error> {
     let (swarm, context, _) = setup_network_and_wallet().await?;
 
     let config = swarm.config().generate_fullnode_config();
-    let node = SuiNode::start(&config).await?;
+    let node = SuiNode::start(&config, Registry::new()).await?;
     let sender = context.keystore.addresses().get(0).cloned().unwrap();
     let (package_ref, counter_id) = publish_basics_package_and_make_counter(&context, sender).await;
     let effects = increment_counter(&context, sender, None, package_ref, counter_id).await;
@@ -213,7 +213,7 @@ async fn test_full_node_indexes() -> Result<(), anyhow::Error> {
     let (swarm, mut context, _) = setup_network_and_wallet().await?;
 
     let config = swarm.config().generate_fullnode_config();
-    let node = SuiNode::start(&config).await?;
+    let node = SuiNode::start(&config, Registry::new()).await?;
 
     let (transferred_object, sender, receiver, digest) = transfer_coin(&mut context).await?;
 
@@ -358,7 +358,7 @@ async fn test_full_node_cold_sync() -> Result<(), anyhow::Error> {
     sleep(Duration::from_millis(1000)).await;
 
     let config = swarm.config().generate_fullnode_config();
-    let node = SuiNode::start(&config).await?;
+    let node = SuiNode::start(&config, Registry::new()).await?;
 
     wait_for_tx(digest, node.state().clone()).await;
 
@@ -380,7 +380,7 @@ async fn test_full_node_sync_flood() -> Result<(), anyhow::Error> {
     let (swarm, context, _) = setup_network_and_wallet().await?;
 
     let config = swarm.config().generate_fullnode_config();
-    let node = SuiNode::start(&config).await?;
+    let node = SuiNode::start(&config, Registry::new()).await?;
 
     let mut futures = Vec::new();
 
@@ -476,7 +476,7 @@ async fn set_up_subscription(swarm: &Swarm) -> Result<(SuiNode, Client), anyhow:
     let mut config = swarm.config().generate_fullnode_config();
     config.websocket_address = Some(ws_addr);
 
-    let node = SuiNode::start(&config).await?;
+    let node = SuiNode::start(&config, Registry::new()).await?;
 
     let client = WsClientBuilder::default()
         .build(&format!("ws://{}", ws_server_url))
@@ -499,7 +499,7 @@ async fn set_up_jsonrpc(
         .generate_fullnode_config_with_custom_db_path(fullnode_db_path, false);
     config.json_rpc_address = jsonrpc_addr;
 
-    let node = SuiNode::start(&config).await?;
+    let node = SuiNode::start(&config, Registry::new()).await?;
 
     let client = HttpClientBuilder::default().build(&format!("http://{}", jsonrpc_server_url))?;
     Ok((node, client))
@@ -884,7 +884,9 @@ async fn test_full_node_quorum_driver_basic() -> Result<(), anyhow::Error> {
 async fn test_validator_node_has_no_quorum_driver() {
     let configs = test_and_configure_authority_configs(1);
     let validator_config = &configs.validator_configs()[0];
-    let node = SuiNode::start(validator_config).await.unwrap();
+    let node = SuiNode::start(validator_config, Registry::new())
+        .await
+        .unwrap();
     assert!(node.quorum_driver().is_none());
     assert!(node.subscribe_to_quorum_driver_effects().is_err());
 }

crates/sui/tests/reconfiguration_tests.rs

@@ -3,6 +3,7 @@
 
 use futures::future::join_all;
 use multiaddr::Multiaddr;
+use prometheus::Registry;
 use sui_config::ValidatorInfo;
 use sui_core::authority_active::checkpoint_driver::{
     checkpoint_process_step, CheckpointProcessControl,
@@ -41,7 +42,7 @@ async fn reconfig_end_to_end_tests() {
     let mut states = Vec::new();
     let mut nodes = Vec::new();
     for validator in configs.validator_configs() {
-        let node = SuiNode::start(validator).await.unwrap();
+        let node = SuiNode::start(validator, Registry::new()).await.unwrap();
         let state = node.state();
 
         for gas in gas_objects.clone() {

crates/test-utils/src/authority.rs

@@ -1,6 +1,7 @@
 // Copyright (c) 2022, Mysten Labs, Inc.
 // SPDX-License-Identifier: Apache-2.0
 use crate::TEST_COMMITTEE_SIZE;
+use prometheus::Registry;
 use rand::{prelude::StdRng, SeedableRng};
 use std::collections::BTreeMap;
 use std::sync::Arc;
@@ -51,8 +52,8 @@ pub fn test_and_configure_authority_configs(committee_size: usize) -> NetworkCon
     configs
 }
 
-pub async fn start_node(config: &NodeConfig) -> SuiNode {
-    SuiNode::start(config).await.unwrap()
+pub async fn start_node(config: &NodeConfig, prom_registry: Registry) -> SuiNode {
+    SuiNode::start(config, prom_registry).await.unwrap()
 }
 
 /// Spawn all authorities in the test committee into a separate tokio task.
@@ -62,7 +63,8 @@ where
 {
     let mut handles = Vec::new();
     for validator in config.validator_configs() {
-        let node = start_node(validator).await;
+        let prom_registry = Registry::new();
+        let node = start_node(validator, prom_registry).await;
         let state = node.state();
 
         for o in objects.clone() {