Merge pull request #4719 from stacks-network/feat/signer-metrics

feat: signer monitoring server

libsigner/Cargo.toml

@@ -18,8 +18,10 @@ path = "./src/libsigner.rs"
 [dependencies]
 clarity = { path = "../clarity" }
 hashbrown = { workspace = true }
+lazy_static = "1.4.0"
 libc = "0.2"
 libstackerdb = { path = "../libstackerdb" }
+prometheus = { version = "0.9", optional = true }
 serde = "1"
 serde_derive = "1"
 serde_stacker = "0.1"
@@ -50,3 +52,6 @@ sha2 = { version = "0.10", features = ["asm"] }
 
 [target.'cfg(any(not(any(target_arch = "x86_64", target_arch = "x86", target_arch = "aarch64")), any(target_os = "windows")))'.dependencies]
 sha2 = { version = "0.10" }
+
+[features]
+monitoring_prom = ["prometheus"]

stacks-signer/Cargo.toml

@@ -24,8 +24,10 @@ backoff = "0.4"
 clarity = { path = "../clarity" }
 clap = { version = "4.1.1", features = ["derive", "env"] }
 hashbrown = { workspace = true }
+lazy_static = "1.4.0"
 libsigner = { path = "../libsigner" }
 libstackerdb = { path = "../libstackerdb" }
+prometheus = { version = "0.9", optional = true }
 rand_core = "0.6"
 reqwest = { version = "0.11.22", default-features = false, features = ["blocking", "json", "rustls-tls"] }
 serde = "1"
@@ -37,6 +39,7 @@ slog-term = "2.6.0"
 stacks-common = { path = "../stacks-common" }
 stackslib = { path = "../stackslib" }
 thiserror = "1.0"
+tiny_http = { version = "0.12", optional = true }
 toml = "0.5.6"
 tracing = "0.1.37"
 tracing-subscriber = { version = "0.3.17", features = ["env-filter"] }
@@ -60,3 +63,6 @@ features = ["arbitrary_precision", "unbounded_depth"]
 [dependencies.secp256k1]
 version = "0.24.3"
 features = ["serde", "recovery"]
+
+[features]
+monitoring_prom = ["libsigner/monitoring_prom", "prometheus", "tiny_http"]

stacks-signer/src/client/stacks_client.rs

@@ -207,6 +207,8 @@ impl StacksClient {
             estimated_len: Some(tx.tx_len()),
             transaction_payload: to_hex(&tx.payload.serialize_to_vec()),
         };
+        let timer =
+            crate::monitoring::new_rpc_call_timer(&self.fees_transaction_path(), &self.http_origin);
         let send_request = || {
             self.stacks_node_client
                 .post(self.fees_transaction_path())
@@ -219,6 +221,7 @@ impl StacksClient {
         if !response.status().is_success() {
             return Err(ClientError::RequestFailure(response.status()));
         }
+        timer.stop_and_record();
         let fee_estimate_response = response.json::<RPCFeeEstimateResponse>()?;
         let fee = fee_estimate_response
             .estimations
@@ -268,6 +271,8 @@ impl StacksClient {
             block,
             chain_id: self.chain_id,
         };
+        let timer =
+            crate::monitoring::new_rpc_call_timer(&self.block_proposal_path(), &self.http_origin);
         let send_request = || {
             self.stacks_node_client
                 .post(self.block_proposal_path())
@@ -279,6 +284,7 @@ impl StacksClient {
         };
 
         let response = retry_with_exponential_backoff(send_request)?;
+        timer.stop_and_record();
         if !response.status().is_success() {
             return Err(ClientError::RequestFailure(response.status()));
         }
@@ -355,13 +361,16 @@ impl StacksClient {
     /// Get the current peer info data from the stacks node
     pub fn get_peer_info(&self) -> Result<RPCPeerInfoData, ClientError> {
         debug!("Getting stacks node info...");
+        let timer =
+            crate::monitoring::new_rpc_call_timer(&self.core_info_path(), &self.http_origin);
         let send_request = || {
             self.stacks_node_client
                 .get(self.core_info_path())
                 .send()
                 .map_err(backoff::Error::transient)
         };
         let response = retry_with_exponential_backoff(send_request)?;
+        timer.stop_and_record();
         if !response.status().is_success() {
             return Err(ClientError::RequestFailure(response.status()));
         }
@@ -402,13 +411,18 @@ impl StacksClient {
         reward_cycle: u64,
     ) -> Result<Option<Vec<NakamotoSignerEntry>>, ClientError> {
         debug!("Getting reward set for reward cycle {reward_cycle}...");
+        let timer = crate::monitoring::new_rpc_call_timer(
+            &self.reward_set_path(reward_cycle),
+            &self.http_origin,
+        );
         let send_request = || {
             self.stacks_node_client
                 .get(self.reward_set_path(reward_cycle))
                 .send()
                 .map_err(backoff::Error::transient)
         };
         let response = retry_with_exponential_backoff(send_request)?;
+        timer.stop_and_record();
         if !response.status().is_success() {
             return Err(ClientError::RequestFailure(response.status()));
         }
@@ -419,13 +433,17 @@ impl StacksClient {
     /// Retreive the current pox data from the stacks node
     pub fn get_pox_data(&self) -> Result<RPCPoxInfoData, ClientError> {
         debug!("Getting pox data...");
+        #[cfg(feature = "monitoring_prom")]
+        let timer = crate::monitoring::new_rpc_call_timer(&self.pox_path(), &self.http_origin);
         let send_request = || {
             self.stacks_node_client
                 .get(self.pox_path())
                 .send()
                 .map_err(backoff::Error::transient)
         };
         let response = retry_with_exponential_backoff(send_request)?;
+        #[cfg(feature = "monitoring_prom")]
+        timer.stop_and_record();
         if !response.status().is_success() {
             return Err(ClientError::RequestFailure(response.status()));
         }
@@ -458,18 +476,21 @@ impl StacksClient {
     }
 
     /// Helper function to retrieve the account info from the stacks node for a specific address
-    fn get_account_entry(
+    pub fn get_account_entry(
         &self,
         address: &StacksAddress,
     ) -> Result<AccountEntryResponse, ClientError> {
         debug!("Getting account info...");
+        let timer =
+            crate::monitoring::new_rpc_call_timer(&self.accounts_path(address), &self.http_origin);
         let send_request = || {
             self.stacks_node_client
                 .get(self.accounts_path(address))
                 .send()
                 .map_err(backoff::Error::transient)
         };
         let response = retry_with_exponential_backoff(send_request)?;
+        timer.stop_and_record();
         if !response.status().is_success() {
             return Err(ClientError::RequestFailure(response.status()));
         }
@@ -536,6 +557,8 @@ impl StacksClient {
     pub fn submit_transaction(&self, tx: &StacksTransaction) -> Result<Txid, ClientError> {
         let txid = tx.txid();
         let tx = tx.serialize_to_vec();
+        let timer =
+            crate::monitoring::new_rpc_call_timer(&self.transaction_path(), &self.http_origin);
         let send_request = || {
             self.stacks_node_client
                 .post(self.transaction_path())
@@ -548,6 +571,7 @@ impl StacksClient {
                 })
         };
         let response = retry_with_exponential_backoff(send_request)?;
+        timer.stop_and_record();
         if !response.status().is_success() {
             return Err(ClientError::RequestFailure(response.status()));
         }
@@ -576,12 +600,14 @@ impl StacksClient {
         let body =
             json!({"sender": self.stacks_address.to_string(), "arguments": args}).to_string();
         let path = self.read_only_path(contract_addr, contract_name, function_name);
+        let timer = crate::monitoring::new_rpc_call_timer(&path, &self.http_origin);
         let response = self
             .stacks_node_client
             .post(path)
             .header("Content-Type", "application/json")
             .body(body)
             .send()?;
+        timer.stop_and_record();
         if !response.status().is_success() {
             return Err(ClientError::RequestFailure(response.status()));
         }

stacks-signer/src/config.rs

@@ -186,6 +186,8 @@ pub struct GlobalConfig {
     pub auth_password: String,
     /// The path to the signer's database file
     pub db_path: PathBuf,
+    /// Metrics endpoint
+    pub metrics_endpoint: Option<SocketAddr>,
 }
 
 /// Internal struct for loading up the config file
@@ -221,6 +223,8 @@ struct RawConfigFile {
     pub auth_password: String,
     /// The path to the signer's database file or :memory: for an in-memory database
     pub db_path: String,
+    /// Metrics endpoint
+    pub metrics_endpoint: Option<String>,
 }
 
 impl RawConfigFile {
@@ -298,6 +302,19 @@ impl TryFrom<RawConfigFile> for GlobalConfig {
         let sign_timeout = raw_data.sign_timeout_ms.map(Duration::from_millis);
         let db_path = raw_data.db_path.into();
 
+        let metrics_endpoint = match raw_data.metrics_endpoint {
+            Some(endpoint) => Some(
+                endpoint
+                    .to_socket_addrs()
+                    .map_err(|_| ConfigError::BadField("endpoint".to_string(), endpoint.clone()))?
+                    .next()
+                    .ok_or_else(|| {
+                        ConfigError::BadField("endpoint".to_string(), endpoint.clone())
+                    })?,
+            ),
+            None => None,
+        };
+
         Ok(Self {
             node_host: raw_data.node_host,
             endpoint,
@@ -315,6 +332,7 @@ impl TryFrom<RawConfigFile> for GlobalConfig {
             max_tx_fee_ustx: raw_data.max_tx_fee_ustx,
             auth_password: raw_data.auth_password,
             db_path,
+            metrics_endpoint,
         })
     }
 }
@@ -345,6 +363,10 @@ impl GlobalConfig {
             0 => "default".to_string(),
             _ => (self.tx_fee_ustx as f64 / 1_000_000.0).to_string(),
         };
+        let metrics_endpoint = match &self.metrics_endpoint {
+            Some(endpoint) => endpoint.to_string(),
+            None => "None".to_string(),
+        };
         format!(
             r#"
 Stacks node host: {node_host}
@@ -354,14 +376,16 @@ Public key: {public_key}
 Network: {network}
 Database path: {db_path}
 DKG transaction fee: {tx_fee} uSTX
+Metrics endpoint: {metrics_endpoint}
 "#,
             node_host = self.node_host,
             endpoint = self.endpoint,
             stacks_address = self.stacks_address,
             public_key = StacksPublicKey::from_private(&self.stacks_private_key).to_hex(),
             network = self.network,
             db_path = self.db_path.to_str().unwrap_or_default(),
-            tx_fee = tx_fee
+            tx_fee = tx_fee,
+            metrics_endpoint = metrics_endpoint,
         )
     }
 }
@@ -384,6 +408,7 @@ pub fn build_signer_config_tomls(
     mut port_start: usize,
     max_tx_fee_ustx: Option<u64>,
     tx_fee_ustx: Option<u64>,
+    mut metrics_port_start: Option<usize>,
 ) -> Vec<String> {
     let mut signer_config_tomls = vec![];
 
@@ -438,6 +463,17 @@ tx_fee_ustx = {tx_fee_ustx}
             )
         }
 
+        if let Some(metrics_port) = metrics_port_start {
+            let metrics_endpoint = format!("localhost:{}", metrics_port);
+            signer_config_toml = format!(
+                r#"
+{signer_config_toml}
+metrics_endpoint = "{metrics_endpoint}"
+"#
+            );
+            metrics_port_start = Some(metrics_port + 1);
+        }
+
         signer_config_tomls.push(signer_config_toml);
     }
 
@@ -469,6 +505,7 @@ mod tests {
             3000,
             None,
             None,
+            Some(4000),
         );
 
         let config =
@@ -477,6 +514,7 @@ mod tests {
         assert_eq!(config.auth_password, "melon");
         assert!(config.max_tx_fee_ustx.is_none());
         assert!(config.tx_fee_ustx.is_none());
+        assert_eq!(config.metrics_endpoint, Some("localhost:4000".to_string()));
     }
 
     #[test]
@@ -501,6 +539,7 @@ mod tests {
             3000,
             None,
             None,
+            None,
         );
 
         let config =
@@ -526,6 +565,7 @@ mod tests {
             3000,
             max_tx_fee_ustx,
             tx_fee_ustx,
+            None,
         );
 
         let config =
@@ -546,6 +586,7 @@ mod tests {
             3000,
             max_tx_fee_ustx,
             None,
+            None,
         );
 
         let config =
@@ -570,6 +611,7 @@ mod tests {
             3000,
             None,
             tx_fee_ustx,
+            None,
         );
 
         let config =
@@ -598,6 +640,7 @@ Public key: 03bc489f27da3701d9f9e577c88de5567cf4023111b7577042d55cde4d823a3505
 Network: testnet
 Database path: :memory:
 DKG transaction fee: 0.01 uSTX
+Metrics endpoint: 0.0.0.0:9090
 "#
             )
         );

stacks-signer/src/lib.rs

@@ -34,3 +34,6 @@ pub mod runloop;
 pub mod signer;
 /// The state module for the signer
 pub mod signerdb;
+
+/// The monitoring server for the signer
+pub mod monitoring;