fix (backend): separate retry policies for sequencer and rpc endpoint

Author: krsnapaudel

backend/src/configs/network.rs

@@ -5,6 +5,15 @@ use super::env::EnvVarParser;
 /// Default network status refetch interval in seconds
 const DEFAULT_NETWORK_STATUS_REFETCH_INTERVAL_S: u64 = 10;
 
+/// Default retry policy max retries
+const DEFAULT_RETRY_POLICY_MAX_RETRIES: u64 = 5;
+
+/// Default retry policy total retry time
+const DEFAULT_RETRY_POLICY_TOTAL_RETRY_TIME: u64 = 60;
+
+/// Default retry policy base
+const DEFAULT_RETRY_POLICY_BASE: f64 = 1.5;
+
 #[derive(Debug, Clone)]
 pub(crate) struct NetworkConfig {
     /// JSON-RPC Endpoint for Strata sequencer
@@ -16,11 +25,11 @@ pub(crate) struct NetworkConfig {
     /// Bundler health check URL (overrides `.env`)
     bundler_url: String,
 
-    /// Max retries in querying status
-    max_retries: u64,
+    /// Max retries for status queries
+    retry_policy_max_retries: u64,
 
-    /// Total time in seconds to spend retrying
-    total_retry_time: u64,
+    /// Total time in seconds to spend retrying status queries
+    retry_policy_total_time_s: u64,
 
     /// Network status refetch interval in seconds
     status_refetch_interval_s: u64,
@@ -37,9 +46,10 @@ impl NetworkConfig {
         let bundler_url = String::parse_env_var("BUNDLER_URL")
             .unwrap_or_else(|| "http://localhost:8434".to_string());
 
-        let max_retries = u64::parse_env_var("MAX_STATUS_RETRIES").unwrap_or(5);
-
-        let total_retry_time = u64::parse_env_var("TOTAL_RETRY_TIME").unwrap_or(60);
+        let retry_policy_max_retries = u64::parse_env_var("NETWORK_STATUS_MAX_RETRIES")
+            .unwrap_or(DEFAULT_RETRY_POLICY_MAX_RETRIES);
+        let retry_policy_total_time_s = u64::parse_env_var("NETWORK_STATUS_TOTAL_RETRY_TIME_S")
+            .unwrap_or(DEFAULT_RETRY_POLICY_TOTAL_RETRY_TIME);
 
         let status_refetch_interval_s = u64::parse_env_var("NETWORK_STATUS_REFETCH_INTERVAL_S")
             .unwrap_or(DEFAULT_NETWORK_STATUS_REFETCH_INTERVAL_S);
@@ -50,8 +60,8 @@ impl NetworkConfig {
             sequencer_url,
             rpc_url,
             bundler_url,
-            max_retries,
-            total_retry_time,
+            retry_policy_max_retries,
+            retry_policy_total_time_s,
             status_refetch_interval_s,
         }
     }
@@ -71,18 +81,26 @@ impl NetworkConfig {
         &self.bundler_url
     }
 
-    /// Getter for `max_retries`
-    pub(crate) fn max_retries(&self) -> u64 {
-        self.max_retries
+    /// Getter for `status_refetch_interval_s`
+    pub(crate) fn status_refetch_interval(&self) -> u64 {
+        self.status_refetch_interval_s
     }
 
-    /// Getter for `total_retry_time`
-    pub(crate) fn total_retry_time(&self) -> u64 {
-        self.total_retry_time
+    /// Retry policy for sequencer status queries
+    pub(crate) fn sequencer_retry_policy(&self) -> crate::utils::retry_policy::ExponentialBackoff {
+        crate::utils::retry_policy::ExponentialBackoff::new(
+            self.retry_policy_max_retries,
+            self.retry_policy_total_time_s,
+            DEFAULT_RETRY_POLICY_BASE,
+        )
     }
 
-    /// Getter for `status_refetch_interval_s`
-    pub(crate) fn status_refetch_interval(&self) -> u64 {
-        self.status_refetch_interval_s
+    /// Retry policy for RPC endpoint status queries
+    pub(crate) fn rpc_retry_policy(&self) -> crate::utils::retry_policy::ExponentialBackoff {
+        crate::utils::retry_policy::ExponentialBackoff::new(
+            self.retry_policy_max_retries,
+            self.retry_policy_total_time_s,
+            DEFAULT_RETRY_POLICY_BASE,
+        )
     }
 }

backend/src/network/status.rs

@@ -12,11 +12,7 @@ use crate::{
 };
 
 /// Calls `strata_syncStatus` using `jsonrpsee`
-async fn call_rpc_status(
-    config: &NetworkConfig,
-    client: &HttpClient,
-    retry_policy: ExponentialBackoff,
-) -> Status {
+async fn call_rpc_status(client: &HttpClient, retry_policy: ExponentialBackoff) -> Status {
     let mut retry_count: u64 = 0;
 
     loop {
@@ -32,7 +28,7 @@ async fn call_rpc_status(
                 }
             }
             Err(e) => {
-                if retry_count < config.max_retries() {
+                if retry_count < retry_policy.max_retries() {
                     let delay_seconds = retry_policy.get_delay(retry_count);
                     if delay_seconds > 0 {
                         info!(?delay_seconds, "Retrying `strata_syncStatus` after");
@@ -69,17 +65,13 @@ pub(crate) async fn fetch_statuses_task(context: Arc<NetworkMonitoringContext>)
     let sequencer_client = create_rpc_client(context.config.sequencer_url());
     let rpc_client = create_rpc_client(context.config.rpc_url());
     let http_client = reqwest::Client::new();
-    let retry_policy = ExponentialBackoff::new(
-        context.config.max_retries(),
-        context.config.total_retry_time(),
-        1.5,
-    );
 
     loop {
         interval.tick().await;
 
-        let sequencer = call_rpc_status(&context.config, &sequencer_client, retry_policy).await;
-        let rpc_endpoint = call_rpc_status(&context.config, &rpc_client, retry_policy).await;
+        let sequencer =
+            call_rpc_status(&sequencer_client, context.config.sequencer_retry_policy()).await;
+        let rpc_endpoint = call_rpc_status(&rpc_client, context.config.rpc_retry_policy()).await;
         let bundler_endpoint = check_bundler_health(&http_client, &context.config).await;
 
         let new_status = NetworkStatus {

backend/src/utils/retry_policy.rs

@@ -20,6 +20,11 @@ impl ExponentialBackoff {
         }
     }
 
+    /// Returns the maximum number of retries.
+    pub fn max_retries(&self) -> u64 {
+        self.max_retries
+    }
+
     /// Returns the delay in seconds.
     pub fn get_delay(&self, retry_counter: u64) -> u64 {
         if retry_counter == 0 {

docker-compose.yml

@@ -8,6 +8,9 @@ services:
       RUST_LOG: INFO
       STRATA_SEQUENCER_URL: https://strataseq.testnet.alpenlabs.io
       RPC_URL: https://rpc.testnet.alpenlabs.io
+      NETWORK_STATUS_MAX_RETRIES: 5
+      NETWORK_STATUS_TOTAL_RETRY_TIME_S: 60
+      NETWORK_STATUS_REFETCH_INTERVAL_S: 10
       STRATA_BRIDGE_OPERATORS_COUNT: 4
       STRATA_BRIDGE_1_RPC_URL: https://bridge.testnet.alpenlabs.io/1
       STRATA_BRIDGE_1_PUBLIC_KEY: 0273441f2ba801b557b23c15829f4a87c02332d59a71499da1479048e6175ff4e0