chore: Provide heartbeat retries in config (#614)

* chore: Provide hearbeat retries in config

* fmt

Author: wojciechsromek

deploy/prod/common-values-iris-mpc.yaml

@@ -1,4 +1,4 @@
-image: "ghcr.io/worldcoin/iris-mpc:v0.8.34"
+image: "ghcr.io/worldcoin/iris-mpc:v0.8.35"
 
 environment: prod
 replicaCount: 1

deploy/prod/smpcv2-0-prod/values-iris-mpc.yaml

@@ -56,6 +56,9 @@ env:
   - name: SMPC__HEARTBEAT_INTERVAL_SECS
     value: "2"
 
+  - name: SMPC__HEARTBEAT_INITIAL_RETRIES
+    value: "50"
+
   - name: SMPC__PATH
     value: "/data/"
 

deploy/prod/smpcv2-1-prod/values-iris-mpc.yaml

@@ -56,6 +56,9 @@ env:
   - name: SMPC__HEARTBEAT_INTERVAL_SECS
     value: "2"
 
+  - name: SMPC__HEARTBEAT_INITIAL_RETRIES
+    value: "50"
+    
   - name: SMPC__PATH
     value: "/data/"
 

deploy/prod/smpcv2-2-prod/values-iris-mpc.yaml

@@ -56,6 +56,9 @@ env:
   - name: SMPC__HEARTBEAT_INTERVAL_SECS
     value: "2"
 
+  - name: SMPC__HEARTBEAT_INITIAL_RETRIES
+    value: "50"
+
   - name: SMPC__PATH
     value: "/data/"
 

iris-mpc-common/src/config/mod.rs

@@ -64,6 +64,9 @@ pub struct Config {
     #[serde(default = "default_heartbeat_interval_secs")]
     pub heartbeat_interval_secs: u64,
 
+    #[serde(default = "default_heartbeat_initial_retries")]
+    pub heartbeat_initial_retries: u64,
+
     #[serde(default)]
     pub fake_db_size: usize,
 
@@ -89,6 +92,10 @@ fn default_heartbeat_interval_secs() -> u64 {
     30
 }
 
+fn default_heartbeat_initial_retries() -> u64 {
+    10
+}
+
 impl Config {
     pub fn load_config(prefix: &str) -> eyre::Result<Config> {
         let settings = config::Config::builder();

iris-mpc/src/bin/server.rs

@@ -54,7 +54,6 @@ const REGION: &str = "eu-north-1";
 const RNG_SEED_INIT_DB: u64 = 42;
 const SQS_POLLING_INTERVAL: Duration = Duration::from_secs(1);
 const MAX_CONCURRENT_REQUESTS: usize = 32;
-const HEARTBEAT_INITIAL_RETRIES: usize = 5;
 
 static CURRENT_BATCH_SIZE: LazyLock<Mutex<usize>> = LazyLock::new(|| Mutex::new(0));
 
@@ -976,7 +975,7 @@ async fn server_main(config: Config) -> eyre::Result<()> {
                     // If it's the first time after startup, we allow a few retries to let the other
                     // nodes start up as well.
                     if last_response[i] == String::default()
-                        && retries[i] < HEARTBEAT_INITIAL_RETRIES
+                        && retries[i] < config.heartbeat_initial_retries
                     {
                         retries[i] += 1;
                         tracing::warn!("Node {} did not respond with success, retrying...", host);