fix: refresh worker models every 20 seconds by default

py_src/vllm_router/router.py

@@ -90,7 +90,7 @@ class Router:
         health_failure_threshold: Number of consecutive health check failures before marking worker unhealthy. Default: 3
         health_success_threshold: Number of consecutive health check successes before marking worker healthy. Default: 2
         health_check_timeout_secs: Timeout in seconds for health check requests. Default: 5
-        health_check_interval_secs: Interval in seconds between runtime health checks. Default: 60
+        health_check_interval_secs: Interval in seconds between runtime health checks. Default: 20
         health_check_endpoint: Health check endpoint path. Default: '/health'
     """
 

py_src/vllm_router/router_args.py

@@ -81,7 +81,7 @@ class RouterArgs:
     health_failure_threshold: int = 3
     health_success_threshold: int = 2
     health_check_timeout_secs: int = 5
-    health_check_interval_secs: int = 60
+    health_check_interval_secs: int = 20
     health_check_endpoint: str = "/health"
     # Circuit breaker configuration
     cb_failure_threshold: int = 10

src/config/types.rs

@@ -363,7 +363,7 @@ impl Default for HealthCheckConfig {
             failure_threshold: 3,
             success_threshold: 2,
             timeout_secs: 5,
-            check_interval_secs: 60,
+            check_interval_secs: 20,
             endpoint: "/health".to_string(),
         }
     }

src/core/worker_registry.rs

@@ -474,7 +474,7 @@ impl WorkerRegistry {
 
         let handle = tokio::spawn(async move {
             const LOAD_RESET_INTERVAL: u64 = 10;
-            const MODEL_REFRESH_INTERVAL: u64 = 5;
+            const MODEL_REFRESH_INTERVAL: u64 = 1;
 
             let mut interval =
                 tokio::time::interval(tokio::time::Duration::from_secs(check_interval_secs));

src/lib.rs

@@ -302,7 +302,7 @@ impl Router {
         health_failure_threshold = 3,
         health_success_threshold = 2,
         health_check_timeout_secs = 5,
-        health_check_interval_secs = 60,
+        health_check_interval_secs = 20,
         health_check_endpoint = String::from("/health"),
         // IGW defaults
         enable_igw = false,

src/main.rs

@@ -321,7 +321,7 @@ struct CliArgs {
     health_check_timeout_secs: u64,
 
     /// Interval in seconds between runtime health checks
-    #[arg(long, default_value_t = 60)]
+    #[arg(long, default_value_t = 20)]
     health_check_interval_secs: u64,
 
     /// Health check endpoint path