[router] Worker Management Workflow Engine (#11868)

sgl-router/src/server.rs

@@ -22,8 +22,8 @@ use tracing::{error, info, warn, Level};
 use crate::{
     config::{ConnectionMode, HistoryBackend, RouterConfig, RoutingMode},
     core::{
-        worker_to_info, Job, JobQueue, JobQueueConfig, LoadMonitor, WorkerManager, WorkerRegistry,
-        WorkerType,
+        worker_to_info, workflow::WorkflowEngine, Job, JobQueue, JobQueueConfig, LoadMonitor,
+        WorkerManager, WorkerRegistry, WorkerType,
     },
     data_connector::{
         MemoryConversationItemStorage, MemoryConversationStorage, MemoryResponseStorage,
@@ -77,6 +77,7 @@ pub struct AppContext {
     pub configured_reasoning_parser: Option<String>,
     pub configured_tool_parser: Option<String>,
     pub worker_job_queue: Arc<OnceLock<Arc<JobQueue>>>,
+    pub workflow_engine: Arc<OnceLock<Arc<WorkflowEngine>>>,
 }
 
 impl AppContext {
@@ -95,6 +96,7 @@ impl AppContext {
         conversation_item_storage: SharedConversationItemStorage,
         load_monitor: Option<Arc<LoadMonitor>>,
         worker_job_queue: Arc<OnceLock<Arc<JobQueue>>>,
+        workflow_engine: Arc<OnceLock<Arc<WorkflowEngine>>>,
     ) -> Self {
         let configured_reasoning_parser = router_config.reasoning_parser.clone();
         let configured_tool_parser = router_config.tool_call_parser.clone();
@@ -116,6 +118,7 @@ impl AppContext {
             configured_reasoning_parser,
             configured_tool_parser,
             worker_job_queue,
+            workflow_engine,
         }
     }
 }
@@ -979,8 +982,9 @@ pub async fn startup(config: ServerConfig) -> Result<(), Box<dyn std::error::Err
         config.router_config.worker_startup_check_interval_secs,
     )));
 
-    // Create empty OnceLock for worker job queue (will be initialized below)
+    // Create empty OnceLock for worker job queue and workflow engine (will be initialized below)
     let worker_job_queue = Arc::new(OnceLock::new());
+    let workflow_engine = Arc::new(OnceLock::new());
 
     // Create AppContext with all initialized components
     let app_context = AppContext::new(
@@ -997,6 +1001,7 @@ pub async fn startup(config: ServerConfig) -> Result<(), Box<dyn std::error::Err
         conversation_item_storage,
         load_monitor,
         worker_job_queue,
+        workflow_engine,
     );
 
     let app_context = Arc::new(app_context);
@@ -1008,17 +1013,38 @@ pub async fn startup(config: ServerConfig) -> Result<(), Box<dyn std::error::Err
         .set(worker_job_queue)
         .expect("JobQueue should only be initialized once");
 
+    // Initialize workflow engine and register workflows
+    let engine = Arc::new(WorkflowEngine::new());
+
+    engine
+        .event_bus()
+        .subscribe(Arc::new(crate::core::workflow::LoggingSubscriber))
+        .await;
+
+    engine.register_workflow(crate::core::workflow::create_worker_registration_workflow());
+    app_context
+        .workflow_engine
+        .set(engine)
+        .expect("WorkflowEngine should only be initialized once");
+    info!("Workflow engine initialized with worker registration workflow");
+
     info!(
         "Initializing workers for routing mode: {:?}",
         config.router_config.mode
     );
-    WorkerManager::initialize_workers(
-        &config.router_config,
-        &app_context.worker_registry,
-        Some(&app_context.policy_registry),
-    )
-    .await
-    .map_err(|e| format!("Failed to initialize workers: {}", e))?;
+
+    // Submit worker initialization job to queue
+    let job_queue = app_context
+        .worker_job_queue
+        .get()
+        .expect("JobQueue should be initialized");
+    let job = Job::InitializeWorkersFromConfig {
+        router_config: Box::new(config.router_config.clone()),
+    };
+    job_queue
+        .submit(job)
+        .await
+        .map_err(|e| format!("Failed to submit worker initialization job: {}", e))?;
 
     let worker_stats = app_context.worker_registry.stats();
     info!(