fix: dummy el race condition

Author: fbwoolf

beacon_node/client/src/builder.rs

@@ -205,12 +205,24 @@ where
                 jwt_secret_path: None,
             };
 
+            // Create a channel to wait for the dummy EL to be ready
+            let (ready_tx, ready_rx) = tokio::sync::oneshot::channel();
+
             // Spawn the dummy EL in a background task
             tokio::spawn(async move {
-                if let Err(e) = dummy_el::start_dummy_el(dummy_el_config).await {
+                if let Err(e) = dummy_el::prepare_and_start_dummy_el(dummy_el_config, ready_tx).await
+                {
                     eprintln!("Error starting dummy execution layer: {:?}", e);
                 }
             });
+
+            // Wait for the dummy EL to be ready before continuing
+            if let Err(_) = ready_rx.await {
+                return Err(
+                    "Dummy execution layer failed to start or signal readiness".to_string(),
+                );
+            }
+            info!("Dummy execution layer is ready");
         }
 
         let kzg_err_msg = |e| format!("Failed to load trusted setup: {:?}", e);

dummy_el/src/lib.rs

@@ -20,6 +20,7 @@ use serde_json::{json, Value as JsonValue};
 use std::net::SocketAddr;
 use std::path::PathBuf;
 use std::sync::Arc;
+use tokio::sync::oneshot;
 use tracing::{debug, error, warn};
 
 const JSONRPC_VERSION: &str = "2.0";
@@ -36,6 +37,20 @@ pub struct DummyElConfig {
     pub jwt_secret_path: Option<PathBuf>,
 }
 
+/// Represents a prepared dummy execution layer ready to run
+pub struct PreparedDummyEl {
+    engine_listener: tokio::net::TcpListener,
+    engine_app: Router,
+    rpc_listener: tokio::net::TcpListener,
+    rpc_app: Router,
+    ws_listener: tokio::net::TcpListener,
+    ws_app: Router,
+    metrics_listener: tokio::net::TcpListener,
+    metrics_app: Router,
+    p2p_tcp_task: tokio::task::JoinHandle<()>,
+    p2p_udp_task: tokio::task::JoinHandle<()>,
+}
+
 #[derive(Debug, Clone)]
 struct AppState {
     jwt_secret: Option<Vec<u8>>,
@@ -322,11 +337,29 @@ fn read_jwt_secret(path: &PathBuf) -> anyhow::Result<Vec<u8>> {
     Ok(bytes)
 }
 
-/// Start the dummy execution layer server
+/// Prepare the dummy execution layer for startup
 ///
-/// This spawns the dummy EL HTTP servers on the configured ports.
-/// Returns a task handle that should be spawned or awaited.
-pub async fn start_dummy_el(config: DummyElConfig) -> anyhow::Result<()> {
+/// This function binds all necessary ports and prepares the servers,
+/// then signals readiness via the oneshot channel before running the servers.
+/// The function does not return until the servers are shut down.
+pub async fn prepare_and_start_dummy_el(
+    config: DummyElConfig,
+    ready_tx: oneshot::Sender<()>,
+) -> anyhow::Result<()> {
+    let prepared = prepare_dummy_el(config).await?;
+
+    // Signal that we're ready
+    let _ = ready_tx.send(());
+
+    // Now run the servers
+    prepared.run().await
+}
+
+/// Prepare the dummy execution layer server without starting it
+///
+/// This binds all ports and prepares the servers but does not start accepting connections.
+/// Returns a `PreparedDummyEl` that can be run with the `run()` method.
+pub async fn prepare_dummy_el(config: DummyElConfig) -> anyhow::Result<PreparedDummyEl> {
     // Read JWT secret if provided
     let jwt_secret = match &config.jwt_secret_path {
         Some(path) => match read_jwt_secret(path) {
@@ -426,22 +459,51 @@ pub async fn start_dummy_el(config: DummyElConfig) -> anyhow::Result<()> {
         }
     });
 
-    debug!("Ready to accept requests on all ports");
-
-    // Spawn all servers concurrently
+    // Bind all servers without starting them
     let engine_listener = tokio::net::TcpListener::bind(engine_addr).await?;
     let rpc_listener = tokio::net::TcpListener::bind(rpc_addr).await?;
     let ws_listener = tokio::net::TcpListener::bind(ws_addr).await?;
     let metrics_listener = tokio::net::TcpListener::bind(metrics_addr).await?;
 
-    tokio::select! {
-        result = axum::serve(engine_listener, engine_app) => result?,
-        result = axum::serve(rpc_listener, rpc_app) => result?,
-        result = axum::serve(ws_listener, ws_app) => result?,
-        result = axum::serve(metrics_listener, metrics_app) => result?,
-        _ = p2p_tcp_task => {},
-        _ = p2p_udp_task => {},
+    debug!("All listeners bound and ready");
+
+    Ok(PreparedDummyEl {
+        engine_listener,
+        engine_app,
+        rpc_listener,
+        rpc_app,
+        ws_listener,
+        ws_app,
+        metrics_listener,
+        metrics_app,
+        p2p_tcp_task,
+        p2p_udp_task,
+    })
+}
+
+impl PreparedDummyEl {
+    /// Run the prepared dummy execution layer servers
+    pub async fn run(self) -> anyhow::Result<()> {
+        debug!("Running dummy execution layer servers");
+
+        tokio::select! {
+            result = axum::serve(self.engine_listener, self.engine_app) => result?,
+            result = axum::serve(self.rpc_listener, self.rpc_app) => result?,
+            result = axum::serve(self.ws_listener, self.ws_app) => result?,
+            result = axum::serve(self.metrics_listener, self.metrics_app) => result?,
+            _ = self.p2p_tcp_task => {},
+            _ = self.p2p_udp_task => {},
+        }
+
+        Ok(())
     }
+}
 
-    Ok(())
+/// Start the dummy execution layer server (legacy function)
+///
+/// This is a convenience function that prepares and starts the dummy EL.
+/// For more control, use `prepare_dummy_el()` and `prepare_and_start_dummy_el()`.
+pub async fn start_dummy_el(config: DummyElConfig) -> anyhow::Result<()> {
+    let prepared = prepare_dummy_el(config).await?;
+    prepared.run().await
 }