feat: made runtime timeout configurable

Author: raphael-goetz

crates/taurus-manual/src/main.rs

@@ -1,5 +1,5 @@
 use std::path::Path;
-use std::time::Instant;
+use std::time::{Duration, Instant};
 
 use clap::Parser;
 use log::error;
@@ -207,7 +207,8 @@ async fn main() {
         return;
     }
 
-    let remote = NATSRemoteRuntime::new(client.clone());
+    let remote =
+        NATSRemoteRuntime::with_execution_result_timeout(client.clone(), Duration::from_secs(30));
     let emitter = NATSRespondEmitter::new(client);
     let engine = ExecutionEngine::new();
 

crates/taurus-provider/src/providers/remote/nats_remote_runtime.rs

@@ -11,13 +11,18 @@ use tucana::shared::NodeExecutionResult;
 
 pub struct NATSRemoteRuntime {
     client: Client,
+    execution_result_timeout: Duration,
 }
 
-const REMOTE_EXECUTION_RESULT_TIMEOUT: Duration = Duration::from_secs(30);
-
 impl NATSRemoteRuntime {
-    pub fn new(client: Client) -> Self {
-        NATSRemoteRuntime { client }
+    pub fn with_execution_result_timeout(
+        client: Client,
+        execution_result_timeout: Duration,
+    ) -> Self {
+        NATSRemoteRuntime {
+            client,
+            execution_result_timeout,
+        }
     }
 }
 
@@ -64,7 +69,7 @@ impl RemoteRuntime for NATSRemoteRuntime {
                 "Failed to receive any response messages from a remote runtime.",
             ));
         }
-        match tokio::time::timeout(REMOTE_EXECUTION_RESULT_TIMEOUT, self.client.flush()).await {
+        match tokio::time::timeout(self.execution_result_timeout, self.client.flush()).await {
             Ok(Ok(())) => {}
             Ok(Err(err)) => {
                 log::error!(
@@ -90,8 +95,7 @@ impl RemoteRuntime for NATSRemoteRuntime {
             }
         }
 
-        let message = match tokio::time::timeout(REMOTE_EXECUTION_RESULT_TIMEOUT, sub.next()).await
-        {
+        let message = match tokio::time::timeout(self.execution_result_timeout, sub.next()).await {
             Ok(Some(message)) => message,
             Ok(None) => {
                 log::error!("RemoteRuntimeException: NATS reply subscription closed");

crates/taurus/src/app/mod.rs

@@ -31,7 +31,10 @@ pub async fn run() {
     let (runtime_status_service, runtime_execution_service, mut runtime_status_heartbeat_task) =
         setup_dynamic_services_if_needed(&config).await;
 
-    let nats_remote = NATSRemoteRuntime::new(client.clone());
+    let nats_remote = NATSRemoteRuntime::with_execution_result_timeout(
+        client.clone(),
+        Duration::from_secs(config.remote_runtime_timeout_secs),
+    );
     let runtime_emitter = NATSRespondEmitter::new(client.clone());
     let mut worker_task = worker::spawn_worker(
         client,

crates/taurus/src/config/mod.rs

@@ -36,6 +36,9 @@ pub struct Config {
 
     /// Timeout in seconds for Aquila gRPC requests.
     pub aquila_grpc_request_timeout_secs: u64,
+
+    /// Timeout in seconds for remote runtime NATS flush and response waits.
+    pub remote_runtime_timeout_secs: u64,
 }
 
 /// Implementation for all relevant `Taurus` startup configurations
@@ -66,6 +69,7 @@ impl Config {
                 "AQUILA_GRPC_REQUEST_TIMEOUT_SECS",
                 10_u64,
             ),
+            remote_runtime_timeout_secs: env_with_default("REMOTE_RUNTIME_TIMEOUT_SECS", 30_u64),
         }
     }
 }

docs/installation.mdx

@@ -103,3 +103,4 @@ Set `MODE=dynamic` when the IDE is required/needed.
 | `RUNTIME_STATUS_UPDATE_INTERVAL_SECONDS` | Interval (in seconds) of updating the current runtime status. | `30s` |
 | `AQUILA_GRPC_CONNECT_TIMEOUT_SECS` | Timeout in seconds for establishing Aquila gRPC channels. | `2s` |
 | `AQUILA_GRPC_REQUEST_TIMEOUT_SECS` | Timeout in seconds for Aquila gRPC requests. | `10s` |
+| `REMOTE_RUNTIME_TIMEOUT_SECS` | Timeout in seconds for remote runtime NATS flush and response waits. | `30s` |