[router] responses api POST and GET with local storage (#10581)

Co-authored-by: key4ng <rukeyang@gmail.com>

sgl-router/src/data_connector/response_memory_store.rs

@@ -74,13 +74,16 @@ impl ResponseStorage for MemoryResponseStorage {
 
         // Store the response
         store.responses.insert(response_id.clone(), response);
+        tracing::info!("memory_store_size" = store.responses.len());
 
         Ok(response_id)
     }
 
     async fn get_response(&self, response_id: &ResponseId) -> Result<Option<StoredResponse>> {
         let store = self.store.read();
-        Ok(store.responses.get(response_id).cloned())
+        let result = store.responses.get(response_id).cloned();
+        tracing::info!("memory_get_response" = %response_id.0, found = result.is_some());
+        Ok(result)
     }
 
     async fn delete_response(&self, response_id: &ResponseId) -> Result<()> {
@@ -200,6 +203,20 @@ pub struct MemoryStoreStats {
 mod tests {
     use super::*;
 
+    #[tokio::test]
+    async fn test_store_with_custom_id() {
+        let store = MemoryResponseStorage::new();
+        let mut response = StoredResponse::new("Input".to_string(), "Output".to_string(), None);
+        response.id = ResponseId::from_string("resp_custom".to_string());
+        store.store_response(response.clone()).await.unwrap();
+        let retrieved = store
+            .get_response(&ResponseId::from_string("resp_custom".to_string()))
+            .await
+            .unwrap();
+        assert!(retrieved.is_some());
+        assert_eq!(retrieved.unwrap().output, "Output");
+    }
+
     #[tokio::test]
     async fn test_memory_store_basic() {
         let store = MemoryResponseStorage::new();

sgl-router/src/data_connector/responses.rs

@@ -1,5 +1,6 @@
 use async_trait::async_trait;
 use serde::{Deserialize, Serialize};
+use serde_json::Value;
 use std::collections::HashMap;
 use std::sync::Arc;
 
@@ -55,6 +56,10 @@ pub struct StoredResponse {
 
     /// Model used for generation
     pub model: Option<String>,
+
+    /// Raw OpenAI response payload
+    #[serde(default)]
+    pub raw_response: Value,
 }
 
 impl StoredResponse {
@@ -70,6 +75,7 @@ impl StoredResponse {
             created_at: chrono::Utc::now(),
             user: None,
             model: None,
+            raw_response: Value::Null,
         }
     }
 }
@@ -175,3 +181,9 @@ pub trait ResponseStorage: Send + Sync {
 
 /// Type alias for shared storage
 pub type SharedResponseStorage = Arc<dyn ResponseStorage>;
+
+impl Default for StoredResponse {
+    fn default() -> Self {
+        Self::new(String::new(), String::new(), None)
+    }
+}