Update proxy_server.py

proxy_server.py

@@ -12,7 +12,7 @@ from fastapi.responses import StreamingResponse, JSONResponse
 from fastapi.middleware.cors import CORSMiddleware
 from loguru import logger
 from typing import AsyncGenerator, Set, Optional, Dict, Any, List
-from urllib.parse import urlparse # Import để phân tích URL
+# from urllib.parse import urlparse # Removed: No longer needed for logging
 
 # --- Logging Configuration ---
 logger.remove()
@@ -50,19 +50,12 @@ async def lifespan(app: FastAPI):
     timeout_config = httpx.Timeout(connect=CONNECT_TIMEOUT, read=READ_TIMEOUT, write=WRITE_TIMEOUT, pool=POOL_TIMEOUT)
     proxy_config = {"http://": HTTP_PROXY, "https://": HTTP_PROXY} if HTTP_PROXY else None
 
-    # --- FIX: Hide full target endpoint URL from logs ---
-    # Parse the URL to get only the hostname for logging
-    try:
-        parsed_url = urlparse(OPENAI_API_ENDPOINT)
-        target_host = parsed_url.netloc # e.g., api.openai.com
-    except Exception:
-        target_host = "[Invalid Target URL]" # Handle potential parsing errors
-
-    logger.info(f"Initializing httpx client. Target Host: {target_host}") # Log only the host
+    # --- FIX: Completely hide target endpoint info from logs ---
+    logger.info("Initializing httpx client for upstream requests.") # Generic message
     # --- End Fix ---
 
     if proxy_config:
-        logger.info(f"Using outbound proxy: {HTTP_PROXY}") # Proxy URL might still be sensitive depending on config
+        logger.info(f"Using outbound proxy: {HTTP_PROXY}") # Proxy URL might still be sensitive
     if not OPENAI_API_KEY:
         logger.warning("OPENAI_API_KEY is not set. Requests to the target endpoint might fail if it requires authentication.")
     if not VALID_API_KEYS:
@@ -128,7 +121,17 @@ def claude_request_to_openai_payload(claude_request: Dict[str, Any]) -> Dict[str
     messages = []
     system_prompt = claude_request.get("system")
     if system_prompt:
-        messages.append({"role": "system", "content": system_prompt})
+        # Ensure system prompt content is a string
+        if isinstance(system_prompt, list):
+             # Combine text blocks if system prompt is a list (like in Claude's format)
+            system_content = "\n".join(block.get("text", "") for block in system_prompt if block.get("type") == "text")
+        elif isinstance(system_prompt, str):
+            system_content = system_prompt
+        else:
+             system_content = "" # Handle other unexpected types if necessary
+        if system_content:
+             messages.append({"role": "system", "content": system_content})
+
 
     for msg in claude_request.get("messages", []):
         role = msg.get("role")
@@ -158,9 +161,7 @@ def claude_request_to_openai_payload(claude_request: Dict[str, Any]) -> Dict[str
         # Add other relevant parameter mappings here (e.g., presence_penalty, frequency_penalty)
     }
 
-    # --- FIX: Avoid logging potentially large/sensitive payload ---
     # logger.debug("Converted Claude request to OpenAI payload.") # Keep this simple
-    # --- End Fix ---
     return openai_payload
 
 def openai_response_to_claude_response(openai_response: Dict[str, Any], claude_request_id: str) -> Dict[str, Any]:
@@ -346,9 +347,8 @@ async def proxy_claude_to_openai(request: Request):
     request_id = f"msg_{uuid.uuid4().hex[:24]}" # Generate a unique ID for logging/tracking
     try:
         claude_request_data = await request.json()
-        # --- FIX: Avoid logging potentially large/sensitive request data ---
+        # Avoid logging potentially large/sensitive request data
         logger.info(f"[{request_id}] Received request. Stream: {claude_request_data.get('stream', False)}. Model: {claude_request_data.get('model')}")
-        # --- End Fix ---
     except json.JSONDecodeError:
         logger.error(f"[{request_id}] Invalid JSON received in request body.")
         return create_error_response(400, "invalid_request_error", "Invalid JSON data in request body.")
@@ -369,15 +369,13 @@ async def proxy_claude_to_openai(request: Request):
         # Add other headers if needed
     }
     if OPENAI_API_KEY:
-        # --- FIX: Avoid logging API key ---
+        # Avoid logging API key
         # logger.debug(f"[{request_id}] Adding Authorization header to upstream request.") # Log presence, not value
         headers["Authorization"] = f"Bearer {OPENAI_API_KEY}"
-        # --- End Fix ---
 
     try:
-        # --- FIX: Hide full target endpoint URL and payload from logs ---
+        # Hide full target endpoint URL and payload from logs
         logger.debug(f"[{request_id}] Sending request to upstream API...")
-        # --- End Fix ---
 
         # Build the request to the target endpoint
         target_request = client.build_request(
@@ -385,8 +383,7 @@ async def proxy_claude_to_openai(request: Request):
             url=OPENAI_API_ENDPOINT,
             headers=headers,
             json=openai_payload,
-            # Timeout is handled globally by the client, but can be overridden here
-            # timeout=httpx.Timeout(READ_TIMEOUT)
+            # Timeout is handled globally by the client
         )
 
         # Send the request and handle the response
@@ -411,10 +408,8 @@ async def proxy_claude_to_openai(request: Request):
             logger.info(f"[{request_id}] Upstream response is non-streaming. Converting.")
             # FIX: Remove await here
             openai_response_data = response.json()
-            # --- FIX: Avoid logging full response data ---
-            # logger.debug(f"[{request_id}] Non-streaming response from OpenAI: {json.dumps(openai_response_data)}")
+            # Avoid logging full response data
             logger.debug(f"[{request_id}] Received non-streaming response from upstream.")
-            # --- End Fix ---
             try:
                 claude_response_data = openai_response_to_claude_response(openai_response_data, request_id)
                 return JSONResponse(content=claude_response_data)
@@ -436,11 +431,15 @@ async def proxy_claude_to_openai(request: Request):
             error_detail_text = json.dumps(error_detail) # Convert back to string for logging snippet
         except json.JSONDecodeError:
             # If not JSON, read as text
-            error_detail_text = e.response.text # Use .text instead of await .aread() as body is likely read
+            try:
+                # Read text synchronously since the body is likely already loaded for non-streaming error responses
+                error_detail_text = e.response.text
+            except Exception:
+                 logger.warning(f"[{request_id}] Could not read error response body as text.")
+
 
-        # --- FIX: Log error snippet, avoid full potentially sensitive detail ---
+        # Log error snippet, avoid full potentially sensitive detail
         logger.error(f"[{request_id}] HTTP error from target endpoint ({status_code}). Response snippet: {error_detail_text[:200]}...")
-        # --- End Fix ---
 
         # Map common HTTP errors to Claude error types
         # Use generic messages in production to avoid leaking upstream details
@@ -461,15 +460,13 @@ async def proxy_claude_to_openai(request: Request):
         return create_error_response(status_code, err_type, msg)
 
     except httpx.TimeoutException:
-         # --- FIX: Hide target endpoint URL from timeout log ---
+         # Hide target endpoint URL from timeout log
         logger.error(f"[{request_id}] Request to target endpoint timed out ({READ_TIMEOUT}s).")
-        # --- End Fix ---
         return create_error_response(504, "api_error", "Gateway Timeout: Request to upstream API timed out.")
     except httpx.RequestError as e:
-        # --- FIX: Hide target endpoint URL from request error log ---
+        # Hide target endpoint URL from request error log
         # The exception 'e' might contain the URL, so log a generic message
         logger.error(f"[{request_id}] Network error connecting to target endpoint: {type(e).__name__}")
-        # --- End Fix ---
         return create_error_response(502, "api_error", f"Bad Gateway: Network error connecting to upstream API.")
     except Exception as e:
         logger.exception(f"[{request_id}] Unexpected error during proxy operation: {e}") # Use logger.exception to include traceback