fix(proxy_server.py): setup depedencies on server startup

litellm/integrations/traceloop.py

@@ -1,7 +1,8 @@
 class TraceloopLogger:
     def __init__(self):
         from traceloop.sdk.tracing.tracing import TracerWrapper
-
+        from traceloop.sdk import Traceloop
+        Traceloop.init(app_name="Litellm-Server", disable_batch=True)
         self.tracer_wrapper = TracerWrapper()
 
     def log_event(self, kwargs, response_obj, start_time, end_time, print_verbose):

litellm/proxy/proxy_server.py

@@ -319,7 +319,6 @@ async def user_api_key_auth(request: Request, api_key: str = fastapi.Security(ap
 def prisma_setup(database_url: Optional[str]): 
     global prisma_client, proxy_logging_obj, user_api_key_cache
 
-    proxy_logging_obj._init_litellm_callbacks()
     if database_url is not None:
         try: 
             prisma_client = PrismaClient(database_url=database_url, proxy_logging_obj=proxy_logging_obj)
@@ -527,10 +526,6 @@ def load_router_config(router: Optional[litellm.Router], config_file_path: str):
                     # these are litellm callbacks - "langfuse", "sentry", "wandb"
                     else:
                         litellm.success_callback.append(callback)
-                        if callback == "traceloop":
-                            from traceloop.sdk import Traceloop
-                            print_verbose(f"{blue_color_code} Initializing Traceloop SDK - \nRunning:`Traceloop.init(app_name='Litellm-Server', disable_batch=True)`")
-                            Traceloop.init(app_name="Litellm-Server", disable_batch=True)
                 print_verbose(f"{blue_color_code} Initialized Success Callbacks - {litellm.success_callback} {reset_color_code}")
             elif key == "failure_callback":
                 litellm.failure_callback = []
@@ -816,11 +811,9 @@ def get_litellm_model_info(model: dict = {}):
     
 @router.on_event("startup")
 async def startup_event():
-    global prisma_client, master_key, use_background_health_checks, use_queue
+    global prisma_client, master_key, use_background_health_checks
     import json
 
-    print(f"VALUE OF USE_QUEUE: {use_queue}")
-
     ### LOAD CONFIG ### 
     worker_config = litellm.get_secret("WORKER_CONFIG")
     print_verbose(f"worker_config: {worker_config}")
@@ -833,6 +826,8 @@ async def startup_event():
         initialize(**worker_config)
     
     
+    proxy_logging_obj._init_litellm_callbacks() # INITIALIZE LITELLM CALLBACKS ON SERVER STARTUP <- do this to catch any logging errors on startup, not when calls are being made
+
     if use_background_health_checks:
         asyncio.create_task(_run_background_health_check()) # start the background health check coroutine. 
 

litellm/proxy/utils.py

@@ -7,7 +7,7 @@ from litellm.proxy.hooks.parallel_request_limiter import MaxParallelRequestsHand
 from litellm.integrations.custom_logger import CustomLogger
 def print_verbose(print_statement):
     if litellm.set_verbose:
-        print(print_statement) # noqa
+        print(f"LiteLLM Proxy: {print_statement}") # noqa
 ### LOGGING ### 
 class ProxyLogging: 
     """
@@ -26,7 +26,7 @@ class ProxyLogging:
         pass
 
     def _init_litellm_callbacks(self):
-        
+        print_verbose(f"INITIALIZING LITELLM CALLBACKS!")
         litellm.callbacks.append(self.max_parallel_request_limiter)
         for callback in litellm.callbacks: 
             if callback not in litellm.input_callback:

litellm/router.py

@@ -386,7 +386,7 @@ class Router:
             self.print_verbose(f'Async Response: {response}')
             return response
         except Exception as e: 
-            self.print_verbose(f"An exception occurs: {e}")
+            self.print_verbose(f"An exception occurs: {e}\n\n Traceback{traceback.format_exc()}")
             original_exception = e
             try: 
                 self.print_verbose(f"Trying to fallback b/w models")

requirements.txt

@@ -18,4 +18,5 @@ celery
 psutil
 mangum
 google-generativeai 
-async_generator # for ollama 
+traceloop-sdk==0.5.3
+langfuse==1.14.0