Merge pull request #14816 from leventov/no-streaming-logging

Fix concurrency/scaling when many Python threads do streaming using *sync* completions

Author: krrishdholakia

litellm/litellm_core_utils/streaming_handler.py

@@ -1619,11 +1619,12 @@ def __next__(self):  # noqa: PLR0915
                             completion_start_time=datetime.datetime.now()
                         )
                     ## LOGGING
-                    executor.submit(
-                        self.run_success_logging_and_cache_storage,
-                        response,
-                        cache_hit,
-                    )  # log response
+                    if not litellm.disable_streaming_logging:
+                        executor.submit(
+                            self.run_success_logging_and_cache_storage,
+                            response,
+                            cache_hit,
+                        )  # log response
                     choice = response.choices[0]
                     if isinstance(choice, StreamingChoices):
                         self.response_uptil_now += choice.delta.get("content", "") or ""

litellm/llms/custom_httpx/http_handler.py

@@ -935,7 +935,7 @@ def _create_sync_transport(self) -> Optional[HTTPTransport]:
         if litellm.force_ipv4:
             return HTTPTransport(local_address="0.0.0.0")
         else:
-            return None
+            return getattr(litellm, 'sync_transport', None)
 
 
 def get_async_httpx_client(