openai: Remove is_raw_response attribute from StreamWrapper

Author: SrdjanLL

instrumentation/elastic-opentelemetry-instrumentation-openai/src/opentelemetry/instrumentation/openai/__init__.py

@@ -196,7 +196,6 @@ def _chat_completion_wrapper(self, wrapped, instance, args, kwargs):
                 _record_operation_duration_metric(self.operation_duration_metric, error_attributes, start_time)
                 raise
 
-            is_raw_response = _is_raw_response(result)
             if kwargs.get("stream"):
                 return StreamWrapper(
                     stream=result,
@@ -208,12 +207,12 @@ def _chat_completion_wrapper(self, wrapped, instance, args, kwargs):
                     start_time=start_time,
                     token_usage_metric=self.token_usage_metric,
                     operation_duration_metric=self.operation_duration_metric,
-                    is_raw_response=is_raw_response,
                 )
 
             logger.debug(f"openai.resources.chat.completions.Completions.create result: {result}")
 
             # if the caller is using with_raw_response we need to parse the output to get the response class we expect
+            is_raw_response = _is_raw_response(result)
             if is_raw_response:
                 result = result.parse()
             response_attributes = _get_attributes_from_response(
@@ -271,7 +270,6 @@ async def _async_chat_completion_wrapper(self, wrapped, instance, args, kwargs):
                 _record_operation_duration_metric(self.operation_duration_metric, error_attributes, start_time)
                 raise
 
-            is_raw_response = _is_raw_response(result)
             if kwargs.get("stream"):
                 return StreamWrapper(
                     stream=result,
@@ -283,12 +281,12 @@ async def _async_chat_completion_wrapper(self, wrapped, instance, args, kwargs):
                     start_time=start_time,
                     token_usage_metric=self.token_usage_metric,
                     operation_duration_metric=self.operation_duration_metric,
-                    is_raw_response=is_raw_response,
                 )
 
             logger.debug(f"openai.resources.chat.completions.AsyncCompletions.create result: {result}")
 
             # if the caller is using with_raw_response we need to parse the output to get the response class we expect
+            is_raw_response = _is_raw_response(result)
             if is_raw_response:
                 result = result.parse()
             response_attributes = _get_attributes_from_response(

instrumentation/elastic-opentelemetry-instrumentation-openai/src/opentelemetry/instrumentation/openai/wrappers.py

@@ -47,7 +47,6 @@ def __init__(
         start_time: float,
         token_usage_metric: Histogram,
         operation_duration_metric: Histogram,
-        is_raw_response: bool,
     ):
         # we need to wrap the original response even in case of raw_responses
         super().__init__(stream)
@@ -60,7 +59,6 @@ def __init__(
         self.token_usage_metric = token_usage_metric
         self.operation_duration_metric = operation_duration_metric
         self.start_time = start_time
-        self.is_raw_response = is_raw_response
 
         self.response_id = None
         self.model = None
@@ -125,8 +123,6 @@ def __exit__(self, exc_type, exc_value, traceback):
     def __iter__(self):
         stream = self.__wrapped__
         try:
-            if self.is_raw_response:
-                stream = stream.parse()
             for chunk in stream:
                 self.process_chunk(chunk)
                 yield chunk
@@ -145,8 +141,6 @@ async def __aexit__(self, exc_type, exc_value, traceback):
     async def __aiter__(self):
         stream = self.__wrapped__
         try:
-            if self.is_raw_response:
-                stream = stream.parse()
             async for chunk in stream:
                 self.process_chunk(chunk)
                 yield chunk
@@ -171,8 +165,6 @@ def parse(self):
             start_time=self.start_time,
             token_usage_metric=self.token_usage_metric,
             operation_duration_metric=self.operation_duration_metric,
-            # Crucially, mark the new wrapper as NOT raw after parsing
-            is_raw_response=False,
         )
 
         # Handle original sync/async iterators accordingly