fix(pydantic-ai): Stop emitting duplicate spans

sentry_sdk/integrations/pydantic_ai/patches/graph_nodes.py

@@ -59,6 +59,14 @@ def _patch_graph_nodes() -> None:
 
     @wraps(original_model_request_run)
     async def wrapped_model_request_run(self: "Any", ctx: "Any") -> "Any":
+        # Avoid creating a duplicate span if run() is invoked after stream().
+        # This fails here: https://github.com/pydantic/pydantic-ai/blob/916fc83e8929470679db5ac1b3065bda5d5f4253/pydantic_ai_slim/pydantic_ai/_agent_graph.py#L1119
+        did_stream = getattr(self, "_did_stream", False)
+        # Do not create a duplicate span when a cached result is served.
+        cached_result = getattr(self, "_result", None)
+        if did_stream or cached_result is not None:
+            return await original_model_request_run(self, ctx)
+
         messages, model, model_settings = _extract_span_data(self, ctx)
 
         with ai_client_span(messages, None, model, model_settings) as span:
@@ -85,6 +93,14 @@ def create_wrapped_stream(
         @asynccontextmanager
         @wraps(original_stream_method)
         async def wrapped_model_request_stream(self: "Any", ctx: "Any") -> "Any":
+            # Avoid creating a duplicate span if the function is invoked twice.
+            # This fails here: https://github.com/pydantic/pydantic-ai/blob/916fc83e8929470679db5ac1b3065bda5d5f4253/pydantic_ai_slim/pydantic_ai/_agent_graph.py#L1128
+            did_stream = getattr(self, "_did_stream", False)
+            if did_stream:
+                async with original_stream_method(self, ctx) as stream:
+                    yield stream
+                return
+
             messages, model, model_settings = _extract_span_data(self, ctx)
 
             # Create chat span for streaming request