gen-ai instrumentation(feat): OpenAI responses create instrumentation (#4474)

Author: eternalcuriouslearner

.github/workflows/test.yml

@@ -214,25 +214,6 @@ jobs:
       - name: Run tests
         run: tox -e py314-test-instrumentation-openai-v2-latest -- -ra
 
-  py313-test-instrumentation-openai-v2-pydantic1_ubuntu-latest:
-    name: instrumentation-openai-v2-pydantic1 3.13 Ubuntu
-    runs-on: ubuntu-latest
-    timeout-minutes: 30
-    steps:
-      - name: Checkout repo @ SHA - ${{ github.sha }}
-        uses: actions/checkout@v4
-
-      - name: Set up Python 3.13
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.13"
-
-      - name: Install tox
-        run: pip install tox-uv
-
-      - name: Run tests
-        run: tox -e py313-test-instrumentation-openai-v2-pydantic1 -- -ra
-
   pypy3-test-instrumentation-openai-v2-oldest_ubuntu-latest:
     name: instrumentation-openai-v2-oldest pypy-3.10 Ubuntu
     runs-on: ubuntu-latest

instrumentation-genai/opentelemetry-instrumentation-openai-v2/CHANGELOG.md

@@ -41,6 +41,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Add strongly typed Responses API extractors with validation and content
   extraction improvements
   ([#4337](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4337))
+- Add instrumentation for OpenAI Responses API `create`
+  ([#4474](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4474))
 - Add completion hook support.
   ([#4315](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4315))
 - Fix `response_format` handling: map `json_object`/`json_schema` to `json` output type.

instrumentation-genai/opentelemetry-instrumentation-openai-v2/src/opentelemetry/instrumentation/openai_v2/__init__.py

@@ -64,6 +64,7 @@
 ---
 """
 
+from importlib import import_module
 from typing import Collection
 
 from wrapt import wrap_function_wrapper
@@ -91,6 +92,9 @@
     chat_completions_create_v_old,
     embeddings_create,
 )
+from .patch_responses import (
+    responses_create,
+)
 
 
 class OpenAIInstrumentor(BaseInstrumentor):
@@ -177,10 +181,33 @@ def _instrument(self, **kwargs):
             ),
         )
 
+        responses_module = _get_responses_module()
+        # Responses instrumentation is intentionally limited to the latest
+        # experimental semconv path. Unlike chat completions, we do not carry
+        # a second legacy wrapper here; the current implementation is built on
+        # the inference handler lifecycle and would need a separate old-path
+        # implementation to support legacy semconv mode.
+        if responses_module is not None and latest_experimental_enabled:
+            wrap_function_wrapper(
+                "openai.resources.responses.responses",
+                "Responses.create",
+                responses_create(handler),
+            )
+
     def _uninstrument(self, **kwargs):
         import openai  # pylint: disable=import-outside-toplevel  # noqa: PLC0415
 
         unwrap(openai.resources.chat.completions.Completions, "create")
         unwrap(openai.resources.chat.completions.AsyncCompletions, "create")
         unwrap(openai.resources.embeddings.Embeddings, "create")
         unwrap(openai.resources.embeddings.AsyncEmbeddings, "create")
+        responses_module = _get_responses_module()
+        if responses_module is not None:
+            unwrap(responses_module.Responses, "create")
+
+
+def _get_responses_module():
+    try:
+        return import_module("openai.resources.responses.responses")
+    except ImportError:
+        return None

instrumentation-genai/opentelemetry-instrumentation-openai-v2/src/opentelemetry/instrumentation/openai_v2/patch_responses.py

@@ -0,0 +1,57 @@
+# Copyright The OpenTelemetry Authors
+# SPDX-License-Identifier: Apache-2.0
+
+from __future__ import annotations
+
+from opentelemetry.util.genai.handler import TelemetryHandler
+from opentelemetry.util.genai.types import Error
+
+from .response_extractors import (
+    apply_request_attributes,
+    extract_params,
+    get_inference_creation_kwargs,
+    set_invocation_response_attributes,
+)
+from .response_wrappers import ResponseStreamWrapper
+from .utils import is_streaming
+
+
+def responses_create(handler: TelemetryHandler):
+    """Wrap the `create` method of the `Responses` class to trace it."""
+
+    capture_content = handler.should_capture_content()
+
+    def traced_method(wrapped, instance, args, kwargs):
+        params = extract_params(**kwargs)
+        invocation = handler.start_inference(
+            **get_inference_creation_kwargs(params, instance)
+        )
+        apply_request_attributes(invocation, params, capture_content)
+
+        try:
+            result = wrapped(*args, **kwargs)
+            parsed_result = _get_response_stream_result(result)
+
+            if is_streaming(kwargs):
+                return ResponseStreamWrapper(
+                    parsed_result,
+                    invocation,
+                    capture_content,
+                )
+
+            set_invocation_response_attributes(
+                invocation, parsed_result, capture_content
+            )
+            invocation.stop()
+            return result
+        except Exception as error:
+            invocation.fail(Error(type=type(error), message=str(error)))
+            raise
+
+    return traced_method
+
+
+def _get_response_stream_result(result):
+    if hasattr(result, "parse"):
+        return result.parse()
+    return result