feat(llmobs): [MLOB-2680] add apm span for openai responses endpoint

ddtrace/contrib/internal/openai/_endpoint_hooks.py

@@ -715,3 +715,77 @@ def _record_response(self, pin, integration, span, args, kwargs, resp, error):
         else:
             span.set_metric("openai.response.total_bytes", getattr(resp, "total_bytes", 0))
         return resp
+
+
+class _ResponseHook(_EndpointHook):
+    _request_arg_params = ("api_key", "api_base", "api_type", "request_id", "api_version", "organization")
+    _request_kwarg_params = (
+        "model",
+        "include",
+        "instructions",
+        "max_output_tokens",
+        "metadata",
+        "parallel_tool_calls",
+        "previous_response_id",
+        "reasoning",
+        "service_tier",
+        "store",
+        "stream",
+        "temperature",
+        "text",
+        "tool_choice",
+        "tools",
+        "top_p",
+        "truncation",
+        "user",
+    )
+    _response_attrs = ("created_at", "id", "model", "tools")
+    ENDPOINT_NAME = "responses"
+    HTTP_METHOD_TYPE = "POST"
+    OPERATION_ID = "createResponseCompletion"
+
+    def _record_request(self, pin, integration, instance, span, args, kwargs):
+        super()._record_request(pin, integration, instance, span, args, kwargs)
+
+        input_data = kwargs.get("input", [])
+        if input_data:
+            if isinstance(input_data, str):
+                input_data = [input_data]
+
+            span._set_ctx_item("llmobs.response.input", input_data)
+
+        if parse_version(OPENAI_VERSION) >= (1, 26) and kwargs.get("stream"):
+            if kwargs.get("stream_options", {}).get("include_usage", None) is not None:
+                # Only perform token chunk auto-extraction if this option is not explicitly set
+                return
+            span._set_ctx_item("_dd.auto_extract_token_chunk", True)
+            stream_options = kwargs.get("stream_options", {})
+            stream_options["include_usage"] = True
+            kwargs["stream_options"] = stream_options
+
+    def _record_response(self, pin, integration, span, args, kwargs, resp, error):
+        resp = super()._record_response(pin, integration, span, args, kwargs, resp, error)
+        if kwargs.get("stream") and error is None:
+            return self._handle_streamed_response(integration, span, kwargs, resp, is_completion=False)
+        integration.llmobs_set_tags(span, args=[], kwargs=kwargs, response=resp, operation="chat")
+        if not resp:
+            return resp
+        span._set_ctx_item("llmobs.response.output", resp.output)
+        if getattr(resp, "tools", None):
+            response_tools = []
+            for tool in resp.tools:
+                tool_dict = {}
+
+                if hasattr(tool, "type"):
+                    tool_dict["type"] = getattr(tool, "type")
+                if hasattr(tool, "name"):
+                    tool_dict["name"] = getattr(tool, "name")
+                if tool_dict:
+                    response_tools.append(tool_dict)
+
+            if response_tools:
+                span.set_tag("openai.response.tools", response_tools)
+
+        integration.llmobs_set_tags(span, args=[], kwargs=kwargs, response=resp, operation="responses")
+        integration.record_usage(span, resp.usage)
+        return resp

ddtrace/contrib/internal/openai/patch.py

@@ -72,6 +72,9 @@ def get_version():
         "delete": _endpoint_hooks._FileDeleteHook,
         "retrieve_content": _endpoint_hooks._FileDownloadHook,
     },
+    "responses.Responses": {
+        "create": _endpoint_hooks._ResponseHook,
+    },
 }
 
 OPENAI_WITH_RAW_RESPONSE_ARG = "_dd.with_raw_response"

tests/contrib/openai/cassettes/v1/response_create.yaml

@@ -0,0 +1,97 @@
+interactions:
+- request:
+    body: '{"input":[{"role":"system","content":"You are a helpful assistant."},{"role":"user","content":"Who
+      won the world series in 2020?"},{"role":"assistant","content":"The Los Angeles
+      Dodgers won the World Series in 2020."},{"role":"user","content":"Where was
+      it played?"}],"model":"gpt-4.1","max_output_tokens":100,"top_p":0.9,"user":"ddtrace-test"}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate, zstd
+      connection:
+      - keep-alive
+      content-length:
+      - '344'
+      content-type:
+      - application/json
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.76.0
+      x-stainless-arch:
+      - arm64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - MacOS
+      x-stainless-package-version:
+      - 1.76.0
+      x-stainless-read-timeout:
+      - '600'
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.12.10
+    method: POST
+    uri: https://api.openai.com/v1/responses
+  response:
+    body:
+      string: !!binary |
+        H4sIAAAAAAAAA3RTwW7bMAy95ysInZvAcpykzm1YsWHAgB1WbIdhMBiZjrXJkiFRXYsi/z5Ydpxm
+        ay9B/B5JPT6SzwsAoWuxB+Ep9NX2VkrZkNwV6+xWlvJ2t6EdNvKwa9ZlqVSmcswLmRfbYkPNWoqb
+        oYA7/CLF5yLOBhpx5QmZ6goHTu6KTbnJd+td4gIjxzDkKNf1hpjqMYm8d17swUZjEqDtOaKqiVGb
+        cM0G9lGxdvYK7/CxcpH7yBW735RImWUj52oyw9PHnpfFSi7zLN8ss2Ipi6mhlCj28GMBAPCcfmen
+        unA8G5Vnt6UajDpQc8i3tN7IJt9tUb5qVKrBTz2lKhQCHulCvOVIIpWzTPYi6aWsq7LnnumR5+wU
+        gNY6xrNPP35ekSl8D+K+JcizPIPvzpsavpLXFOAPBugNPlENyPDRuAPBZ90QfNBkatAW3nmj7ZGd
+        vYF7esSwgrtIwA64JXj/5dunu6UsoUdbU6fVTYLJsvYE4fJGO1RDBm51AEuRPRoImgk8ckseuEWb
+        clvXEQTGWscugGsS2KNnrXSPrO0RmLALKzF3eZr+zY0L70yyDEPQgdHyGDwEpiDRo0djyFTsnKkU
+        mrR67CONtKcH7WKozltfpf2YV9ATBme1PYr9NCtBTeM8vwga5h67Dv3TBC4ATuOBkH/QiirWNJyD
+        qKnBaMahisDOD9obNGEUw9T15JFjwuUqm9A01+n1xvkOL98vtibFzd2PAsamW6fV6FJkJ2biskOC
+        XV/1Yg/Zqhy/fbQqLVqSrQMezPm4Y1r5WZC2V/e5Wd/8j784+lm2QtVSfUnMRulT9r9nX2xfI16r
+        O0/srdLsGM2FLMvZrRimMdXsUdGSKUyz6oixRsbhodPi9BcAAP//AwBg/xKFcwUAAA==
+    headers:
+      CF-RAY:
+      - 9380fee08936c5a6-IAD
+      Connection:
+      - keep-alive
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json
+      Date:
+      - Tue, 29 Apr 2025 18:52:18 GMT
+      Server:
+      - cloudflare
+      Set-Cookie:
+      - __cf_bm=ZBZ_gGawWFZR6g_amHJEvmq6eTrwHn.6T0arKW9gVW0-1745952738-1.0.1.1-P.TmPJmOrF49vIO_.9StvI17Ray1uycDvUw73C88oGotVOmLTyPooIHWvf0v2Q.CuZw8REvaclx9qcKBnHXVhxGr_2MsX2_o8TMbSVj4Wb8;
+        path=/; expires=Tue, 29-Apr-25 19:22:18 GMT; domain=.api.openai.com; HttpOnly;
+        Secure; SameSite=None
+      - _cfuvid=EOcZHViEbTa_tPSVA8uhDzehfVRpUdeTnoMPw8C9f3I-1745952738788-0.0.1.1-604800000;
+        path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      openai-organization:
+      - datadog-staging
+      openai-processing-ms:
+      - '1371'
+      openai-version:
+      - '2020-10-01'
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      x-request-id:
+      - req_0320a94becf7871df72beab48c3ecc6b
+    status:
+      code: 200
+      message: OK
+version: 1

tests/contrib/openai/test_openai_llmobs.py

@@ -682,6 +682,35 @@ def test_deepseek_as_provider(self, openai, mock_llmobs_writer, mock_tracer):
         assert span_event["meta"]["model_provider"] == "deepseek"
         assert span_event["meta"]["model_name"] == "deepseek-chat"
 
+    @mock.patch("openai._base_client.SyncAPIClient.post")
+    def test_response_completion_proxy(
+        self, mock_completions_post, openai, ddtrace_global_config, mock_llmobs_writer, mock_tracer
+    ):
+        """Ensure llmobs records are not emitted for response endpoints when the base_url is specified."""
+        model = "gpt-4.1"
+        input_messages = multi_message_input
+        client = openai.OpenAI(base_url="http://0.0.0.0:4000")
+        client.responses.create(
+            model=model, input=input_messages, top_p=0.9, max_output_tokens=100, user="ddtrace-test"
+        )
+        assert mock_llmobs_writer.enqueue.call_count == 0
+
+    @pytest.mark.snapshot(token="tests.contrib.openai.test_openai_llmobs.test_response_completion")
+    def test_response_completion(self, openai, ddtrace_global_config, mock_llmobs_writer, mock_tracer):
+        """Ensure llmobs records are emitted for response completion endpoints when configured."""
+        # Create a new cassette for this test
+        with get_openai_vcr(subdirectory_name="v1").use_cassette("response_create.yaml"):
+            model = "gpt-4.1"
+            input_messages = multi_message_input
+            client = openai.OpenAI()
+            client.responses.create(
+                model=model, input=input_messages, top_p=0.9, max_output_tokens=100, user="ddtrace-test"
+            )
+        span = mock_tracer.pop_traces()[0][0]
+        assert span.name == "openai.request"
+        assert span.resource == "createResponseCompletion"
+        assert span.get_tag("openai.request.model") == "gpt-4.1"
+
 
 @pytest.mark.parametrize(
     "ddtrace_global_config",

tests/snapshots/tests.contrib.openai.test_openai_llmobs.test_response_completion.json

@@ -0,0 +1,44 @@
+[[
+  {
+    "name": "openai.request",
+    "service": "tests.contrib.openai",
+    "resource": "createResponseCompletion",
+    "trace_id": 0,
+    "span_id": 1,
+    "parent_id": 0,
+    "type": "llm",
+    "error": 0,
+    "meta": {
+      "_dd.p.dm": "-0",
+      "_dd.p.tid": "68112ff900000000",
+      "component": "openai",
+      "language": "python",
+      "openai.base_url": "https://api.openai.com/v1/",
+      "openai.organization.name": "datadog-staging",
+      "openai.request.client": "OpenAI",
+      "openai.request.endpoint": "/v1/responses",
+      "openai.request.max_output_tokens": "100",
+      "openai.request.method": "POST",
+      "openai.request.model": "gpt-4.1",
+      "openai.request.top_p": "0.9",
+      "openai.request.user": "ddtrace-test",
+      "openai.response.created_at": "1745952737.0",
+      "openai.response.id": "resp_68111fe174308191875e7af1b7f399cc0c2a24124645ef31",
+      "openai.response.model": "gpt-4.1-2025-04-14",
+      "openai.response.tools": "[]",
+      "openai.user.api_key": "sk-...key>",
+      "runtime-id": "a052b5091f1b4828ac710e8760bee8e5"
+    },
+    "metrics": {
+      "_dd.measured": 1,
+      "_dd.top_level": 1,
+      "_dd.tracer_kr": 1.0,
+      "_sampling_priority_v1": 1,
+      "openai.response.usage.input_tokens": 53,
+      "openai.response.usage.output_tokens": 46,
+      "openai.response.usage.total_tokens": 99,
+      "process_id": 8758
+    },
+    "duration": 27859000,
+    "start": 1745956857134697000
+  }]]