openai: handle message with developer role (elastic#58)

Send an system event with an explicit role attribute set to developer in the body

Author: xrmx

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

@@ -255,7 +255,9 @@ def _send_log_events_from_messages(
             content = message.get("content")
             if content:
                 body["content"] = content
-        if message["role"] == "system":
+        if message["role"] == "system" or message["role"] == "developer":
+            if message["role"] == "developer":
+                body["role"] = message["role"]
             event = Event(name=EVENT_GEN_AI_SYSTEM_MESSAGE, body=body, attributes=attributes)
             event_logger.emit(event)
         elif message["role"] == "user":

instrumentation/elastic-opentelemetry-instrumentation-openai/tests/cassettes/test_chat_with_developer_role_message.yaml

@@ -0,0 +1,138 @@
+interactions:
+- request:
+    body: |-
+      {
+        "messages": [
+          {
+            "role": "developer",
+            "content": "You are a friendly assistant"
+          },
+          {
+            "role": "user",
+            "content": "Answer in up to 3 words: Which ocean contains Bouvet Island?"
+          }
+        ],
+        "model": "gpt-4o-mini"
+      }
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      authorization:
+      - Bearer test_openai_api_key
+      connection:
+      - keep-alive
+      content-length:
+      - '197'
+      content-type:
+      - application/json
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.54.4
+      x-stainless-arch:
+      - x64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - Linux
+      x-stainless-package-version:
+      - 1.54.4
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.12.8
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    body:
+      string: |-
+        {
+          "id": "chatcmpl-B6vdHtqgT6rj4cj7itn9bNlaUlqHg",
+          "object": "chat.completion",
+          "created": 1740991207,
+          "model": "gpt-4o-mini-2024-07-18",
+          "choices": [
+            {
+              "index": 0,
+              "message": {
+                "role": "assistant",
+                "content": "Atlantic Ocean.",
+                "refusal": null
+              },
+              "logprobs": null,
+              "finish_reason": "stop"
+            }
+          ],
+          "usage": {
+            "prompt_tokens": 31,
+            "completion_tokens": 4,
+            "total_tokens": 35,
+            "prompt_tokens_details": {
+              "cached_tokens": 0,
+              "audio_tokens": 0
+            },
+            "completion_tokens_details": {
+              "reasoning_tokens": 0,
+              "audio_tokens": 0,
+              "accepted_prediction_tokens": 0,
+              "rejected_prediction_tokens": 0
+            }
+          },
+          "service_tier": "default",
+          "system_fingerprint": "fp_06737a9306"
+        }
+    headers:
+      CF-RAY:
+      - 91a7d3c46d41eda3-MXP
+      Connection:
+      - keep-alive
+      Content-Type:
+      - application/json
+      Date:
+      - Mon, 03 Mar 2025 08:40:08 GMT
+      Server:
+      - cloudflare
+      Set-Cookie: test_set_cookie
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      access-control-expose-headers:
+      - X-Request-ID
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      content-length:
+      - '794'
+      openai-organization: test_openai_org_id
+      openai-processing-ms:
+      - '169'
+      openai-version:
+      - '2020-10-01'
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      x-ratelimit-limit-requests:
+      - '10000'
+      x-ratelimit-limit-tokens:
+      - '200000'
+      x-ratelimit-remaining-requests:
+      - '9999'
+      x-ratelimit-remaining-tokens:
+      - '199960'
+      x-ratelimit-reset-requests:
+      - 8.64s
+      x-ratelimit-reset-tokens:
+      - 12ms
+      x-request-id:
+      - req_70b661026c20278c4d79838fbba06bde
+    status:
+      code: 200
+      message: OK
+version: 1

instrumentation/elastic-opentelemetry-instrumentation-openai/tests/test_chat_completions.py

@@ -130,6 +130,78 @@ def test_chat(default_openai_env, trace_exporter, metrics_reader, logs_exporter)
     )
 
 
+@pytest.mark.vcr()
+def test_chat_with_developer_role_message(default_openai_env, trace_exporter, metrics_reader, logs_exporter):
+    client = openai.OpenAI()
+
+    messages = [
+        {
+            "role": "developer",
+            "content": "You are a friendly assistant",
+        },
+        {
+            "role": "user",
+            "content": TEST_CHAT_INPUT,
+        },
+    ]
+
+    chat_completion = client.chat.completions.create(model=TEST_CHAT_MODEL, messages=messages)
+
+    assert chat_completion.choices[0].message.content == "Atlantic Ocean."
+
+    spans = trace_exporter.get_finished_spans()
+    assert len(spans) == 1
+
+    span = spans[0]
+    assert span.name == f"chat {TEST_CHAT_MODEL}"
+    assert span.kind == SpanKind.CLIENT
+    assert span.status.status_code == StatusCode.UNSET
+
+    address, port = address_and_port(client)
+    assert dict(span.attributes) == {
+        GEN_AI_OPENAI_RESPONSE_SERVICE_TIER: "default",
+        GEN_AI_OPERATION_NAME: "chat",
+        GEN_AI_REQUEST_MODEL: TEST_CHAT_MODEL,
+        GEN_AI_SYSTEM: "openai",
+        GEN_AI_RESPONSE_ID: "chatcmpl-B6vdHtqgT6rj4cj7itn9bNlaUlqHg",
+        GEN_AI_RESPONSE_MODEL: TEST_CHAT_RESPONSE_MODEL,
+        GEN_AI_RESPONSE_FINISH_REASONS: ("stop",),
+        GEN_AI_USAGE_INPUT_TOKENS: 31,
+        GEN_AI_USAGE_OUTPUT_TOKENS: 4,
+        SERVER_ADDRESS: address,
+        SERVER_PORT: port,
+    }
+
+    logs = logs_exporter.get_finished_logs()
+    assert len(logs) == 3
+    log_records = logrecords_from_logs(logs)
+    system_message, user_message, choice = log_records
+    assert dict(system_message.attributes) == {"gen_ai.system": "openai", "event.name": "gen_ai.system.message"}
+    assert dict(system_message.body) == {"role": "developer"}
+
+    assert dict(user_message.attributes) == {"gen_ai.system": "openai", "event.name": "gen_ai.user.message"}
+    assert dict(user_message.body) == {}
+
+    assert_stop_log_record(choice)
+
+    operation_duration_metric, token_usage_metric = get_sorted_metrics(metrics_reader)
+    attributes = {
+        GEN_AI_REQUEST_MODEL: TEST_CHAT_MODEL,
+        GEN_AI_RESPONSE_MODEL: "gpt-4o-mini-2024-07-18",
+    }
+    assert_operation_duration_metric(
+        client, "chat", operation_duration_metric, attributes=attributes, min_data_point=0.006761051714420319
+    )
+    assert_token_usage_metric(
+        client,
+        "chat",
+        token_usage_metric,
+        attributes=attributes,
+        input_data_point=span.attributes[GEN_AI_USAGE_INPUT_TOKENS],
+        output_data_point=span.attributes[GEN_AI_USAGE_OUTPUT_TOKENS],
+    )
+
+
 @pytest.mark.skipif(OPENAI_VERSION < (1, 35, 0), reason="service tier added in 1.35.0")
 @pytest.mark.vcr()
 def test_chat_all_the_client_options(default_openai_env, trace_exporter, metrics_reader, logs_exporter):