openai: Use semconv-1.31.0 (#77)

* Use "gen_ai.output.type" instead of deprecated "gen_ai.openai.request.response_format" attribute.

* Use json instead of json_schema and json_object for gen_ai.output.type attribute value

* Bump schema version number

Author: SrdjanLL

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

@@ -87,14 +87,14 @@ def _instrument(self, **kwargs):
             __name__,
             __version__,
             tracer_provider,
-            schema_url=Schemas.V1_28_0.value,
+            schema_url=Schemas.V1_31_0.value,
         )
         meter_provider = kwargs.get("meter_provider")
         self.meter = get_meter(
             __name__,
             __version__,
             meter_provider,
-            schema_url=Schemas.V1_28_0.value,
+            schema_url=Schemas.V1_31_0.value,
         )
         event_logger_provider = kwargs.get("event_logger_provider")
         self.event_logger = get_event_logger(__name__, event_logger_provider)

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

@@ -20,10 +20,10 @@
 
 from opentelemetry._events import Event, EventLogger
 from opentelemetry.semconv._incubating.attributes.gen_ai_attributes import (
-    GEN_AI_OPENAI_REQUEST_RESPONSE_FORMAT,
     GEN_AI_OPENAI_REQUEST_SERVICE_TIER,
     GEN_AI_OPENAI_RESPONSE_SERVICE_TIER,
     GEN_AI_OPERATION_NAME,
+    GEN_AI_OUTPUT_TYPE,
     GEN_AI_REQUEST_CHOICE_COUNT,
     GEN_AI_REQUEST_FREQUENCY_PENALTY,
     GEN_AI_REQUEST_MAX_TOKENS,
@@ -166,13 +166,16 @@ def _is_set(value):
         # response_format may be string or object with a string in the `type` key
         if isinstance(response_format, Mapping):
             if _is_set(response_format_type := response_format.get("type")):
-                span_attributes[GEN_AI_OPENAI_REQUEST_RESPONSE_FORMAT] = response_format_type
+                if response_format_type in ("json_object", "json_schema"):
+                    span_attributes[GEN_AI_OUTPUT_TYPE] = "json"
+                else:
+                    span_attributes[GEN_AI_OUTPUT_TYPE] = response_format_type
         elif isinstance(response_format, str):
-            span_attributes[GEN_AI_OPENAI_REQUEST_RESPONSE_FORMAT] = response_format
+            span_attributes[GEN_AI_OUTPUT_TYPE] = response_format
         else:
             # Assume structured output lazily parsed to a schema via type_to_response_format_param or similar.
             # e.g. pydantic._internal._model_construction.ModelMetaclass
-            span_attributes[GEN_AI_OPENAI_REQUEST_RESPONSE_FORMAT] = "json_schema"
+            span_attributes[GEN_AI_OUTPUT_TYPE] = "json"
 
     return span_attributes
 

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

@@ -28,10 +28,10 @@
 from opentelemetry._logs import LogRecord
 from opentelemetry.instrumentation.openai import OpenAIInstrumentor
 from opentelemetry.semconv._incubating.attributes.gen_ai_attributes import (
-    GEN_AI_OPENAI_REQUEST_RESPONSE_FORMAT,
     GEN_AI_OPENAI_REQUEST_SERVICE_TIER,
     GEN_AI_OPENAI_RESPONSE_SERVICE_TIER,
     GEN_AI_OPERATION_NAME,
+    GEN_AI_OUTPUT_TYPE,
     GEN_AI_REQUEST_CHOICE_COUNT,
     GEN_AI_REQUEST_FREQUENCY_PENALTY,
     GEN_AI_REQUEST_MAX_TOKENS,
@@ -248,7 +248,7 @@ def test_chat_all_the_client_options(default_openai_env, trace_exporter, metrics
     expected_attrs = {
         GEN_AI_REQUEST_SEED: 100,
         GEN_AI_OPENAI_REQUEST_SERVICE_TIER: "default",
-        GEN_AI_OPENAI_REQUEST_RESPONSE_FORMAT: "text",
+        GEN_AI_OUTPUT_TYPE: "text",
         GEN_AI_OPENAI_RESPONSE_SERVICE_TIER: "default",
         GEN_AI_OPERATION_NAME: "chat",
         GEN_AI_REQUEST_FREQUENCY_PENALTY: 0,
@@ -1502,14 +1502,14 @@ def test_chat_exported_schema_version(default_openai_env, trace_exporter, metric
 
     spans = trace_exporter.get_finished_spans()
     (span,) = spans
-    assert span.instrumentation_scope.schema_url == "https://opentelemetry.io/schemas/1.28.0"
+    assert span.instrumentation_scope.schema_url == "https://opentelemetry.io/schemas/1.31.0"
 
     metrics_data = metrics_reader.get_metrics_data()
     resource_metrics = metrics_data.resource_metrics
 
     for metrics in resource_metrics:
         for scope_metrics in metrics.scope_metrics:
-            assert scope_metrics.schema_url == "https://opentelemetry.io/schemas/1.28.0"
+            assert scope_metrics.schema_url == "https://opentelemetry.io/schemas/1.31.0"
 
 
 @pytest.mark.skipif(OPENAI_VERSION < (1, 40, 0), reason="beta completions added in 1.40.0")
@@ -1545,7 +1545,7 @@ def test_parse_response_format_json_object_with_capture_message_content(
     address, port = address_and_port(client)
     assert dict(span.attributes) == {
         GEN_AI_OPENAI_RESPONSE_SERVICE_TIER: "default",
-        GEN_AI_OPENAI_REQUEST_RESPONSE_FORMAT: "json_object",
+        GEN_AI_OUTPUT_TYPE: "json",
         GEN_AI_OPERATION_NAME: "chat",
         GEN_AI_REQUEST_MODEL: TEST_CHAT_MODEL,
         GEN_AI_SYSTEM: "openai",
@@ -1620,7 +1620,7 @@ def test_parse_response_format_structured_output_with_capture_message_content(
     address, port = address_and_port(client)
     assert dict(span.attributes) == {
         GEN_AI_OPENAI_RESPONSE_SERVICE_TIER: "default",
-        GEN_AI_OPENAI_REQUEST_RESPONSE_FORMAT: "json_schema",
+        GEN_AI_OUTPUT_TYPE: "json",
         GEN_AI_OPERATION_NAME: "chat",
         GEN_AI_REQUEST_MODEL: TEST_CHAT_MODEL,
         GEN_AI_SYSTEM: "openai",

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

@@ -27,10 +27,10 @@
 from opentelemetry._logs import LogRecord
 from opentelemetry.instrumentation.openai import OpenAIInstrumentor
 from opentelemetry.semconv._incubating.attributes.gen_ai_attributes import (
-    GEN_AI_OPENAI_REQUEST_RESPONSE_FORMAT,
     GEN_AI_OPENAI_REQUEST_SERVICE_TIER,
     GEN_AI_OPENAI_RESPONSE_SERVICE_TIER,
     GEN_AI_OPERATION_NAME,
+    GEN_AI_OUTPUT_TYPE,
     GEN_AI_REQUEST_CHOICE_COUNT,
     GEN_AI_REQUEST_FREQUENCY_PENALTY,
     GEN_AI_REQUEST_MAX_TOKENS,
@@ -332,7 +332,7 @@ def test_chat_all_the_client_options(default_openai_env, trace_exporter, metrics
     expected_attrs = {
         GEN_AI_REQUEST_SEED: 100,
         GEN_AI_OPENAI_REQUEST_SERVICE_TIER: "default",
-        GEN_AI_OPENAI_REQUEST_RESPONSE_FORMAT: "text",
+        GEN_AI_OUTPUT_TYPE: "text",
         GEN_AI_OPENAI_RESPONSE_SERVICE_TIER: "default",
         GEN_AI_OPERATION_NAME: "chat",
         GEN_AI_REQUEST_FREQUENCY_PENALTY: 0,
@@ -1204,7 +1204,7 @@ def test_chat_stream_all_the_client_options(default_openai_env, trace_exporter,
     address, port = address_and_port(client)
     expected_attrs = {
         GEN_AI_REQUEST_SEED: 100,
-        GEN_AI_OPENAI_REQUEST_RESPONSE_FORMAT: "text",
+        GEN_AI_OUTPUT_TYPE: "text",
         GEN_AI_OPENAI_REQUEST_SERVICE_TIER: "default",
         GEN_AI_OPENAI_RESPONSE_SERVICE_TIER: "default",
         GEN_AI_OPERATION_NAME: "chat",
@@ -2444,14 +2444,14 @@ def test_chat_exported_schema_version(default_openai_env, trace_exporter, metric
 
     spans = trace_exporter.get_finished_spans()
     (span,) = spans
-    assert span.instrumentation_scope.schema_url == "https://opentelemetry.io/schemas/1.28.0"
+    assert span.instrumentation_scope.schema_url == "https://opentelemetry.io/schemas/1.31.0"
 
     metrics_data = metrics_reader.get_metrics_data()
     resource_metrics = metrics_data.resource_metrics
 
     for metrics in resource_metrics:
         for scope_metrics in metrics.scope_metrics:
-            assert scope_metrics.schema_url == "https://opentelemetry.io/schemas/1.28.0"
+            assert scope_metrics.schema_url == "https://opentelemetry.io/schemas/1.31.0"
 
 
 @dataclass