opentelemetry-instrumentation-openai 0.16.4__tar.gz → 0.16.5__tar.gz

{opentelemetry_instrumentation_openai-0.16.4 → opentelemetry_instrumentation_openai-0.16.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: opentelemetry-instrumentation-openai
-Version: 0.16.4
+Version: 0.16.5
 Summary: OpenTelemetry OpenAI instrumentation
 Home-page: https://github.com/traceloop/openllmetry/tree/main/packages/opentelemetry-instrumentation-openai
 License: Apache-2.0

{opentelemetry_instrumentation_openai-0.16.4 → opentelemetry_instrumentation_openai-0.16.5}/opentelemetry/instrumentation/openai/shared/embeddings_wrappers.py

@@ -7,7 +7,6 @@ from opentelemetry.semconv.ai import SpanAttributes, LLMRequestTypeValues
 
 from opentelemetry.instrumentation.utils import _SUPPRESS_INSTRUMENTATION_KEY
 from opentelemetry.instrumentation.openai.utils import (
-    _with_tracer_wrapper,
     start_as_current_span_async,
     _with_embeddings_telemetry_wrapper,
 )
@@ -33,12 +32,17 @@ logger = logging.getLogger(__name__)
 
 
 @_with_embeddings_telemetry_wrapper
-def embeddings_wrapper(tracer,
-                       token_counter: Counter,
-                       vector_size_counter: Counter,
-                       duration_histogram: Histogram,
-                       exception_counter: Counter,
-                       wrapped, instance, args, kwargs):
+def embeddings_wrapper(
+    tracer,
+    token_counter: Counter,
+    vector_size_counter: Counter,
+    duration_histogram: Histogram,
+    exception_counter: Counter,
+    wrapped,
+    instance,
+    args,
+    kwargs,
+):
     if context_api.get_value(_SUPPRESS_INSTRUMENTATION_KEY):
         return wrapped(*args, **kwargs)
 
@@ -56,7 +60,7 @@ def embeddings_wrapper(tracer,
             end_time = time.time()
         except Exception as e:  # pylint: disable=broad-except
             end_time = time.time()
-            duration = end_time - start_time if 'start_time' in locals() else 0
+            duration = end_time - start_time if "start_time" in locals() else 0
             attributes = {
                 "error.type": e.__class__.__name__,
             }
@@ -71,25 +75,71 @@ def embeddings_wrapper(tracer,
 
         duration = end_time - start_time
 
-        _handle_response(response, span, instance, token_counter, vector_size_counter, duration_histogram, duration)
+        _handle_response(
+            response,
+            span,
+            instance,
+            token_counter,
+            vector_size_counter,
+            duration_histogram,
+            duration,
+        )
 
         return response
 
 
-@_with_tracer_wrapper
-async def aembeddings_wrapper(tracer, wrapped, instance, args, kwargs):
+@_with_embeddings_telemetry_wrapper
+async def aembeddings_wrapper(
+    tracer,
+    token_counter: Counter,
+    vector_size_counter: Counter,
+    duration_histogram: Histogram,
+    exception_counter: Counter,
+    wrapped,
+    instance,
+    args,
+    kwargs,
+):
     if context_api.get_value(_SUPPRESS_INSTRUMENTATION_KEY):
         return wrapped(*args, **kwargs)
 
     async with start_as_current_span_async(
-            tracer=tracer,
-            name=SPAN_NAME,
-            kind=SpanKind.CLIENT,
-            attributes={SpanAttributes.LLM_REQUEST_TYPE: LLM_REQUEST_TYPE.value},
+        tracer=tracer,
+        name=SPAN_NAME,
+        kind=SpanKind.CLIENT,
+        attributes={SpanAttributes.LLM_REQUEST_TYPE: LLM_REQUEST_TYPE.value},
     ) as span:
         _handle_request(span, kwargs, instance)
-        response = await wrapped(*args, **kwargs)
-        _handle_response(response, span)
+        try:
+            # record time for duration
+            start_time = time.time()
+            response = wrapped(*args, **kwargs)
+            end_time = time.time()
+        except Exception as e:  # pylint: disable=broad-except
+            end_time = time.time()
+            duration = end_time - start_time if "start_time" in locals() else 0
+            attributes = {
+                "error.type": e.__class__.__name__,
+            }
+
+            # if there are legal duration, record it
+            if duration > 0 and duration_histogram:
+                duration_histogram.record(duration, attributes=attributes)
+            if exception_counter:
+                exception_counter.add(1, attributes=attributes)
+
+            raise e
+
+        duration = end_time - start_time
+        _handle_response(
+            response,
+            span,
+            instance,
+            token_counter,
+            vector_size_counter,
+            duration_histogram,
+            duration,
+        )
 
         return response
 
@@ -101,19 +151,40 @@ def _handle_request(span, kwargs, instance):
     _set_client_attributes(span, instance)
 
 
-def _handle_response(response, span, instance=None, token_counter=None, vector_size_counter=None,
-                     duration_histogram=None, duration=None):
+def _handle_response(
+    response,
+    span,
+    instance=None,
+    token_counter=None,
+    vector_size_counter=None,
+    duration_histogram=None,
+    duration=None,
+):
     if is_openai_v1():
         response_dict = model_as_dict(response)
     else:
         response_dict = response
     # metrics record
-    _set_embeddings_metrics(instance, token_counter, vector_size_counter, duration_histogram, response_dict, duration)
+    _set_embeddings_metrics(
+        instance,
+        token_counter,
+        vector_size_counter,
+        duration_histogram,
+        response_dict,
+        duration,
+    )
     # span attributes
     _set_response_attributes(span, response_dict)
 
 
-def _set_embeddings_metrics(instance, token_counter, vector_size_counter, duration_histogram, response_dict, duration):
+def _set_embeddings_metrics(
+    instance,
+    token_counter,
+    vector_size_counter,
+    duration_histogram,
+    response_dict,
+    duration,
+):
     shared_attributes = {
         "llm.response.model": response_dict.get("model") or None,
         "server.address": _get_openai_base_url(instance),
@@ -124,7 +195,10 @@ def _set_embeddings_metrics(instance, token_counter, vector_size_counter, durati
     if usage and token_counter:
         for name, val in usage.items():
             if name in OPENAI_LLM_USAGE_TOKEN_TYPES:
-                attributes_with_token_type = {**shared_attributes, "llm.usage.token_type": name.split('_')[0]}
+                attributes_with_token_type = {
+                    **shared_attributes,
+                    "llm.usage.token_type": name.split("_")[0],
+                }
                 token_counter.add(val, attributes=attributes_with_token_type)
 
     # vec size metrics

{opentelemetry_instrumentation_openai-0.16.4 → opentelemetry_instrumentation_openai-0.16.5}/opentelemetry/instrumentation/openai/v1/__init__.py

@@ -178,7 +178,13 @@ class OpenAIV1Instrumentor(BaseInstrumentor):
         wrap_function_wrapper(
             "openai.resources.embeddings",
             "AsyncEmbeddings.create",
-            aembeddings_wrapper(tracer),
+            aembeddings_wrapper(
+                tracer,
+                embeddings_token_counter,
+                embeddings_vector_size_counter,
+                embeddings_duration_histogram,
+                embeddings_exception_counter,
+            ),
         )
 
         if is_metrics_enabled():

opentelemetry_instrumentation_openai-0.16.5/opentelemetry/instrumentation/openai/version.py

@@ -0,0 +1 @@
+__version__ = "0.16.5"

{opentelemetry_instrumentation_openai-0.16.4 → opentelemetry_instrumentation_openai-0.16.5}/pyproject.toml

@@ -8,7 +8,7 @@ show_missing = true
 
 [tool.poetry]
 name = "opentelemetry-instrumentation-openai"
-version = "0.16.4"
+version = "0.16.5"
 description = "OpenTelemetry OpenAI instrumentation"
 authors = [
   "Gal Kleinman <gal@traceloop.com>",

opentelemetry_instrumentation_openai-0.16.4/opentelemetry/instrumentation/openai/version.py

@@ -1 +0,0 @@
-__version__ = "0.16.4"