feat(google-generativeai): Add metrics support

packages/opentelemetry-instrumentation-google-generativeai/opentelemetry/instrumentation/google_generativeai/__init__.py

@@ -1,6 +1,8 @@
 """OpenTelemetry Google Generative AI API instrumentation"""
 
 import logging
+import os
+import time
 import types
 from typing import Collection
 
@@ -31,8 +33,9 @@
 from opentelemetry.semconv_ai import (
     SUPPRESS_LANGUAGE_MODEL_INSTRUMENTATION_KEY,
     LLMRequestTypeValues,
-    SpanAttributes,
+    SpanAttributes
 )
+from opentelemetry.metrics import Meter, get_meter
 from opentelemetry.trace import SpanKind, get_tracer
 from wrapt import wrap_function_wrapper
 
@@ -79,6 +82,7 @@ def _build_from_streaming_response(
     response: GenerateContentResponse,
     llm_model,
     event_logger,
+    token_histogram,
 ):
     complete_response = ""
     last_chunk = None
@@ -93,12 +97,12 @@ def _build_from_streaming_response(
         emit_choice_events(response, event_logger)
     else:
         set_response_attributes(span, complete_response, llm_model)
-    set_model_response_attributes(span, last_chunk or response, llm_model)
+    set_model_response_attributes(span, last_chunk or response, llm_model, token_histogram)
     span.end()
 
 
 async def _abuild_from_streaming_response(
-    span, response: GenerateContentResponse, llm_model, event_logger
+    span, response: GenerateContentResponse, llm_model, event_logger, token_histogram
 ):
     complete_response = ""
     last_chunk = None
@@ -113,7 +117,7 @@ async def _abuild_from_streaming_response(
         emit_choice_events(response, event_logger)
     else:
         set_response_attributes(span, complete_response, llm_model)
-    set_model_response_attributes(span, last_chunk if last_chunk else response, llm_model)
+    set_model_response_attributes(span, last_chunk if last_chunk else response, llm_model, token_histogram)
     span.end()
 
 
@@ -128,21 +132,31 @@ def _handle_request(span, args, kwargs, llm_model, event_logger):
 
 
 @dont_throw
-def _handle_response(span, response, llm_model, event_logger):
+def _handle_response(span, response, llm_model, event_logger, token_histogram):
     if should_emit_events() and event_logger:
         emit_choice_events(response, event_logger)
     else:
         set_response_attributes(span, response, llm_model)
 
-    set_model_response_attributes(span, response, llm_model)
+    set_model_response_attributes(span, response, llm_model, token_histogram)
 
 
 def _with_tracer_wrapper(func):
     """Helper for providing tracer for wrapper functions."""
 
-    def _with_tracer(tracer, event_logger, to_wrap):
+    def _with_tracer(tracer, event_logger, to_wrap, token_histogram, duration_histogram):
         def wrapper(wrapped, instance, args, kwargs):
-            return func(tracer, event_logger, to_wrap, wrapped, instance, args, kwargs)
+            return func(
+                tracer,
+                event_logger,
+                to_wrap,
+                token_histogram,
+                duration_histogram,
+                wrapped,
+                instance,
+                args,
+                kwargs,
+            )
 
         return wrapper
 
@@ -154,10 +168,13 @@ async def _awrap(
     tracer,
     event_logger,
     to_wrap,
+    token_histogram,
+    duration_histogram,
     wrapped,
     instance,
     args,
     kwargs,
+
 ):
     """Instruments and calls every function defined in TO_WRAP."""
     if context_api.get_value(_SUPPRESS_INSTRUMENTATION_KEY) or context_api.get_value(
@@ -186,22 +203,31 @@ async def _awrap(
             SpanAttributes.LLM_REQUEST_TYPE: LLMRequestTypeValues.COMPLETION.value,
         },
     )
-
+    start_time = time.perf_counter()
     _handle_request(span, args, kwargs, llm_model, event_logger)
 
     response = await wrapped(*args, **kwargs)
 
+    if duration_histogram:
+        duration = time.perf_counter() - start_time
+        duration_histogram.record(
+            duration,
+            attributes={
+                "gen_ai.provider.name": "Google",
+                "gen_ai.response.model": llm_model
+            },
+        )
     if response:
         if is_streaming_response(response):
             return _build_from_streaming_response(
-                span, response, llm_model, event_logger
+                span, response, llm_model, event_logger, token_histogram
             )
         elif is_async_streaming_response(response):
             return _abuild_from_streaming_response(
-                span, response, llm_model, event_logger
+                span, response, llm_model, event_logger, token_histogram
             )
         else:
-            _handle_response(span, response, llm_model, event_logger)
+            _handle_response(span, response, llm_model, event_logger, token_histogram)
 
     span.end()
     return response
@@ -212,10 +238,13 @@ def _wrap(
     tracer,
     event_logger,
     to_wrap,
+    token_histogram,
+    duration_histogram,
     wrapped,
     instance,
     args,
     kwargs,
+
 ):
     """Instruments and calls every function defined in TO_WRAP."""
     if context_api.get_value(_SUPPRESS_INSTRUMENTATION_KEY) or context_api.get_value(
@@ -245,26 +274,56 @@ def _wrap(
         },
     )
 
+    start_time = time.perf_counter()
     _handle_request(span, args, kwargs, llm_model, event_logger)
 
     response = wrapped(*args, **kwargs)
 
+    if duration_histogram:
+        duration = time.perf_counter() - start_time
+        duration_histogram.record(
+            duration,
+            attributes={
+                "gen_ai.provider.name": "Google",
+                "gen_ai.response.model": llm_model
+            },
+        )
     if response:
         if is_streaming_response(response):
             return _build_from_streaming_response(
-                span, response, llm_model, event_logger
+                span, response, llm_model, event_logger, token_histogram
             )
         elif is_async_streaming_response(response):
             return _abuild_from_streaming_response(
-                span, response, llm_model, event_logger
+                span, response, llm_model, event_logger, token_histogram
             )
         else:
-            _handle_response(span, response, llm_model, event_logger)
+            _handle_response(span, response, llm_model, event_logger, token_histogram)
 
     span.end()
     return response
 
 
+def is_metrics_enabled() -> bool:
+    return (os.getenv("TRACELOOP_METRICS_ENABLED") or "true").lower() == "true"
+
+
+def _create_metrics(meter: Meter):
+    token_histogram = meter.create_histogram(
+        name="gen_ai.client.token.usage",
+        unit="token",
+        description="Measures number of input and output tokens used",
+    )
+
+    duration_histogram = meter.create_histogram(
+        name="gen_ai.client.operation.duration",
+        unit="s",
+        description="GenAI operation duration",
+    )
+
+    return token_histogram, duration_histogram
+
+
 class GoogleGenerativeAiInstrumentor(BaseInstrumentor):
     """An instrumentor for Google Generative AI's client library."""
 
@@ -285,6 +344,15 @@ def _instrument(self, **kwargs):
         tracer_provider = kwargs.get("tracer_provider")
         tracer = get_tracer(__name__, __version__, tracer_provider)
 
+        meter_provider = kwargs.get("meter_provider")
+        meter = get_meter(__name__, __version__, meter_provider)
+
+        token_histogram = None
+        duration_histogram = None
+
+        if is_metrics_enabled():
+            token_histogram, duration_histogram = _create_metrics(meter)
+
         event_logger = None
         if not Config.use_legacy_attributes:
             logger_provider = kwargs.get("logger_provider")
@@ -297,14 +365,24 @@ def _instrument(self, **kwargs):
             wrap_object = wrapped_method.get("object")
             wrap_method = wrapped_method.get("method")
 
+            wrapper_args = (
+                tracer,
+                event_logger,
+                wrapped_method,
+                token_histogram,
+                duration_histogram,
+            )
+
+            wrapper = (
+                _awrap(*wrapper_args)
+                if wrap_object == "AsyncModels"
+                else _wrap(*wrapper_args)
+            )
+
             wrap_function_wrapper(
                 wrap_package,
                 f"{wrap_object}.{wrap_method}",
-                (
-                    _awrap(tracer, event_logger, wrapped_method)
-                    if wrap_object == "AsyncModels"
-                    else _wrap(tracer, event_logger, wrapped_method)
-                ),
+                wrapper,
             )
 
     def _uninstrument(self, **kwargs):

packages/opentelemetry-instrumentation-google-generativeai/opentelemetry/instrumentation/google_generativeai/span_utils.py

@@ -446,7 +446,7 @@ def set_response_attributes(span, response, llm_model):
             )
 
 
-def set_model_response_attributes(span, response, llm_model):
+def set_model_response_attributes(span, response, llm_model, token_histogram):
     if not span.is_recording():
         return
 
@@ -469,4 +469,22 @@ def set_model_response_attributes(span, response, llm_model):
             response.usage_metadata.prompt_token_count,
         )
 
+    if token_histogram and hasattr(response, "usage_metadata"):
+        token_histogram.record(
+            response.usage_metadata.prompt_token_count,
+            attributes={
+                "gen_ai.provider.name": "Google",
+                "gen_ai.token.type": "input",
+                "gen_ai.response.model": llm_model,
+            }
+        )
+        token_histogram.record(
+                    response.usage_metadata.candidates_token_count,
+                    attributes={
+                        "gen_ai.provider.name": "Google",
+                        "gen_ai.token.type": "output",
+                        "gen_ai.response.model": llm_model,
+                    },
+                )
+
     span.set_status(Status(StatusCode.OK))

packages/opentelemetry-instrumentation-google-generativeai/tests/cassettes/test_generate_content/test_client_spans.yaml

@@ -0,0 +1,94 @@
+interactions:
+- request:
+    body: '{"contents": [{"parts": [{"text": "What is ai?"}], "role": "user"}]}'
+    headers:
+      Accept:
+      - '*/*'
+      Accept-Encoding:
+      - gzip, deflate, zstd
+      Connection:
+      - keep-alive
+      Content-Length:
+      - '68'
+      Content-Type:
+      - application/json
+      Host:
+      - generativelanguage.googleapis.com
+      user-agent:
+      - google-genai-sdk/1.52.0 gl-python/3.11.11
+      x-goog-api-client:
+      - google-genai-sdk/1.52.0 gl-python/3.11.11
+      x-goog-api-key:
+      - DUMMY_GOOGLE_API_KEY
+    method: POST
+    uri: https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent
+  response:
+    body:
+      string: !!binary |
+        H4sIAAAAAAAC/5VY73PbNhL97r8Coy+1NZLGyTXT1l9uNLbP9dVOPJUvuenlPkDkSkREAiwAymYy
+        +d/7dkFSlK8/5jKdVCHAxe7bt28X/HKi1CTTNje5jhQmF+o/eKLUF/mb15yNZCMW+kd4WGsfD3vT
+        ny+j39gS6ZlfmixvVYiwH9TGeTWdLn00G5MZXapbWC5LsyWb0XS6+Gg/2mVUJgaVOU8zhVdNUFpt
+        DJW5chs8ruomklchM/wSTGZNIKxZlXnS0ditqnRWGEtBxUJHhdBUTR5nVyrqsOsex7Y2mS7LVnn6
+        tTGeVNFU2GpGLi3UY4Hzjc3KJhd7MB9UaXZ0wb5OEeZ0ekfaWyxcTKdqmbEtdsJYPhEOwTMEr3xT
+        UkKgCbIeF72Bn0kH11v4l6ym3dHBOQSjdF1792xgjhQs5LQx1uA3UgPXAs4Ig7UH79YlVfPgyn1n
+        8x8G2YVVPGrYIbHMUJb0rLICKJDd0sgE+Yxq3ike2Zy8ZJBtcCyMka89CdqBbHC+PQr4lCFS8BdW
+        2eFQE2XF2XDAnbbbBouqGdvmwy7hlaeCDochE+RTYlOGyu7lwdoVZYYxmFd615spnBMgY0FqTQE0
+        cI0PxBzSGfsoZJtOf3RPKnfwElR7cn6nTo1VwTA0CkFW4ezv06nwEixuQ6QK+eeEq413ldprmNaV
+        ayw4C9soIT1TJke9mE2rah1hxIaZRAJfLLIPw4XD3/0iJwOe44Gn3GQpQZLlFFZYoE5wspYKWBOH
+        Rc91iRKKoC+YsfW6qlAEzC7aE5JRuxAMaIA6Iau9cTM+vVUaPAeTzdZiO85NobBzOtd17ED5CTsv
+        QS1wIMjiY1uTxLe8vUhwvFow8m+19wAQ2Jzi6A+kd/h9JltYAlJ6PnC5GfwXODOpooIkpnKSmKoC
+        ZVCPlE5QT6RQfYwpCj26XLdAIH4TDp4Lml4b2wWtmWEZq4qU+GJ8/PWz5mzK4e+dgWZoYMOcQ3Cn
+        K1QrdAaFoM9mqDb2BuRLNEZViI6cvqW4Kc0zNlb6s7PYGWpdQZZKuIjsBio389wbLjgIjscrWeM9
+        WIAEuZr5y8cq2+PF6XDNtuirEBazgkKY16Vu2UiX1XAUymWhPdhLHu6bjAO6Tfq2JiVp3zoHcJKC
+        HiEyA7QQcRUaX3P4QzFlutZrU5poCHGsG8kUF4T9JgqL4FRXguWRNjJDIT/QYdcEgBSEO6+FFiOF
+        v+leHSu9Ol3e3PL7q+gdPPkryixV0dYOhIms2GOmiJJ3XiJPEtK8RAmU8G0LiTR74N7Hh8r32Au2
+        PKFClYeOiCHpCkwxKECDPrPmd7j+u/qYjWQqVTL0uGyP4cBmbftnlMWGPYXh5KMe4LZ/SM53KUNE
+        3IQODS4pgjqlxXYxUz8u79QP5+fnSX5en5+/ugA+q1pj67u8BQztTF1BhdKGVdRePXranf01kUQG
+        rYsQFzxWLcUk9gxYaNafEBSjha4OxWdgPQXAw+0J2+rClC44/I9zlNMalBdK/O0lJVYNCsIc82H1
+        /1GgT33KV0fq32ng4tmY4ozs3njODhKYtFKGi1nX5aWAZZDYmwgo6+NuOhu1pFLtrHsqKZdW9Gfg
+        Xu+R003j4b9nD1j7Nk1sPLcCuIxySIYhacxjVjlGCxKlKCZEeVkSw60F//61QVfrGn+v2atmLcF0
+        oo2Waw3vA995dDmod3L0Ps1Jqh9g1On93VlCm8cvTnqg2IsyfKffa4Hc8uQA18ShLw1dCWaPwLki
+        qkcHXvUHHs66v0u5bRgJSw0DbSlyc06BoFmi0KCTxDprbKgxvaFqW8E1kWDNzeHs0OH6YefQkEn0
+        UWgAZCSI0dACSUf/6aqdB9lFX4AbLSTmTiEC4yAPVkdxs59MOPqMRGYPM8/bbtMw+zwMm9Bg7h4E
+        iGvLyhOGOVeGg7H6DKPXmImRXoxGvbOY7eLaRYQzrKBt2lDq5HhgMqHfcaXrskVbPPh72Y/a72UK
+        +UPvPk4gWR8nx4Mhiiw00jEOQ6GwpZsJefMeIuxC76pLApNTpCw5V/E0xDbwShoGX7r4s0NwJhOF
+        uLV7lCilwSINCjNpUNE3ncHUiOUnn9+kcdC7hFBqkFAB52vXDZzgPY8XVu/NtnsTMJu6KUd2JOhh
+        qOx1HWdFtUrlIupeIolqNEX2JJdZf77WfIvx/U1AZkJTcfeWcI4G3K6jtTLMdpknOXA4/6YbmqHS
+        UvSosJdtk+eGgT6WnlR3zwMrGtb0ILzvRnloWrx5eMSY0iUwPb5a3t3Nr7mV35v8EyZsSy32aOgo
+        Jk48lhwfsOWWUSMnqV1y9J5rpRomzw9Fyw2HYXI4CongOe0wghc6JRi9oNNBuSPtu4uN+cyhaD9o
+        Oy4+DRjQpq7NJR/TmG6wpTQbXA7aiwTasomuSmn3VKPRCHppNuh2MAE/8wYZ+1kzoFldiWZuL+IO
+        4hWxfwMXKeRTl+ZzN7SjKXBf6tevupFRmr00SJWbIKbMsOkWNyIn22jDLRQGUjwvaCEQ4j7Dkc+4
+        deoSIK693FkHLIeGwrWB7KSKCIesr40O3PfMXmdogJ/cmn3CXJpRJfTorjIAupu+u6TKfSPn4cvV
+        vHMxGX0P+Dr8/u/s8BXBu5L4E0Hlcir77V/7DRO+5YYiXY952+rx3cNkWEVu6RmPz0/6A8T0pAlg
+        6D1FzQkavlpMEs8e3Y7sJd8tsPIm2Rp9/Tha/v6777oN0UVdHq29+uHb89n/GA5XONaU4w8jo28m
+        iFJz3XIoj9f/fpyMkIhHfvVInIwAm6QbQzz28dX5m+9POswSjO/BOZPw2hL6r5m/XryZb0odCjlw
+        0tfgbc575sXuW715t7r/5afdp+b+wf7zw81uuZucfD35DXc/SmwaEgAA
+    headers:
+      Alt-Svc:
+      - h3=":443"; ma=2592000,h3-29=":443"; ma=2592000
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json; charset=UTF-8
+      Date:
+      - Tue, 09 Dec 2025 12:45:46 GMT
+      Server:
+      - scaffolding on HTTPServer2
+      Server-Timing:
+      - gfet4t7; dur=11009
+      Transfer-Encoding:
+      - chunked
+      Vary:
+      - Origin
+      - X-Origin
+      - Referer
+      X-Content-Type-Options:
+      - nosniff
+      X-Frame-Options:
+      - SAMEORIGIN
+      X-XSS-Protection:
+      - '0'
+    status:
+      code: 200
+      message: OK
+version: 1