GenAI Utils | Adding Embedding metrics (#4377)

* linting fix

* Updating changelog

* Updating changelog PR number

* Removing embedding events emission

* Refatoring and adding input token metric for Embedding invocation

* Merging and removing unused import

* build(deps): bump aiohttp from 3.13.3 to 3.13.4 (#4386)

---
updated-dependencies:
- dependency-name: aiohttp
  dependency-version: 3.13.4
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* Revert "Build list of required jobs in generate-workflow (#4326)" (#4413)

This reverts commit 22879d6.

Now that we have just one job to check we don't need to build the list anymore.

* Drop Python 3.9 support (#4412)

* Drop Python 3.9 support

Signed-off-by: emdneto <9735060+emdneto@users.noreply.github.com>

* generate-workflows

Signed-off-by: emdneto <9735060+emdneto@users.noreply.github.com>

* fixes

Signed-off-by: emdneto <9735060+emdneto@users.noreply.github.com>

* remove extra reference to pypy310

Signed-off-by: emdneto <9735060+emdneto@users.noreply.github.com>

* changelog

Signed-off-by: emdneto <9735060+emdneto@users.noreply.github.com>

* fix flask tests

Signed-off-by: emdneto <9735060+emdneto@users.noreply.github.com>

* fix google-genai tests

Signed-off-by: emdneto <9735060+emdneto@users.noreply.github.com>

* fix google-genai tests

Signed-off-by: emdneto <9735060+emdneto@users.noreply.github.com>

* fix

Signed-off-by: emdneto <9735060+emdneto@users.noreply.github.com>

* remove unused _ensure_gzip_single_response

Signed-off-by: emdneto <9735060+emdneto@users.noreply.github.com>

---------

Signed-off-by: emdneto <9735060+emdneto@users.noreply.github.com>
Co-authored-by: Riccardo Magliocchetti <riccardo.magliocchetti@gmail.com>

* Add AGENTS.md with project structure and commands (#4233)

* Add CLAUDE.md with project structure and commands

* Add AGENTS.md symlink to CLAUDE.md

* Move guidance to AGENTS.md and address review feedback

- Move main content from CLAUDE.md to AGENTS.md so all AI agents
  (not only Claude) pick up the guidance; CLAUDE.md now just references
  it via `@AGENTS.md`.
- Add general rules, PR scoping, and `Assisted-by:` commit trailer
  guidance (inspired by the Collector's AGENTS.md).
- Clarify that only instrumentation packages live under
  `src/opentelemetry/instrumentation/{name}/`; other package types use
  their own namespace.

---------

Co-authored-by: Riccardo Magliocchetti <riccardo.magliocchetti@gmail.com>

* scripts: drop update_sha (#4430)

It's buggy and unused.

* Fix pylint false positives for ThreadPoolExecutor (#4244)

* Bump pylint to 4.0.5 to fix Python 3.14 concurrent.futures false positives

* Fix too-many-positional-arguments pylint failures - Add max-positional-arguments=10, Add pylint: disable=too-many-positional-arguments to functions that legitimately exceed the limit

* Bump max-positional-arguments to 12 and remove unnecessary disable comments

* Address review comments: fix CHANGELOG, remove stale pylintrc comment, add openai-agents disable

* Fix formatting: restore blank line in CHANGELOG and remove extra blank line in .pylintrc

* Update processor.py

* Update test_botocore_bedrock.py

---------

Co-authored-by: Riccardo Magliocchetti <riccardo.magliocchetti@gmail.com>

* feat(util-genai): refactor and make API smaller and more user-friendly (#4391)

* Refactor public API on GenAI utils

* more lint

* review feedback

* update tests to use named params

* address some of the comments

* up

* fix failing checks and clean up imports

* lint

* lint

* fix lint

* replace @deprecated with docstring info to avoid warnings for users

* up

* common code for context manager

* Adding metrics call for Embedding type after refactoring

* Updating metrics tests with embedding

* Adding fix for markdown-link-check

---------

Signed-off-by: dependabot[bot] <support@github.com>
Signed-off-by: emdneto <9735060+emdneto@users.noreply.github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Riccardo Magliocchetti <riccardo.magliocchetti@gmail.com>
Co-authored-by: Emídio Neto <9735060+emdneto@users.noreply.github.com>
Co-authored-by: Marcelo Trylesinski <marcelotryle@gmail.com>
Co-authored-by: Sri Kaaviya <107148069+srikaaviya@users.noreply.github.com>
Co-authored-by: Liudmila Molkova <neskazu@gmail.com>

Author: 7 people

util/opentelemetry-util-genai/CHANGELOG.md

@@ -7,18 +7,20 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## Unreleased
 
+- Add metrics support for EmbeddingInvocation
+  ([#4377](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4377))
 - Add support for workflow in genAI utils handler.
-  ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4366](#4366))
+  ([#4366](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4366))
 - Enrich ToolCall type, breaking change: usage of ToolCall class renamed to ToolCallRequest
   ([#4218](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4218))
 - Add EmbeddingInvocation span lifecycle support
-  ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4219](#4219))
+  ([#4219](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4219))
 - Populate schema_url on metrics
   ([#4320](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4320))
 - Add workflow invocation type to genAI utils
-  ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4310](#4310))
+  ([#4310](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4310))
 - Check if upload works at startup in initializer of the `UploadCompletionHook`, instead
-of repeatedly failing on every upload ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4390](#4390)).
+of repeatedly failing on every upload ([#4390](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4390)).
 - Refactor public API: add factory methods (`start_inference`, `start_embedding`, `start_tool`, `start_workflow`) and invocation-owned lifecycle (`invocation.stop()` / `invocation.fail(exc)`); rename `LLMInvocation` → `InferenceInvocation` and `ToolCall` → `ToolInvocation`. Existing usages remain fully functional via deprecated aliases.
   ([#4391](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4391))
 
@@ -32,28 +34,28 @@ of repeatedly failing on every upload ([https://github.com/open-telemetry/opente
 - Log error when `fsspec` fails to be imported instead of silently failing ([#4037](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4037)).
 - Minor change to check LRU cache in Completion Hook before acquiring semaphore/thread ([#3907](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3907)).
 - Add environment variable for genai upload hook queue size
-  ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3943](#3943))
+  ([#3943](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3943))
 - Add more Semconv attributes to LLMInvocation spans.
-  ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3862](#3862))
+  ([#3862](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3862))
 - Limit the upload hook thread pool to 64 workers
-  ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3944](#3944))
+  ([#3944](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3944))
 - Add metrics to LLMInvocation traces
-  ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3891](#3891))
+  ([#3891](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3891))
 - Add parent class genAI invocation
-  ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3889](#3889))
+  ([#3889](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3889))
 
 ## Version 0.2b0 (2025-10-14)
 
 - Add jsonlines support to fsspec uploader
-  ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3791](#3791))
+  ([#3791](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3791))
 - Rename "fsspec_upload" entry point and classes to more generic "upload"
-  ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3798](#3798))
+  ([#3798](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3798))
 - Record content-type and use canonical paths in fsspec genai uploader
-  ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3795](#3795))
+  ([#3795](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3795))
 - Make inputs / outputs / system instructions optional params to `on_completion`,
-  ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3802](#3802)).
+  ([#3802](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3802)).
 - Use a SHA256 hash of the system instructions as it's upload filename, and check
-  if the file exists before re-uploading it, ([https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3814](#3814)).
+  if the file exists before re-uploading it, ([#3814](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3814)).
 
 ## Version 0.1b0 (2025-09-25)
 

util/opentelemetry-util-genai/src/opentelemetry/util/genai/_embedding_invocation.py

@@ -120,5 +120,4 @@ def _apply_finish(self, error: Error | None = None) -> None:
             self._apply_error_attributes(error)
         attributes.update(self.attributes)
         self.span.set_attributes(attributes)
-        # Metrics recorder currently supports InferenceInvocation fields only.
-        # No-op until dedicated embedding metric support is added.
+        self._metrics_recorder.record(self)

util/opentelemetry-util-genai/tests/test_handler_metrics.py

@@ -181,3 +181,145 @@ def _assert_metric_scope_schema_urls(
                 self.assertEqual(
                     scope_metric.scope.schema_url, expected_schema_url
                 )
+
+    def test_stop_embedding_records_duration_and_tokens(self) -> None:
+        """Verify embedding invocations record duration and input token metrics."""
+        handler = TelemetryHandler(
+            tracer_provider=self.tracer_provider,
+            meter_provider=self.meter_provider,
+        )
+        # Patch default_timer during start to ensure monotonic_start_s
+        with patch("timeit.default_timer", return_value=1000.0):
+            invocation = handler.start_embedding(
+                "embed-prov", request_model="embed-model"
+            )
+        invocation.input_tokens = 100
+
+        # Simulate 1.5 seconds of elapsed monotonic time
+        with patch("timeit.default_timer", return_value=1001.5):
+            invocation.stop()
+
+        self._assert_metric_scope_schema_urls(_DEFAULT_SCHEMA_URL)
+        metrics = self._harvest_metrics()
+
+        # Duration should be recorded
+        self.assertIn("gen_ai.client.operation.duration", metrics)
+        duration_points = metrics["gen_ai.client.operation.duration"]
+        self.assertEqual(len(duration_points), 1)
+        duration_point = duration_points[0]
+        self.assertEqual(
+            duration_point.attributes[GenAI.GEN_AI_OPERATION_NAME],
+            GenAI.GenAiOperationNameValues.EMBEDDINGS.value,
+        )
+        self.assertEqual(
+            duration_point.attributes[GenAI.GEN_AI_REQUEST_MODEL],
+            "embed-model",
+        )
+        self.assertEqual(
+            duration_point.attributes[GenAI.GEN_AI_PROVIDER_NAME], "embed-prov"
+        )
+        self.assertAlmostEqual(duration_point.sum, 1.5, places=3)
+
+        # Token metrics should be recorded for embedding (input only)
+        self.assertIn("gen_ai.client.token.usage", metrics)
+        token_points = metrics["gen_ai.client.token.usage"]
+        self.assertEqual(len(token_points), 1)  # Only input tokens
+        token_point = token_points[0]
+        self.assertEqual(
+            token_point.attributes[GenAI.GEN_AI_TOKEN_TYPE],
+            GenAI.GenAiTokenTypeValues.INPUT.value,
+        )
+        self.assertAlmostEqual(token_point.sum, 100.0, places=3)
+
+    def test_stop_embedding_records_duration_with_additional_attributes(
+        self,
+    ) -> None:
+        """Verify embedding metrics include server and custom attributes."""
+        handler = TelemetryHandler(
+            tracer_provider=self.tracer_provider,
+            meter_provider=self.meter_provider,
+        )
+        invocation = handler.start_embedding(
+            "embed-prov",
+            request_model="embed-model",
+            server_address="embed.server.com",
+            server_port=8080,
+        )
+        invocation.metric_attributes = {"custom.embed.attr": "embed_value"}
+        invocation.response_model_name = "embed-response-model"
+        invocation.stop()
+
+        self._assert_metric_scope_schema_urls(_DEFAULT_SCHEMA_URL)
+        metrics = self._harvest_metrics()
+
+        self.assertIn("gen_ai.client.operation.duration", metrics)
+        duration_points = metrics["gen_ai.client.operation.duration"]
+        self.assertEqual(len(duration_points), 1)
+        duration_point = duration_points[0]
+
+        self.assertEqual(
+            duration_point.attributes["server.address"], "embed.server.com"
+        )
+        self.assertEqual(duration_point.attributes["server.port"], 8080)
+        self.assertEqual(
+            duration_point.attributes["custom.embed.attr"], "embed_value"
+        )
+        self.assertEqual(
+            duration_point.attributes[GenAI.GEN_AI_RESPONSE_MODEL],
+            "embed-response-model",
+        )
+
+    def test_fail_embedding_records_error_and_duration(self) -> None:
+        """Verify embedding failure records error type and duration."""
+        handler = TelemetryHandler(
+            tracer_provider=self.tracer_provider,
+            meter_provider=self.meter_provider,
+        )
+        with patch("timeit.default_timer", return_value=3000.0):
+            invocation = handler.start_embedding(
+                "embed-prov", request_model="embed-err-model"
+            )
+
+        error = Error(message="embedding failed", type=RuntimeError)
+        with patch("timeit.default_timer", return_value=3002.5):
+            invocation.fail(error)
+
+        self._assert_metric_scope_schema_urls(_DEFAULT_SCHEMA_URL)
+        metrics = self._harvest_metrics()
+
+        self.assertIn("gen_ai.client.operation.duration", metrics)
+        duration_points = metrics["gen_ai.client.operation.duration"]
+        self.assertEqual(len(duration_points), 1)
+        duration_point = duration_points[0]
+
+        self.assertEqual(
+            duration_point.attributes.get("error.type"), "RuntimeError"
+        )
+        self.assertEqual(
+            duration_point.attributes.get(GenAI.GEN_AI_REQUEST_MODEL),
+            "embed-err-model",
+        )
+        self.assertAlmostEqual(duration_point.sum, 2.5, places=3)
+
+        # Token metrics should NOT be recorded when input_tokens is not set
+        self.assertNotIn("gen_ai.client.token.usage", metrics)
+
+    def test_stop_embedding_without_tokens(self) -> None:
+        """Verify embedding without input_tokens does not record token metrics."""
+        handler = TelemetryHandler(
+            tracer_provider=self.tracer_provider,
+            meter_provider=self.meter_provider,
+        )
+        invocation = handler.start_embedding(
+            "embed-prov", request_model="embed-model"
+        )
+        # input_tokens is not set
+        invocation.stop()
+
+        metrics = self._harvest_metrics()
+
+        # Duration should be recorded
+        self.assertIn("gen_ai.client.operation.duration", metrics)
+
+        # Token metrics should NOT be recorded when input_tokens is not set
+        self.assertNotIn("gen_ai.client.token.usage", metrics)