Add OpenAI embeddings instrumentation (#3461)

* Initial implementation and tests

* Add embeddings example

* Update documentation

* Changelog entry

* Add comment about custom attributes

* Update PR link in Changelog

* Add input and output events for embeddings

* Fix changelog

* Use gen_ai.embeddings.dimension.count

* Remove total_tokens

* Use end_on_exit

* Don't import conditionally

* Fix heading

* Use gen_ai.request.encoding_formats

* Use gen_ai.request.encoding_formats

* Rename span_attr to request_attr

* Remove embeddings capture via events

* bump versions

* Remove unused event code

* Updates for PR feedback

* Update requirements for example

* dimension.count is added in get_llm_request_attributes

* Use a shared function to reduce duplicate code

* remove return

* Update instrumentation-genai/opentelemetry-instrumentation-openai-v2/examples/embeddings/requirements.txt

---------

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

Author: drewby

instrumentation-genai/opentelemetry-instrumentation-openai-v2/CHANGELOG.md

@@ -7,6 +7,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## Unreleased
 
+- Added support for OpenAI embeddings instrumentation
+  ([#3461](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3461))
 - Record prompt and completion events regardless of span sampling decision.
   ([#3226](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3226))
 - Migrate off the deprecated events API to use the logs API

instrumentation-genai/opentelemetry-instrumentation-openai-v2/README.rst

@@ -56,7 +56,7 @@ Check out the `manual example <examples/manual>`_ for more details.
 Instrumenting all clients
 *************************
 
-When using the instrumentor, all clients will automatically trace OpenAI chat completion operations.
+When using the instrumentor, all clients will automatically trace OpenAI operations including chat completions and embeddings.
 You can also optionally capture prompts and completions as log events.
 
 Make sure to configure OpenTelemetry tracing, logging, and events to capture all telemetry emitted by the instrumentation.
@@ -68,12 +68,19 @@ Make sure to configure OpenTelemetry tracing, logging, and events to capture all
     OpenAIInstrumentor().instrument()
 
     client = OpenAI()
+    # Chat completion example
     response = client.chat.completions.create(
         model="gpt-4o-mini",
         messages=[
             {"role": "user", "content": "Write a short poem on open telemetry."},
         ],
     )
+    
+    # Embeddings example
+    embedding_response = client.embeddings.create(
+        model="text-embedding-3-small",
+        input="Generate vector embeddings for this text"
+    )
 
 Enabling message content
 *************************

instrumentation-genai/opentelemetry-instrumentation-openai-v2/examples/embeddings/.env

@@ -0,0 +1,22 @@
+# Update this with your real OpenAI API key
+OPENAI_API_KEY=sk-YOUR_API_KEY
+
+# Uncomment to use Ollama instead of OpenAI
+# OPENAI_BASE_URL=http://localhost:11434/v1
+# OPENAI_API_KEY=unused
+# CHAT_MODEL=qwen2.5:0.5b
+
+# Uncomment and change to your OTLP endpoint
+# OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
+# OTEL_EXPORTER_OTLP_PROTOCOL=grpc
+
+OTEL_SERVICE_NAME=opentelemetry-python-openai
+
+# Change to 'false' to disable collection of python logging logs
+OTEL_PYTHON_LOGGING_AUTO_INSTRUMENTATION_ENABLED=true
+
+# Uncomment if your OTLP endpoint doesn't support logs
+# OTEL_LOGS_EXPORTER=console
+
+# Change to 'false' to hide prompt and completion content
+OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=true

instrumentation-genai/opentelemetry-instrumentation-openai-v2/examples/embeddings/README.rst

@@ -0,0 +1,44 @@
+OpenTelemetry OpenAI Embeddings API Instrumentation Example
+===========================================================
+
+This is an example of how to instrument OpenAI Embeddings API calls with zero code changes,
+using ``opentelemetry-instrument``.
+
+When ``main.py`` is run, it exports traces and metrics to an OTLP
+compatible endpoint. Traces include details such as the model used,
+dimensions of embeddings, and the duration of the embedding request.
+Metrics capture token usage and performance data.
+
+Note: ``.env`` file configures additional environment variables:
+
+- ``OTEL_PYTHON_LOGGING_AUTO_INSTRUMENTATION_ENABLED=true`` configures OpenTelemetry SDK to export logs and events.
+- ``OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=true`` configures OpenAI instrumentation to capture content on events.
+- ``OTEL_LOGS_EXPORTER=otlp`` to specify exporter type.
+
+Setup
+-----
+
+Minimally, update the ``.env`` file with your ``OPENAI_API_KEY``. An
+OTLP compatible endpoint should be listening for traces and logs on
+http://localhost:4317. If not, update ``OTEL_EXPORTER_OTLP_ENDPOINT`` as well.
+
+Next, set up a virtual environment like this:
+
+::
+
+    python3 -m venv .venv
+    source .venv/bin/activate
+    pip install "python-dotenv[cli]"
+    pip install -r requirements.txt
+
+Run
+---
+
+Run the example like this:
+
+::
+
+    dotenv run -- opentelemetry-instrument python main.py
+
+You should see embedding information printed while traces and metrics export to your
+configured observability tool.

instrumentation-genai/opentelemetry-instrumentation-openai-v2/examples/embeddings/main.py

@@ -0,0 +1,29 @@
+import os
+
+from openai import OpenAI
+
+
+def main():
+    client = OpenAI()
+
+    # Create embeddings with OpenAI API
+    embedding_response = client.embeddings.create(
+        model=os.getenv("EMBEDDING_MODEL", "text-embedding-3-small"),
+        input="OpenTelemetry provides observability for your applications.",
+    )
+
+    # Print embedding information
+    print(f"Model: {embedding_response.model}")
+    print(f"Dimensions: {len(embedding_response.data[0].embedding)}")
+    print(
+        f"Token usage - Prompt: {embedding_response.usage.prompt_tokens}, Total: {embedding_response.usage.total_tokens}"
+    )
+
+    # Print a sample of the embedding vector (first 5 dimensions)
+    print(
+        f"Embedding sample (first 5 dimensions): {embedding_response.data[0].embedding[:5]}"
+    )
+
+
+if __name__ == "__main__":
+    main()

instrumentation-genai/opentelemetry-instrumentation-openai-v2/examples/embeddings/requirements.txt

@@ -0,0 +1,6 @@
+openai~=1.57.3
+
+opentelemetry-sdk~=1.36.0
+opentelemetry-exporter-otlp-proto-grpc~=1.36.0
+opentelemetry-distro~=0.57b0
+opentelemetry-instrumentation-openai-v2~=2.2b0

instrumentation-genai/opentelemetry-instrumentation-openai-v2/src/opentelemetry/instrumentation/openai_v2/__init__.py

@@ -54,7 +54,12 @@
 from opentelemetry.trace import get_tracer
 
 from .instruments import Instruments
-from .patch import async_chat_completions_create, chat_completions_create
+from .patch import (
+    async_chat_completions_create,
+    async_embeddings_create,
+    chat_completions_create,
+    embeddings_create,
+)
 
 
 class OpenAIInstrumentor(BaseInstrumentor):
@@ -106,8 +111,27 @@ def _instrument(self, **kwargs):
             ),
         )
 
+        # Add instrumentation for the embeddings API
+        wrap_function_wrapper(
+            module="openai.resources.embeddings",
+            name="Embeddings.create",
+            wrapper=embeddings_create(
+                tracer, instruments, is_content_enabled()
+            ),
+        )
+
+        wrap_function_wrapper(
+            module="openai.resources.embeddings",
+            name="AsyncEmbeddings.create",
+            wrapper=async_embeddings_create(
+                tracer, instruments, is_content_enabled()
+            ),
+        )
+
     def _uninstrument(self, **kwargs):
         import openai  # pylint: disable=import-outside-toplevel
 
         unwrap(openai.resources.chat.completions.Completions, "create")
         unwrap(openai.resources.chat.completions.AsyncCompletions, "create")
+        unwrap(openai.resources.embeddings.Embeddings, "create")
+        unwrap(openai.resources.embeddings.AsyncEmbeddings, "create")