Added span support for genAI langchain llm invocation (#3665)

* Added span support for llm invocation

* removed invalid code

* added entry point and fixed unwrap

* fixed check runs and updated dependencies

* fixed ruff error

* moved span generation code and added test coverage

* ruff formatting

* ruff formatting again

* removed config exception logger

* removed dontThrow

* fixed span name

* fixed ruff

* fixed typecheck

* added span exist check

* fixed typecheck

* removed start time from span state and moved error handler method to span manager

* fixed ruff

* made SpanManager class and method private

* removed deprecated gen_ai.system attribute

* Moved model to fixture and changed imports

* Fixed ruff errors and renamed method

* Added bedrock support and test

* Fixed ruff errors

* Addressed Aaron's comments

* Reverted versions and ignored typecheck errors

* removed context and added issue

* fixed versions

* skipped telemetry for other than ChatOpenAI and ChatBedrock. Added test for the same.

* Fixed telemetry skipping logic

* Fixed ruff

* added notice file

* fixed conflict

* fixed ruff and typecheck

* fixed ruff

* upgraded semcov version

---------

Co-authored-by: Aaron Abbott <[email protected]>

Author: wrisa

.github/component_owners.yml

@@ -40,3 +40,7 @@ components:
   util/opentelemetry-util-genai:
     - DylanRussell
     - keith-decker
+
+  instrumentation-genai/opentelemetry-instrumentation-langchain:
+    - zhirafovod
+    - wrisa

instrumentation-genai/opentelemetry-instrumentation-langchain/CHANGELOG.md

@@ -5,4 +5,7 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
-## Unreleased
+## Unreleased
+
+- Added span support for genAI langchain llm invocation.
+  ([#3665](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3665))

instrumentation-genai/opentelemetry-instrumentation-langchain/NOTICE

@@ -0,0 +1,3 @@
+This project is inspired by and portions of it are derived from Traceloop OpenLLMetry
+(https://github.com/traceloop/openllmetry).
+Licensed under the Apache License, Version 2.0 (http://www.apache.org/licenses/LICENSE-2.0).

instrumentation-genai/opentelemetry-instrumentation-langchain/examples/manual/.env

@@ -0,0 +1,8 @@
+# Update this with your real OpenAI API key
+OPENAI_API_KEY=sk-YOUR_API_KEY
+
+# Uncomment and change to your OTLP endpoint
+# OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
+# OTEL_EXPORTER_OTLP_PROTOCOL=grpc
+
+OTEL_SERVICE_NAME=opentelemetry-python-langchain-manual

instrumentation-genai/opentelemetry-instrumentation-langchain/examples/manual/README.rst

@@ -0,0 +1,39 @@
+OpenTelemetry Langcahin Instrumentation Example
+============================================
+
+This is an example of how to instrument Langchain when configuring OpenTelemetry SDK and instrumentations manually.
+
+When :code:`main.py <main.py>`_ is run, it exports traces to an OTLP-compatible endpoint.
+Traces include details such as the span name and other attributes.
+
+Note: :code:`.env <.env>`_ file configures additional environment variables:
+- :code:`OTEL_LOGS_EXPORTER=otlp` to specify exporter type.
+- :code:`OPENAI_API_KEY` open AI key for accessing the OpenAI API.
+- :code:`OTEL_EXPORTER_OTLP_ENDPOINT` to specify the endpoint for exporting traces (default is http://localhost:4317).
+
+Setup
+-----
+
+Minimally, update the :code:`.env <.env>`_ file with your :code:`OPENAI_API_KEY`.
+An OTLP compatible endpoint should be listening for traces http://localhost:4317.
+If not, update :code:`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 -- python main.py
+
+You should see the capital of France generated by Langchain ChatOpenAI while traces export to your configured observability tool.

instrumentation-genai/opentelemetry-instrumentation-langchain/examples/manual/main.py

@@ -0,0 +1,48 @@
+from langchain_core.messages import HumanMessage, SystemMessage
+from langchain_openai import ChatOpenAI
+
+from opentelemetry import trace
+from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import (
+    OTLPSpanExporter,
+)
+from opentelemetry.instrumentation.langchain import LangChainInstrumentor
+from opentelemetry.sdk.trace import TracerProvider
+from opentelemetry.sdk.trace.export import BatchSpanProcessor
+
+# Configure tracing
+trace.set_tracer_provider(TracerProvider())
+span_processor = BatchSpanProcessor(OTLPSpanExporter())
+trace.get_tracer_provider().add_span_processor(span_processor)
+
+
+def main():
+    # Set up instrumentation
+    LangChainInstrumentor().instrument()
+
+    # ChatOpenAI
+    llm = ChatOpenAI(
+        model="gpt-3.5-turbo",
+        temperature=0.1,
+        max_tokens=100,
+        top_p=0.9,
+        frequency_penalty=0.5,
+        presence_penalty=0.5,
+        stop_sequences=["\n", "Human:", "AI:"],
+        seed=100,
+    )
+
+    messages = [
+        SystemMessage(content="You are a helpful assistant!"),
+        HumanMessage(content="What is the capital of France?"),
+    ]
+
+    result = llm.invoke(messages)
+
+    print("LLM output:\n", result)
+
+    # Un-instrument after use
+    LangChainInstrumentor().uninstrument()
+
+
+if __name__ == "__main__":
+    main()

instrumentation-genai/opentelemetry-instrumentation-langchain/examples/manual/requirements.txt

@@ -0,0 +1,7 @@
+langchain==0.3.21
+langchain_openai
+opentelemetry-sdk>=1.31.0
+opentelemetry-exporter-otlp-proto-grpc>=1.31.0
+
+# Uncomment after lanchain instrumetation is released
+# opentelemetry-instrumentation-langchain~=2.0b0.dev

instrumentation-genai/opentelemetry-instrumentation-langchain/examples/zero-code/.env

@@ -0,0 +1,8 @@
+# Update this with your real OpenAI API key
+OPENAI_API_KEY=sk-YOUR_API_KEY
+
+# Uncomment and change to your OTLP endpoint
+# OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
+# OTEL_EXPORTER_OTLP_PROTOCOL=grpc
+
+OTEL_SERVICE_NAME=opentelemetry-python-langchain-zero-code

instrumentation-genai/opentelemetry-instrumentation-langchain/examples/zero-code/README.rst

@@ -0,0 +1,40 @@
+OpenTelemetry Langchain Zero-Code Instrumentation Example
+======================================================
+
+This is an example of how to instrument Langchain with zero code changes,
+using `opentelemetry-instrument`.
+
+When :code:`main.py <main.py>`_ is run, it exports traces to an OTLP-compatible endpoint.
+Traces include details such as the span name and other attributes.
+
+Note: :code:`.env <.env>`_ file configures additional environment variables:
+- :code:`OTEL_LOGS_EXPORTER=otlp` to specify exporter type.
+- :code:`OPENAI_API_KEY` open AI key for accessing the OpenAI API.
+- :code:`OTEL_EXPORTER_OTLP_ENDPOINT` to specify the endpoint for exporting traces (default is http://localhost:4317).
+
+Setup
+-----
+
+Minimally, update the :code:`.env <.env>`_ file with your :code:`OPENAI_API_KEY`.
+An OTLP compatible endpoint should be listening for traces http://localhost:4317.
+If not, update :code:`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 the capital of France generated by Langchain ChatOpenAI while traces export to your configured observability tool.

instrumentation-genai/opentelemetry-instrumentation-langchain/examples/zero-code/main.py

@@ -0,0 +1,27 @@
+from langchain_core.messages import HumanMessage, SystemMessage
+from langchain_openai import ChatOpenAI
+
+
+def main():
+    llm = ChatOpenAI(
+        model="gpt-3.5-turbo",
+        temperature=0.1,
+        max_tokens=100,
+        top_p=0.9,
+        frequency_penalty=0.5,
+        presence_penalty=0.5,
+        stop_sequences=["\n", "Human:", "AI:"],
+        seed=100,
+    )
+
+    messages = [
+        SystemMessage(content="You are a helpful assistant!"),
+        HumanMessage(content="What is the capital of France?"),
+    ]
+
+    result = llm.invoke(messages).content
+    print("LLM output:\n", result)
+
+
+if __name__ == "__main__":
+    main()