Spike: Adding Llamaindex llm instrumentation

instrumentation-genai/opentelemetry-instrumentation-llamaindex/README.rst

@@ -0,0 +1,155 @@
+OpenTelemetry LlamaIndex Instrumentation
+=========================================
+
+This library provides automatic instrumentation for LlamaIndex applications using OpenTelemetry.
+
+Installation
+------------
+
+Development installation::
+
+    # Install the package in editable mode
+    cd instrumentation-genai/opentelemetry-instrumentation-llamaindex
+    pip install -e .
+
+    # Install test dependencies
+    pip install -e ".[test]"
+
+    # Install util-genai (required for telemetry)
+    cd ../../util/opentelemetry-util-genai
+    pip install -e .
+
+
+Quick Start
+-----------
+
+.. code-block:: python
+
+    import os
+    from opentelemetry.instrumentation.llamaindex import LlamaindexInstrumentor
+    from opentelemetry import trace, metrics
+    from opentelemetry.sdk.trace import TracerProvider
+    from opentelemetry.sdk.trace.export import ConsoleSpanExporter, SimpleSpanProcessor
+    from opentelemetry.sdk.metrics import MeterProvider
+    from opentelemetry.sdk.metrics.export import InMemoryMetricReader
+
+    # Enable metrics (default is spans only)
+    os.environ["OTEL_INSTRUMENTATION_GENAI_EMITTERS"] = "span_metric"
+
+    # Setup tracing
+    trace.set_tracer_provider(TracerProvider())
+    trace.get_tracer_provider().add_span_processor(
+        SimpleSpanProcessor(ConsoleSpanExporter())
+    )
+
+    # Setup metrics
+    metric_reader = InMemoryMetricReader()
+    meter_provider = MeterProvider(metric_readers=[metric_reader])
+    metrics.set_meter_provider(meter_provider)
+
+    # Enable instrumentation with providers
+    LlamaindexInstrumentor().instrument(
+        tracer_provider=trace.get_tracer_provider(),
+        meter_provider=meter_provider
+    )
+
+    # Use LlamaIndex as normal
+    from llama_index.llms.openai import OpenAI
+    from llama_index.core.llms import ChatMessage, MessageRole
+
+    llm = OpenAI(model="gpt-3.5-turbo")
+    messages = [ChatMessage(role=MessageRole.USER, content="Hello")]
+    response = llm.chat(messages)
+
+
+Running Tests
+-------------
+
+.. code-block:: bash
+
+    # Set environment variables
+    export OPENAI_API_KEY=your-api-key
+    export OTEL_INSTRUMENTATION_GENAI_EMITTERS=span_metric
+
+    # Run the test
+    cd tests
+    python test_llm_instrumentation.py
+
+
+Expected Output
+---------------
+
+**Span Attributes**::
+
+    {
+        "gen_ai.framework": "llamaindex",
+        "gen_ai.request.model": "gpt-3.5-turbo",
+        "gen_ai.operation.name": "chat",
+        "gen_ai.usage.input_tokens": 24,
+        "gen_ai.usage.output_tokens": 7
+    }
+
+**Metrics**::
+
+    Metric: gen_ai.client.operation.duration
+      Duration: 0.6900 seconds
+      Count: 1
+
+    Metric: gen_ai.client.token.usage
+      Token type: input, Sum: 24, Count: 1
+      Token type: output, Sum: 7, Count: 1
+
+
+Key Implementation Differences from LangChain
+----------------------------------------------
+
+**1. Event-Based Callbacks**
+
+LlamaIndex uses ``on_event_start(event_type, ...)`` and ``on_event_end(event_type, ...)`` 
+instead of LangChain's method-based callbacks (``on_llm_start``, ``on_llm_end``).
+
+Event types are dispatched via ``CBEventType`` enum::
+
+    CBEventType.LLM       # LLM invocations
+    CBEventType.AGENT     # Agent steps  
+    CBEventType.EMBEDDING # Embedding operations
+
+**2. Handler Registration**
+
+LlamaIndex uses ``handlers`` list::
+
+    callback_manager.handlers.append(handler)
+
+LangChain uses ``inheritable_handlers``::
+
+    callback_manager.inheritable_handlers.append(handler)
+
+**3. Response Structure**
+
+LlamaIndex ``ChatMessage`` uses ``blocks`` (list of TextBlock objects)::
+
+    message.content  # Computed property from blocks[0].text
+
+LangChain uses simple strings::
+
+    message.content  # Direct string property
+
+**4. Token Usage**
+
+LlamaIndex returns objects (not dicts)::
+
+    response.raw.usage.prompt_tokens      # Object attribute
+    response.raw.usage.completion_tokens  # Object attribute
+
+LangChain returns dicts::
+
+    response["usage"]["prompt_tokens"]      # Dict key
+    response["usage"]["completion_tokens"]  # Dict key
+
+
+References
+----------
+
+* `OpenTelemetry Project <https://opentelemetry.io/>`_
+* `LlamaIndex <https://www.llamaindex.ai/>`_
+* `LlamaIndex Callbacks <https://docs.llamaindex.ai/en/stable/module_guides/observability/callbacks/>`_

instrumentation-genai/opentelemetry-instrumentation-llamaindex/pyproject.toml

@@ -0,0 +1,58 @@
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[project]
+name = "splunk-otel-instrumentation-llamaindex"
+dynamic = ["version"]
+description = "OpenTelemetry LlamaIndex instrumentation"
+readme = "README.rst"
+license = "Apache-2.0"
+requires-python = ">=3.9"
+authors = [
+    { name = "OpenTelemetry Authors", email = "[email protected]" },
+]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: Apache Software License",
+    "Programming Language :: Python",
+    "Programming Language :: Python :: 3.9",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+]
+dependencies = [
+    "opentelemetry-api ~= 1.38.0.dev0",
+    "opentelemetry-instrumentation ~= 0.59b0.dev0",
+    "opentelemetry-semantic-conventions ~= 0.59b0.dev0",
+    "splunk-otel-util-genai>=0.1.4",
+]
+
+[project.optional-dependencies]
+instruments = ["llama-index-core >= 0.14.0"]
+test = [
+    "llama-index-core >= 0.14.0",
+    "llama-index-llms-openai >= 0.6.0",
+    "pytest >= 7.0.0",
+]
+
+[project.entry-points.opentelemetry_instrumentor]
+llamaindex = "opentelemetry.instrumentation.llamaindex:LlamaindexInstrumentor"
+
+[project.urls]
+Homepage = "https://github.com/open-telemetry/opentelemetry-python-contrib/tree/main/instrumentation-genai/opentelemetry-instrumentation-llamaindex"
+Repository = "https://github.com/open-telemetry/opentelemetry-python-contrib"
+
+[tool.hatch.version]
+path = "src/opentelemetry/instrumentation/llamaindex/version.py"
+
+[tool.hatch.build.targets.sdist]
+include = ["/src", "/tests"]
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/opentelemetry"]
+
+[tool.ruff]
+exclude = ["./"]

instrumentation-genai/opentelemetry-instrumentation-llamaindex/src/opentelemetry/instrumentation/llamaindex/__init__.py

@@ -0,0 +1,67 @@
+from opentelemetry.instrumentation.instrumentor import BaseInstrumentor
+from opentelemetry.util.genai.handler import get_telemetry_handler
+from opentelemetry.instrumentation.llamaindex.config import Config
+from opentelemetry.instrumentation.llamaindex.callback_handler import (
+    LlamaindexCallbackHandler,
+)
+from wrapt import wrap_function_wrapper
+
+_instruments = ("llama-index-core >= 0.14.0",)
+
+
+class LlamaindexInstrumentor(BaseInstrumentor):
+    def __init__(
+        self,
+        exception_logger=None,
+        disable_trace_context_propagation=False,
+        use_legacy_attributes: bool = True,
+    ):
+        super().__init__()
+        Config._exception_logger = exception_logger
+        Config.use_legacy_attributes = use_legacy_attributes
+        self._disable_trace_context_propagation = (
+            disable_trace_context_propagation
+        )
+        self._telemetry_handler = None
+
+    def instrumentation_dependencies(self):
+        return _instruments
+
+    def _instrument(self, **kwargs):
+        tracer_provider = kwargs.get("tracer_provider")
+        meter_provider = kwargs.get("meter_provider")
+        logger_provider = kwargs.get("logger_provider")
+
+        self._telemetry_handler = get_telemetry_handler(
+            tracer_provider=tracer_provider,
+            meter_provider=meter_provider,
+            logger_provider=logger_provider,
+        )
+
+        llamaindexCallBackHandler = LlamaindexCallbackHandler(
+            telemetry_handler=self._telemetry_handler
+        )
+
+        wrap_function_wrapper(
+            module="llama_index.core.callbacks.base",
+            name="CallbackManager.__init__",
+            wrapper=_BaseCallbackManagerInitWrapper(llamaindexCallBackHandler),
+        )
+
+    def _uninstrument(self, **kwargs):
+        pass
+
+
+class _BaseCallbackManagerInitWrapper:
+    def __init__(self, callback_handler: "LlamaindexCallbackHandler"):
+        self._callback_handler = callback_handler
+
+    def __call__(self, wrapped, instance, args, kwargs) -> None:
+        wrapped(*args, **kwargs)
+        # LlamaIndex uses 'handlers' instead of 'inheritable_handlers'
+        for handler in instance.handlers:
+            if isinstance(handler, type(self._callback_handler)):
+                break
+        else:
+            self._callback_handler._callback_manager = instance
+            instance.add_handler(self._callback_handler)