opentelemetry-instrumentation-llamaindex 0.26.4__tar.gz → 0.27.0__tar.gz

{opentelemetry_instrumentation_llamaindex-0.26.4 → opentelemetry_instrumentation_llamaindex-0.27.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: opentelemetry-instrumentation-llamaindex
-Version: 0.26.4
+Version: 0.27.0
 Summary: OpenTelemetry LlamaIndex instrumentation
 Home-page: https://github.com/traceloop/openllmetry/tree/main/packages/opentelemetry-instrumentation-llamaindex
 License: Apache-2.0
@@ -18,7 +18,7 @@ Requires-Dist: inflection (>=0.5.1,<0.6.0)
 Requires-Dist: opentelemetry-api (>=1.26.0,<2.0.0)
 Requires-Dist: opentelemetry-instrumentation (>=0.47b0,<0.48)
 Requires-Dist: opentelemetry-semantic-conventions (>=0.47b0,<0.48)
-Requires-Dist: opentelemetry-semantic-conventions-ai (==0.4.0)
+Requires-Dist: opentelemetry-semantic-conventions-ai (==0.4.1)
 Project-URL: Repository, https://github.com/traceloop/openllmetry/tree/main/packages/opentelemetry-instrumentation-llamaindex
 Description-Content-Type: text/markdown
 

{opentelemetry_instrumentation_llamaindex-0.26.4 → opentelemetry_instrumentation_llamaindex-0.27.0}/opentelemetry/instrumentation/llamaindex/__init__.py

@@ -1,6 +1,7 @@
 """OpenTelemetry LlamaIndex instrumentation"""
 
 import logging
+from importlib.metadata import version as import_version
 from typing import Collection
 
 from opentelemetry.instrumentation.llamaindex.config import Config
@@ -33,6 +34,7 @@ from opentelemetry.instrumentation.llamaindex.query_pipeline_instrumentor import
     QueryPipelineInstrumentor,
 )
 from opentelemetry.instrumentation.llamaindex.version import __version__
+from opentelemetry.instrumentation.llamaindex.dispatcher_wrapper import instrument_with_dispatcher
 
 logger = logging.getLogger(__name__)
 
@@ -53,14 +55,17 @@ class LlamaIndexInstrumentor(BaseInstrumentor):
         tracer_provider = kwargs.get("tracer_provider")
         tracer = get_tracer(__name__, __version__, tracer_provider)
 
-        RetrieverQueryEngineInstrumentor(tracer).instrument()
-        BaseRetrieverInstrumentor(tracer).instrument()
-        BaseSynthesizerInstrumentor(tracer).instrument()
-        BaseEmbeddingInstrumentor(tracer).instrument()
-        CustomLLMInstrumentor(tracer).instrument()
-        QueryPipelineInstrumentor(tracer).instrument()
-        BaseAgentInstrumentor(tracer).instrument()
-        BaseToolInstrumentor(tracer).instrument()
+        if import_version("llama-index") >= "0.10.20":
+            instrument_with_dispatcher(tracer)
+        else:
+            RetrieverQueryEngineInstrumentor(tracer).instrument()
+            BaseRetrieverInstrumentor(tracer).instrument()
+            BaseSynthesizerInstrumentor(tracer).instrument()
+            BaseEmbeddingInstrumentor(tracer).instrument()
+            CustomLLMInstrumentor(tracer).instrument()
+            QueryPipelineInstrumentor(tracer).instrument()
+            BaseAgentInstrumentor(tracer).instrument()
+            BaseToolInstrumentor(tracer).instrument()
 
     def _uninstrument(self, **kwargs):
         pass

opentelemetry_instrumentation_llamaindex-0.27.0/opentelemetry/instrumentation/llamaindex/dispatcher_wrapper.py

@@ -0,0 +1,219 @@
+import inspect
+import json
+import re
+from typing import Any, Optional
+from dataclasses import dataclass
+
+from llama_index.core.bridge.pydantic import PrivateAttr
+from llama_index.core.base.llms.types import MessageRole
+from llama_index.core.instrumentation import get_dispatcher
+from llama_index.core.instrumentation.events import BaseEvent
+from llama_index.core.instrumentation.events.llm import (
+    LLMChatEndEvent,
+    LLMChatStartEvent,
+    LLMPredictEndEvent,
+)
+from llama_index.core.instrumentation.event_handlers import BaseEventHandler
+from llama_index.core.instrumentation.span_handlers import BaseSpanHandler
+from opentelemetry import context as context_api
+from opentelemetry.instrumentation.llamaindex.utils import (
+    JSONEncoder,
+    dont_throw,
+    should_send_prompts,
+)
+from opentelemetry.semconv_ai import (
+    SUPPRESS_LANGUAGE_MODEL_INSTRUMENTATION_KEY,
+    SpanAttributes,
+    TraceloopSpanKindValues,
+)
+from opentelemetry.trace import get_current_span, set_span_in_context, Tracer
+from opentelemetry.trace.span import Span
+
+
+LLAMA_INDEX_REGEX = re.compile(r"^([a-zA-Z]+)\.")
+
+
+def instrument_with_dispatcher(tracer: Tracer):
+    dispatcher = get_dispatcher()
+    openll_span_handler = OpenLLSpanHandler(tracer)
+    dispatcher.add_span_handler(openll_span_handler)
+    dispatcher.add_event_handler(OpenLLEventHandler(openll_span_handler))
+
+
+@dont_throw
+def _set_llm_chat_request(event, span) -> None:
+    model_dict = event.model_dict
+    span.set_attribute(SpanAttributes.LLM_REQUEST_MODEL, model_dict.get("model"))
+    span.set_attribute(
+        SpanAttributes.LLM_REQUEST_TEMPERATURE, model_dict.get("temperature")
+    )
+    if should_send_prompts():
+        for idx, message in enumerate(event.messages):
+            span.set_attribute(
+                f"{SpanAttributes.LLM_PROMPTS}.{idx}.role", message.role.value
+            )
+            span.set_attribute(
+                f"{SpanAttributes.LLM_PROMPTS}.{idx}.content", message.content
+            )
+
+
+@dont_throw
+def _set_llm_chat_response(event, span) -> None:
+    if should_send_prompts():
+        for idx, message in enumerate(event.messages):
+            span.set_attribute(
+                f"{SpanAttributes.LLM_PROMPTS}.{idx}.role", message.role.value
+            )
+            span.set_attribute(
+                f"{SpanAttributes.LLM_PROMPTS}.{idx}.content", message.content
+            )
+        response = event.response
+        span.set_attribute(
+            f"{SpanAttributes.LLM_COMPLETIONS}.0.role",
+            response.message.role.value,
+        )
+        span.set_attribute(
+            f"{SpanAttributes.LLM_COMPLETIONS}.0.content",
+            response.message.content,
+        )
+        if not (raw := response.raw):
+            return
+        # raw can be Any, not just ChatCompletion
+    span.set_attribute(
+        SpanAttributes.LLM_RESPONSE_MODEL,
+        raw.get("model") if "model" in raw else raw.model,
+    )
+    if usage := raw.get("usage") if "usage" in raw else raw.usage:
+        span.set_attribute(
+            SpanAttributes.LLM_USAGE_COMPLETION_TOKENS, usage.completion_tokens
+        )
+        span.set_attribute(
+            SpanAttributes.LLM_USAGE_PROMPT_TOKENS, usage.prompt_tokens
+        )
+        span.set_attribute(
+            SpanAttributes.LLM_USAGE_TOTAL_TOKENS, usage.total_tokens
+        )
+
+
+@dont_throw
+def _set_llm_predict_response(event, span) -> None:
+    if should_send_prompts():
+        span.set_attribute(
+            f"{SpanAttributes.LLM_COMPLETIONS}.role",
+            MessageRole.ASSISTANT.value,
+        )
+        span.set_attribute(
+            f"{SpanAttributes.LLM_COMPLETIONS}.content",
+            event.output,
+        )
+
+
+@dataclass
+class SpanHolder:
+    span: Span
+    token: Any
+    context: context_api.context.Context
+
+
+class OpenLLSpanHandler(BaseSpanHandler[SpanHolder]):
+    _tracer: Tracer = PrivateAttr()
+
+    def __init__(self, tracer: Tracer):
+        super().__init__()
+        self._tracer = tracer
+
+    def new_span(
+        self, id_: str, bound_args: inspect.BoundArguments, parent_span_id: Optional[str], **kwargs
+    ) -> Optional[SpanHolder]:
+        """Create a span."""
+        parent = self.open_spans.get(parent_span_id)
+        kind = (
+            TraceloopSpanKindValues.TASK.value
+            if parent
+            else TraceloopSpanKindValues.WORKFLOW.value
+        )
+        # Take the class name from id_ where id_ is e.g.
+        # 'SentenceSplitter.split_text_metadata_aware-a2f2a780-2fa6-4682-a88e-80dc1f1ebe6a'
+        class_name = LLAMA_INDEX_REGEX.match(id_).groups()[0]
+        span_name = f"{class_name}.{kind}"
+        span = self._tracer.start_span(
+            span_name,
+            context=parent.context if parent else None,
+        )
+        current_context = set_span_in_context(
+            span, context=parent.context if parent else None
+        )
+        current_context = context_api.set_value(
+            SUPPRESS_LANGUAGE_MODEL_INSTRUMENTATION_KEY,
+            True,
+            current_context,
+        )
+        token = context_api.attach(current_context)
+        span.set_attribute(SpanAttributes.TRACELOOP_SPAN_KIND, kind)
+        span.set_attribute(SpanAttributes.TRACELOOP_ENTITY_NAME, span_name)
+        try:
+            if should_send_prompts():
+                span.set_attribute(
+                    SpanAttributes.TRACELOOP_ENTITY_INPUT,
+                    json.dumps(bound_args.arguments, cls=JSONEncoder)
+                )
+        except Exception:
+            pass
+
+        return SpanHolder(span, token, current_context)
+
+    def prepare_to_exit_span(
+        self, id_: str, result: Optional[Any] = None, **kwargs
+    ) -> SpanHolder:
+        """Logic for preparing to drop a span."""
+        span_holder = self.open_spans[id_]
+        # I know it's messy, but the typing of result is messy and couldn't find a better way
+        # to get a dictionary I can then use to remove keys
+        try:
+            serialized_output = json.dumps(result, cls=JSONEncoder)
+            # we need to remove some keys like source_nodes as they can be very large
+            output = json.loads(serialized_output)
+            if "source_nodes" in output:
+                del output["source_nodes"]
+            if should_send_prompts():
+                span_holder.span.set_attribute(
+                    SpanAttributes.TRACELOOP_ENTITY_OUTPUT,
+                    json.dumps(output, cls=JSONEncoder),
+                )
+        except Exception:
+            pass
+
+        span_holder.span.end()
+        context_api.detach(span_holder.token)
+        with self.lock:
+            self.completed_spans += [span_holder]
+        return span_holder
+
+    def prepare_to_drop_span(
+        self, id_: str, err: Optional[Exception], **kwargs
+    ) -> Optional[SpanHolder]:
+        """Logic for dropping a span."""
+        if id_ in self.open_spans:
+            with self.lock:
+                span_holder = self.open_spans[id_]
+                self.dropped_spans += [span_holder]
+            return span_holder
+        return None
+
+
+class OpenLLEventHandler(BaseEventHandler):
+    _span_handler: OpenLLSpanHandler = PrivateAttr()
+
+    def __init__(self, span_handler: OpenLLSpanHandler):
+        super().__init__()
+        self._span_handler = span_handler
+
+    def handle(self, event: BaseEvent, **kwargs) -> Any:
+        span = get_current_span()
+        # use case with class_pattern if support for 3.9 is dropped
+        if isinstance(event, LLMChatStartEvent):
+            _set_llm_chat_request(event, span)
+        elif isinstance(event, LLMChatEndEvent):
+            _set_llm_chat_response(event, span)
+        elif isinstance(event, LLMPredictEndEvent):
+            _set_llm_predict_response(event, span)

opentelemetry_instrumentation_llamaindex-0.27.0/opentelemetry/instrumentation/llamaindex/version.py

@@ -0,0 +1 @@
+__version__ = "0.27.0"

{opentelemetry_instrumentation_llamaindex-0.26.4 → opentelemetry_instrumentation_llamaindex-0.27.0}/pyproject.toml

@@ -8,7 +8,7 @@ show_missing = true
 
 [tool.poetry]
 name = "opentelemetry-instrumentation-llamaindex"
-version = "0.26.4"
+version = "0.27.0"
 description = "OpenTelemetry LlamaIndex instrumentation"
 authors = [
   "Gal Kleinman <gal@traceloop.com>",
@@ -27,7 +27,7 @@ python = ">=3.9,<4"
 opentelemetry-api = "^1.26.0"
 opentelemetry-instrumentation = "^0.47b0"
 opentelemetry-semantic-conventions = "^0.47b0"
-opentelemetry-semantic-conventions-ai = "0.4.0"
+opentelemetry-semantic-conventions-ai = "0.4.1"
 inflection = "^0.5.1"
 
 [tool.poetry.group.dev.dependencies]
@@ -43,9 +43,9 @@ openai = "^1.35.0"
 opentelemetry-sdk = "^1.23.0"
 llama-index = "^0.10.46"
 llama-index-postprocessor-cohere-rerank = "^0.1.7"
-opentelemetry-instrumentation-openai = "==0.26.4"
-opentelemetry-instrumentation-cohere = "==0.26.4"
-opentelemetry-instrumentation-chromadb = "==0.26.4"
+opentelemetry-instrumentation-openai = "==0.27.0"
+opentelemetry-instrumentation-cohere = "==0.27.0"
+opentelemetry-instrumentation-chromadb = "==0.27.0"
 sqlalchemy = "^2.0.31"
 llama-index-agent-openai = ">=0.2.7,<0.3.0"
 llama-index-vector-stores-chroma = "^0.1.9"

opentelemetry_instrumentation_llamaindex-0.26.4/opentelemetry/instrumentation/llamaindex/version.py

@@ -1 +0,0 @@
-__version__ = "0.26.4"