paid-python 0.3.5__tar.gz → 0.3.6__tar.gz

{paid_python-0.3.5 → paid_python-0.3.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: paid-python
-Version: 0.3.5
+Version: 0.3.6
 Summary: 
 Requires-Python: >=3.9,<3.14
 Classifier: Intended Audience :: Developers
@@ -21,11 +21,11 @@ Requires-Dist: httpx (>=0.21.0)
 Requires-Dist: mutagen (>=1.47.0)
 Requires-Dist: openinference-instrumentation-bedrock (>=0.1.0)
 Requires-Dist: openinference-instrumentation-google-genai (>=0.1.8)
+Requires-Dist: openinference-instrumentation-langchain (>=0.1.55)
 Requires-Dist: openinference-instrumentation-openai-agents (>=1.0.0)
 Requires-Dist: opentelemetry-api (>=1.23.0)
 Requires-Dist: opentelemetry-exporter-otlp-proto-http (>=1.23.0)
 Requires-Dist: opentelemetry-instrumentation-anthropic (>=0.47.0)
-Requires-Dist: opentelemetry-instrumentation-langchain (>=0.47.0)
 Requires-Dist: opentelemetry-sdk (>=1.23.0)
 Requires-Dist: pydantic (>=1.9.0)
 Requires-Dist: pydantic-core (>=2.18.0)

{paid_python-0.3.5 → paid_python-0.3.6}/pyproject.toml

@@ -3,7 +3,7 @@ name = "paid-python"
 
 [tool.poetry]
 name = "paid-python"
-version = "0.3.5"
+version = "0.3.6"
 description = ""
 readme = "README.md"
 authors = []
@@ -45,7 +45,7 @@ mutagen = ">=1.47.0"
 python-dotenv = ">=0.15.0"
 opentelemetry-instrumentation-anthropic = ">=0.47.0"
 # opentelemetry-instrumentation-openai is vendored from https://github.com/paid-ai/openllmetry-instrumentation-patch (patched version)
-opentelemetry-instrumentation-langchain = ">=0.47.0"
+openinference-instrumentation-langchain = ">=0.1.55"
 openinference-instrumentation-openai-agents = ">=1.0.0"
 openinference-instrumentation-bedrock = ">=0.1.0"
 openinference-instrumentation-google-genai = ">=0.1.8"

{paid_python-0.3.5 → paid_python-0.3.6}/src/paid/tracing/autoinstrumentation.py

@@ -44,7 +44,7 @@ except ImportError:
     BEDROCK_AVAILABLE = False
 
 try:
-    from opentelemetry.instrumentation.langchain import LangchainInstrumentor
+    from openinference.instrumentation.langchain import LangChainInstrumentor
 
     LANGCHAIN_AVAILABLE = True
 except ImportError:
@@ -196,16 +196,14 @@ def _instrument_bedrock() -> None:
 
 def _instrument_langchain() -> None:
     """
-    Instrument LangChain using opentelemetry-instrumentation-langchain.
+    Instrument LangChain using openinference-instrumentation-langchain.
     """
     if not LANGCHAIN_AVAILABLE:
         logger.warning("LangChain instrumentation library not available, skipping instrumentation")
         return
 
     # Instrument LangChain with Paid's tracer provider
-    LangchainInstrumentor(disable_trace_context_propagation=True).instrument(
-        tracer_provider=tracing.paid_tracer_provider
-    )
+    LangChainInstrumentor().instrument(tracer_provider=tracing.paid_tracer_provider)
 
     _initialized_instrumentors.append("langchain")
     logger.info("LangChain auto-instrumentation enabled")

{paid_python-0.3.5 → paid_python-0.3.6}/src/paid/tracing/tracing.py

@@ -63,25 +63,36 @@ class PaidSpanProcessor(SpanProcessor):
     1. Prefixes all span names with 'paid.trace.'
     2. Automatically adds external_customer_id and external_agent_id attributes
        to all spans based on context variables set by the tracing decorator.
+    3. Filters out prompt/response data unless store_prompt=True.
+    4. Filters out duplicate LangChain spans that may duplicate information from other instrumentations.
     """
 
     SPAN_NAME_PREFIX = "paid.trace."
     PROMPT_ATTRIBUTES_SUBSTRINGS = {
-        "prompt",
-        # "gen_ai.prompt",
         "gen_ai.completion",
         "gen_ai.request.messages",
         "gen_ai.response.messages",
         "llm.output_message",
         "llm.input_message",
         "llm.invocation_parameters",
+        "gen_ai.prompt",
+        "langchain.prompt",
         "output.value",
         "input.value",
-        # "langchain.prompt",
     }
 
     def on_start(self, span: Span, parent_context: Optional[Context] = None) -> None:
         """Called when a span is started. Prefix the span name and add attributes."""
+
+        LANGCHAIN_SPAN_FILTERS = ["ChatOpenAI", "ChatAnthropic"]
+        if any(f in span.name for f in LANGCHAIN_SPAN_FILTERS):
+            # HACK TO FILTER DUPLICATE SPANS CREATED BY LANGCHAIN INSTRUMENTATION.
+            # Langchain instrumentation creates spans, that are created by other instrumentations (ex. OpenAI, Anthropic).
+            # Not all spans need filtering (ex. ChatGoogleGenerativeAI), so first test actual telemetry before adding filters.
+            # TODO: maybe consider a dropping sampler for such spans instead of raising an exception?
+            logger.debug(f"Dropping Langchain span: {span.name}")
+            raise Exception(f"Dropping Langchain span: {span.name}")
+
         # Prefix the span name
         if span.name and not span.name.startswith(self.SPAN_NAME_PREFIX):
             span.update_name(f"{self.SPAN_NAME_PREFIX}{span.name}")