fix(langchain): instrument chat models (#741)

packages/opentelemetry-instrumentation-langchain/opentelemetry/instrumentation/langchain/__init__.py

@@ -21,6 +21,10 @@
     llm_wrapper,
     allm_wrapper,
 )
+from opentelemetry.instrumentation.langchain.custom_chat_wrapper import (
+    chat_wrapper,
+    achat_wrapper,
+)
 from opentelemetry.instrumentation.langchain.version import __version__
 
 from opentelemetry.semconv.ai import TraceloopSpanKindValues
@@ -101,14 +105,14 @@
     {
         "package": "langchain.chat_models.base",
         "object": "BaseChatModel",
-        "method": "invoke",
-        "wrapper": task_wrapper,
+        "method": "generate",
+        "wrapper": chat_wrapper,
     },
     {
         "package": "langchain.chat_models.base",
         "object": "BaseChatModel",
-        "method": "ainvoke",
-        "wrapper": atask_wrapper,
+        "method": "agenerate",
+        "wrapper": achat_wrapper,
     },
     {
         "package": "langchain.schema",

packages/opentelemetry-instrumentation-langchain/opentelemetry/instrumentation/langchain/custom_chat_wrapper.py

@@ -0,0 +1,66 @@
+import json
+from opentelemetry import context as context_api
+from opentelemetry.instrumentation.utils import _SUPPRESS_INSTRUMENTATION_KEY
+
+from opentelemetry.semconv.ai import SpanAttributes, LLMRequestTypeValues
+
+from opentelemetry.instrumentation.langchain.utils import _with_tracer_wrapper
+from opentelemetry.instrumentation.langchain.utils import should_send_prompts
+
+
+@_with_tracer_wrapper
+def chat_wrapper(tracer, to_wrap, wrapped, instance, args, kwargs):
+    """Instruments and calls every function defined in TO_WRAP."""
+    if context_api.get_value(_SUPPRESS_INSTRUMENTATION_KEY):
+        return wrapped(*args, **kwargs)
+
+    name = f"langchain.task.{instance.__class__.__name__}"
+    with tracer.start_as_current_span(name) as span:
+        _handle_request(span, args, kwargs, instance)
+        return_value = wrapped(*args, **kwargs)
+        _handle_response(span, return_value)
+
+    return return_value
+
+
+@_with_tracer_wrapper
+async def achat_wrapper(tracer, to_wrap, wrapped, instance, args, kwargs):
+    """Instruments and calls every function defined in TO_WRAP."""
+    if context_api.get_value(_SUPPRESS_INSTRUMENTATION_KEY):
+        return wrapped(*args, **kwargs)
+
+    name = f"langchain.task.{instance.__class__.__name__}"
+    with tracer.start_as_current_span(name) as span:
+        _handle_request(span, args, kwargs, instance)
+        return_value = await wrapped(*args, **kwargs)
+        _handle_response(span, return_value)
+
+    return return_value
+
+
+def _handle_request(span, args, kwargs, instance):
+    model = instance.model if hasattr(instance, "model") else instance.model_name
+    span.set_attribute(SpanAttributes.LLM_REQUEST_TYPE, LLMRequestTypeValues.CHAT.value)
+    span.set_attribute(SpanAttributes.LLM_REQUEST_MODEL, model)
+    span.set_attribute(SpanAttributes.LLM_RESPONSE_MODEL, model)
+
+    if should_send_prompts():
+        for idx, prompt in enumerate(args[0][0]):
+            if isinstance(prompt.content, list):
+                span.set_attribute(
+                    f"{SpanAttributes.LLM_PROMPTS}.{idx}.user",
+                    json.dumps(prompt.content),
+                )
+            else:
+                span.set_attribute(
+                    f"{SpanAttributes.LLM_PROMPTS}.{idx}.user", prompt.content
+                )
+
+
+def _handle_response(span, return_value):
+    if should_send_prompts():
+        for idx, generation in enumerate(return_value.generations):
+            span.set_attribute(
+                f"{SpanAttributes.LLM_COMPLETIONS}.{idx}.content",
+                generation[0].text,
+            )