Logging of custom info in nodes

[rensdebee]

Checked other resources

• This is a feature request, not a bug report or usage question.
• I added a clear and descriptive title that summarizes the feature request.
• I searched for similar feature requests and didn't find a duplicate.

Package

langchain-azure-ai

Component

Tracer / Observability

Feature Description

I would like to log custom attributes inside nodes. For example log only certain parts of an LLM response or action taken based on an LLM response, or even log things in deterministic nodes.

Use Case

Building an application that has deterministic checks and LLM checks and no way to log the outcome of these checks

Proposed Solution

Some way to add metadata to a node during runtime which is automatically logged.

Alternatives Considered

Tried using the tracer

[santiagxf]

Hi rensdebee. Thanks for reaching out. You mentioned logging custom attributes inside a node. You can set attributes either for the invocation, or for the nodes themselves using the key metadata.

Any metadata key starting with gen_ai. is forwarded as a span attribute to the trace:

config = {
    "callbacks": [tracer],
    "metadata": {
        "gen_ai.my_attribute": "PlannerAgent",
        "thread_id": "session-abc-123",
    },
}
result = graph.invoke({"messages": [message]}, config)

When using LangGraph, you can also set metadata per node in the graph definition:

workflow = StateGraph(MessagesState)
workflow.add_node(
    "planner",
    planner_fn,
    metadata={
        "gen_ai.my_attribute": "PlannerAgent",
        "otel_agent_span": True,
    },
)

If you need more control, like for example logging within custom code, then you need to retrieve the tracer object:

from opentelemetry import trace

tracer = trace.getTracer("AzureAIOpenTelemetryTracer")

def my_node(state):
    span = trace.get_current_span()
    span.set_attribute("langgraph.node_name", "my_node")