Tracing with Langgraph Breakpoints

[lucatrovato]

Hello,
I am implementing a chatbot using Langgraph and Human-in-the-loop, which means that there are breakpoints in the graph, the application returns to the user and the graph can be resumed with the following call. This means that every time there is a breakpoint, a new trace is created. How can I configure tracing such that each conversation or thread is properly captured together? Can I use a thread-id or a parameter to uniquely identify and continue tracking in an existing trace?

[dosubot[bot]]

Ciao @lucatrovato! I'm here to help you with any bugs, questions, or contributions you have. If you need assistance, feel free to ask.

To ensure that each conversation or thread is properly captured together in Langgraph, you can use a thread_id or similar parameter to uniquely identify and continue tracking in an existing trace. You can include attributes like session_id, conversation_id, or thread_id in your session metadata to manage and track sessions or threads effectively.

Here's an example of how you can set up the session metadata to include a thread_id:

@pytest.mark.parametrize(
    "session_metadata, expected_session_id",
    [
        (
            {
                "thread_id": "test-langchain-thread-id",
            },
            "test-langchain-thread-id",
        ),
    ],
)
@pytest.mark.parametrize("use_context_attributes", [False, True])

In this setup, you pass a dictionary with a thread_id to the session metadata, which will be used to track the conversation or thread. The expected_session_id will be set to the thread_id value, ensuring that the trace is properly captured and associated with the correct thread ^[1][2].

To continue talking to Dosu, mention @dosu.

[axiomofjoy]

Hey @lucatrovato, we will be releasing sessions support in the Phoenix UI soon. Once released, you should be able to provide your conversation/ thread ID to LangChain as you usually would to view those separate traces in a single conversation.

[lucatrovato]

Hi @axiomofjoy , thanks a lot for your answer. Any idea on the timeline, i.e. what does soon mean?

[axiomofjoy]

Hey @lucatrovato, we don't have a concrete launch date planned, but it will be before the end of the year.

[axiomofjoy]

You can track progress here.

[mikeldking]

Hey @lucatrovato - you should be able to use langchain's native threads (https://docs.smith.langchain.com/old/monitoring/faq/threads) and they will be propagated to phoenix as sessions (https://docs.arize.com/phoenix/tracing/how-to-tracing/setup-sessions)