How to disable new traces in Langsmith during runtime?

[apoorvkk]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

def questions_filter(state: AgentState):
    searches = state["searches"]
    subject_details = state["subject_details"]
    syllabus = generate_syllabus_prompt(subject_details)

    updated_searches = []
    all_already_seen_question_chunk_ids = []

    with ThreadPoolExecutor() as executor:
        futures = [
            executor.submit(process_search, search, syllabus)
            for search in searches
        ]
        results = [future.result() for future in as_completed(futures)]

    for search, already_seen_question_chunk_ids in results:
        updated_searches.append(search)
        all_already_seen_question_chunk_ids.extend(already_seen_question_chunk_ids)

    return {
        "searches": updated_searches,
        "already_seen_question_chunk_ids": list(set(state["already_seen_question_chunk_ids"] + all_already_seen_question_chunk_ids))
    }

Description

Hi all,

I was just curious on how can one disable logging to Langsmith when I create a new custom Python thread which calls an LLM via langchain? Please note these are standard Python threads. I would like to disable tracing for these spawned threads during runtime.

Please see the code reference of how I am spawning threads within a node of my Langgraph.

Thank you!

System Info

langchain = "^0.2.6"
mac
python3

[dosubot[bot]]

Hey there, @apoorvkk! 👋 I'm here to help you with any bugs, questions, or contributions you have in mind. Let's tackle this together!

To disable logging to Langsmith for custom Python threads that call an LLM via Langchain, you can set the tracing_v2_callback_var context variable to None for those threads. This ensures that the Langchain operations within those threads do not log to Langsmith. Here's how you can do it:

from contextvars import ContextVar
from concurrent.futures import ThreadPoolExecutor, as_completed

# Assuming tracing_v2_callback_var is imported from the relevant module
tracing_v2_callback_var: ContextVar[Optional[LangChainTracer]] = ContextVar(
    "tracing_callback_v2", default=None
)

def process_search(search, syllabus):
    # Disable tracing for this thread
    tracing_v2_callback_var.set(None)
    # Your code that processes the search
    # ...

def questions_filter(state: AgentState):
    searches = state["searches"]
    subject_details = state["subject_details"]
    syllabus = generate_syllabus_prompt(subject_details)

    updated_searches = []
    all_already_seen_question_chunk_ids = []

    with ThreadPoolExecutor() as executor:
        futures = [
            executor.submit(process_search, search, syllabus)
            for search in searches
        ]
        results = [future.result() for future in as_completed(futures)]

    for search, already_seen_question_chunk_ids in results:
        updated_searches.append(search)
        all_already_seen_question_chunk_ids.extend(already_seen_question_chunk_ids)

    return {
        "searches": updated_searches,
        "already_seen_question_chunk_ids": list(set(state["already_seen_question_chunk_ids"] + all_already_seen_question_chunk_ids))
    }

By setting tracing_v2_callback_var to None within the process_search function, you ensure that tracing is disabled for each thread spawned by the ThreadPoolExecutor ^[1].

To continue talking to Dosu, mention @dosu.

[apoorvkk]

@dosu I still want the overall trace of my graph to be logged, just not new subtraces when i spawn a thread within a node of my langgraph graph.

[dosubot[bot]]

To achieve this, you can selectively disable tracing for specific parts of your Langgraph using the tracing_v2_enabled context manager. This allows you to control which parts of your code are traced. For your case, you can disable tracing for the threads you spawn while keeping the overall trace of your graph. Here's how you can do it:

from contextvars import ContextVar
from concurrent.futures import ThreadPoolExecutor, as_completed
from langchain_core.tracers.context import tracing_v2_enabled

# Assuming tracing_v2_callback_var is imported from the relevant module
tracing_v2_callback_var: ContextVar[Optional[LangChainTracer]] = ContextVar(
    "tracing_callback_v2", default=None
)

def process_search(search, syllabus):
    # Disable tracing for this thread
    tracing_v2_callback_var.set(None)
    # Your code that processes the search
    # ...

def questions_filter(state: AgentState):
    searches = state["searches"]
    subject_details = state["subject_details"]
    syllabus = generate_syllabus_prompt(subject_details)

    updated_searches = []
    all_already_seen_question_chunk_ids = []

    with tracing_v2_enabled(project_name="my_project"):
        with ThreadPoolExecutor() as executor:
            futures = [
                executor.submit(process_search, search, syllabus)
                for search in searches
            ]
            results = [future.result() for future in as_completed(futures)]

    for search, already_seen_question_chunk_ids in results:
        updated_searches.append(search)
        all_already_seen_question_chunk_ids.extend(already_seen_question_chunk_ids)

    return {
        "searches": updated_searches,
        "already_seen_question_chunk_ids": list(set(state["already_seen_question_chunk_ids"] + all_already_seen_question_chunk_ids))
    }

By setting tracing_v2_callback_var to None within the process_search function, you ensure that tracing is disabled for each thread spawned by the ThreadPoolExecutor, while the overall trace of your graph is still logged ^[1].

To continue talking to Dosu, mention @dosu.

[apoorvkk]

@dosu with tracing_v2_enabled(project_name="my_project"): is this necessary? I already have LANGCHAIN_TRACING_V2 set to True as an env var.

[devstein]

Hey @apoorvkk Dosu only has access to the LangChain repository and documentation. Sorry for the confusion.

For questions about LangSmith, I recommend looking at the LangSmith docs or asking in the LangGraph repo

[dosubot[bot]]

Using with tracing_v2_enabled(project_name="my_project"): is not strictly necessary if you already have the LANGCHAIN_TRACING_V2 environment variable set to True. This environment variable enables tracing globally, so your LangChain operations will be logged to LangSmith without needing to explicitly use the context manager. However, using the context manager allows you to specify additional options like a project name, example ID, or tags, and it provides more granular control over which parts of your code are traced. If you don't need these additional features, you can rely on the environment variable alone ^[1].

To continue talking to Dosu, mention @dosu.