Shouldn't thread_ts be set to a datetime object instead of checkpoint["id"]?

[umithyo]

Checked other resources

• [X] I added a very descriptive title to this issue.
• [X] I searched the LangChain documentation with the integrated search.
• [X] I used the GitHub search to find a similar question and didn't find it.
• [x] I am sure that this is a bug in LangChain rather than my code.

Example Code

The following code:

import datetime
from langgraph.checkpoint.sqlite import SqliteSaver
from langgraph.graph import StateGraph
import uuid

builder = StateGraph(State)

## This part is my implementation, not the part of the problem
primary_assistant = PrimaryAssistant()
primary_assistant.build_workflow(builder)
## End of my implementation

memory = SqliteSaver.from_conn_string("checkpoints")
graph = builder.compile(checkpointer=memory)

config = {
        "configurable": {
            "thread_id": uuid.uuid4(),
            "thread_ts": datetime.datetime.utcnow()
        }
    }

graph.invoke({"messages": "Hello!"}, config, debug=True)

Following graph.invoke(), stream method sets the thread_ts to checkpoint["id"] instead of the passed thread_ts in config

Error Message and Stack Trace (if applicable)

No response

Description

I'm trying to create a persistent storage for LangGraph's messages. While trying it out, I noticed that thread_ts argument that I pass gets changed to checkpoint["id"] by stream

Am I missing something or is this something to be fixed? If so, I'd like to create a PR for this.

System Info

langchain==0.2.1 langchain-community==0.2.1 langchain-core==0.2.1 langchain-mongodb==0.1.5 langchain-openai==0.1.7 langchain-pinecone==0.1.1 langchain-text-splitters==0.2.0

[hinthornw]

Hi @umithyo great q - as you pointed out, thread_ts isn't exposed/really honored as a user-configurable value right now - could you share more about what you plan to gain by managing it outside the system?

If we were to implement this, it would likely only apply to the first checkpoint and then we'd overwrite it in subsequent ones within a trace anyhow

[umithyo]

Hi @umithyo great q - as you pointed out, thread_ts isn't exposed as a user-configurable value right now - could you share more about what you plan to gain by managing it outside the system?

If we were to implement this, it would likely only apply to the first checkpoint and then we'd overwrite it in subsequent ones within a trace anyhow

Hi @hinthornw

Thank you. I feel like it's being represented wrong is all. While it's shown as a datetime value in StateSnapshot that is not the case at all. Also, last message is being fetched by this field although it is a uuid. Maybe I haven't gotten my head around this.

I have no use case at the moment but this just caught my eye while going through the sqlite.py file.

[jty016]

Hi! @hinthornw Actually I have some use case for the thread_ts. Maybe it could be slightly off-topic. When I tried to use LangServe endpoint with a graph built by LangGraph, I'd like to use TimeTravel feature from client. For now, howerver, I don't know what will be the best way to deliver thread_ts from StateSnapShot to a client of LangServe.

FYI, thread_ts is opened to client because it is included in configurable

[hinthornw]

Hi @jty016 , for time travel in a deployed version, we're also working on https://github.com/langchain-ai/langgraph-example , which would add better first-class support for this kind of thing. Still in alpha right now but hoping to get something ready by EOM

Thanks for clarifying, @umithyo ! Will sync with nuno on this before I give a full response. Appreciate both of your patience

[jty016]

Hi @jty016 , for time travel in a deployed version, we're also working on https://github.com/langchain-ai/langgraph-example , which would add better first-class support for this kind of thing. Still in alpha right now but hoping to get something ready by EOM

This is very interesting and the way I really like to
however I think langgraph cli and langgraph sdk repo is not ready or opened.

[Arslan-Mehmood1]

@jty016 Hi, I'm having issue with using langgraph with langserve playground.

I'm stuck and don't know how to pass thread_id required by checkpointer in langgraph.

current_code after trying multiple things:

### LANGSERVE PLAYGROUND

from typing import List, Union
from fastapi import FastAPI
from langserve import add_routes
from langserve.pydantic_v1 import BaseModel, Field
from langchain_core.runnables import RunnableLambda, RunnablePassthrough
from langchain_core.messages import AIMessage, HumanMessage, SystemMessage
from langchain_core.runnables import ConfigurableField

from langchain_core.runnables import RunnableConfig

app = FastAPI(
    title="Biscuit AI LangChain Server",
    version="1.0",
    description="Api server using LangChain's Runnable interfaces",
)

class InputChat(BaseModel):
    """Input for the chat endpoint."""

    messages: List[Union[HumanMessage, AIMessage, SystemMessage]] = Field(
        ...,
        description="The chat messages representing the current conversation.",
    )
    input: str
    # thread_id:int

def output_parsing_for_playground(agent_output):
    data = agent_output[-1]

    content = next(iter(data.values()))['messages'][0].content

    print("graph output : ",agent_output[-1])
    print("content : ",content)

    return content

# def execute_graph(input,thread_id):
#     input_message = [HumanMessage(content=input)]
#     config = {"configurable": {"thread_id": thread_id}}

#     return graph.stream({"messages": input_message}, config, stream_mode="values")

add_routes(
    app,
    (
        # RunnablePassthrough.assign(thread_id=lambda x: x["thread_id"]) 
        graph.with_config(RunnableConfig(configurable={"configurable": {"thread_id": thread_id}})) 
        # | execute_graph
        | RunnableLambda(output_parsing_for_playground)).with_types(input_type=InputChat, output_type=str
    ),
    config_keys=["configurable"],
    playground_type="default"
)

if __name__ == "__main__":
    import uvicorn
    host = "localhost"
    port = 8181

    uvicorn.run(app, host=host, port=port)

[Arslan-Mehmood1]

error:

| ValueError: Checkpointer requires one or more of the following 'configurable' keys: ['thread_id', 'thread_ts']

[Arslan-Mehmood1]

[nfcampos]

thread_ts has been renamed to checkpoint_id in 0.2

[Arslan-Mehmood1]

@hinthornw any solution for above mentioned issue, on using LangGraph with checkpointer with LangServe.

[sandeep135]

@Arslan-Mehmood1 I can help you