Tool calls as separate traces - How do I only trace langgraph?

[RamiAwar]

Not sure where to ask this, gonna start here.

I'm getting a lot of traces from one graph execution. Not sure why it's not just one entry, it's pretty annoying.

I didn't really do anything special in my setup, only added the langsmith env vars locally.

Any ideas?

[runirudh]

I had the same issue. The fix suggested here works, remove tracing from env and pass callbacks to invoke
#720 (comment)

[RamiAwar]

it does look like the same issue, thank you for pointing it out! @runirudh

Fix also worked for me for now.

[hinthornw]

Thanks for reporting (and for your patience)could you share:

• python version
• langgraph version
• langsmith version
• Are you running sync or async. Which method to call?
• sample code for what you're doing

We are triaging a similar issue so want to
a) be able to help y'all
b) see if these are the same issue

[RamiAwar]

Hey @hinthornw , it does look very similar as the fix over there also worked for me.

I'm defining a langgraph graph (build graph, compile) with multiple nodes (mostly sql toolkit tools).

To be fully transparent, I started drifting away from the 'recommended' way of using Langgraph pretty quickly as I had to do some more complicated things. My code is open source actually, but might be a bit hard to read.

Gonna share it anyways, in case you're curious. The LLM flow lives inside backend/dataline/llm_flow/graph, and uses tools from the backend/dataline/llm_flow/toolkit.
https://github.com/RamiAwar/dataline

Sorry I can't be of more help!